consul/agent/pool/pool.go

673 lines
17 KiB
Go
Raw Normal View History

package pool
2013-12-09 20:09:57 +00:00
import (
"container/list"
"crypto/tls"
2013-12-24 20:22:42 +00:00
"fmt"
2014-05-28 23:32:10 +00:00
"io"
2013-12-09 20:09:57 +00:00
"net"
2013-12-09 22:52:22 +00:00
"net/rpc"
2013-12-09 20:09:57 +00:00
"sync"
"sync/atomic"
2013-12-09 20:09:57 +00:00
"time"
2014-10-02 17:26:25 +00:00
"github.com/hashicorp/consul/agent/structs"
"github.com/hashicorp/consul/lib"
"github.com/hashicorp/consul/tlsutil"
msgpackrpc "github.com/hashicorp/net-rpc-msgpackrpc"
2014-10-02 17:26:25 +00:00
"github.com/hashicorp/yamux"
2013-12-09 20:09:57 +00:00
)
const defaultDialTimeout = 10 * time.Second
// muxSession is used to provide an interface for a stream multiplexer.
type muxSession interface {
Open() (net.Conn, error)
Close() error
}
// streamClient is used to wrap a stream with an RPC client
type StreamClient struct {
stream net.Conn
codec rpc.ClientCodec
}
func (sc *StreamClient) Close() {
sc.stream.Close()
sc.codec.Close()
}
2013-12-09 20:09:57 +00:00
// Conn is a pooled connection to a Consul server
type Conn struct {
refCount int32
shouldClose int32
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
nodeName string
addr net.Addr
session muxSession
2013-12-19 23:42:17 +00:00
lastUsed time.Time
pool *ConnPool
clients *list.List
clientLock sync.Mutex
2013-12-09 20:09:57 +00:00
}
2013-12-09 22:58:49 +00:00
func (c *Conn) Close() error {
return c.session.Close()
2013-12-09 22:58:49 +00:00
}
// getClient is used to get a cached or new client
func (c *Conn) getClient() (*StreamClient, error) {
// Check for cached client
c.clientLock.Lock()
front := c.clients.Front()
if front != nil {
c.clients.Remove(front)
}
c.clientLock.Unlock()
if front != nil {
return front.Value.(*StreamClient), nil
}
// Open a new session
stream, err := c.session.Open()
if err != nil {
return nil, err
}
// Create the RPC client
codec := msgpackrpc.NewCodecFromHandle(true, true, stream, structs.MsgpackHandle)
// Return a new stream client
sc := &StreamClient{
stream: stream,
codec: codec,
}
return sc, nil
}
// returnStream is used when done with a stream
// to allow re-use by a future RPC
func (c *Conn) returnClient(client *StreamClient) {
didSave := false
c.clientLock.Lock()
if c.clients.Len() < c.pool.MaxStreams && atomic.LoadInt32(&c.shouldClose) == 0 {
c.clients.PushFront(client)
didSave = true
// If this is a Yamux stream, shrink the internal buffers so that
// we can GC the idle memory
if ys, ok := client.stream.(*yamux.Stream); ok {
ys.Shrink()
}
}
c.clientLock.Unlock()
if !didSave {
client.Close()
}
}
// markForUse does all the bookkeeping required to ready a connection for use.
func (c *Conn) markForUse() {
c.lastUsed = time.Now()
atomic.AddInt32(&c.refCount, 1)
}
// ConnPool is used to maintain a connection pool to other Consul
// servers. This is used to reduce the latency of RPC requests between
// servers. It is only used to pool connections in the rpcConsul mode.
// Raft connections are pooled separately. Maintain at most one
// connection per host, for up to MaxTime. When MaxTime connection
// reaping is disabled. MaxStreams is used to control the number of idle
// streams allowed. If TLS settings are provided outgoing connections
// use TLS.
2013-12-09 20:09:57 +00:00
type ConnPool struct {
// SrcAddr is the source address for outgoing connections.
SrcAddr *net.TCPAddr
2014-05-28 23:32:10 +00:00
// LogOutput is used to control logging
LogOutput io.Writer
2014-05-28 23:32:10 +00:00
2013-12-19 23:42:17 +00:00
// The maximum time to keep a connection open
MaxTime time.Duration
2013-12-19 23:42:17 +00:00
// The maximum number of open streams to keep
MaxStreams int
// TLSConfigurator
TLSConfigurator *tlsutil.Configurator
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
// GatewayResolver is a function that returns a suitable random mesh
// gateway address for dialing servers in a given DC. This is only
// needed if wan federation via mesh gateways is enabled.
GatewayResolver func(string) string
// Datacenter is the datacenter of the current agent.
Datacenter string
// Server should be set to true if this connection pool is configured in a
// server instead of a client.
Server bool
sync.Mutex
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
// pool maps a nodeName+address to a open connection
pool map[string]*Conn
2013-12-09 20:09:57 +00:00
// limiter is used to throttle the number of connect attempts
// to a given address. The first thread will attempt a connection
// and put a channel in here, which all other threads will wait
// on to close.
limiter map[string]chan struct{}
2013-12-09 20:09:57 +00:00
// Used to indicate the pool is shutdown
shutdown bool
shutdownCh chan struct{}
// once initializes the internal data structures and connection
// reaping on first use.
once sync.Once
2013-12-09 20:09:57 +00:00
}
// init configures the initial data structures. It should be called
// by p.once.Do(p.init) in all public methods.
func (p *ConnPool) init() {
p.pool = make(map[string]*Conn)
p.limiter = make(map[string]chan struct{})
p.shutdownCh = make(chan struct{})
if p.MaxTime > 0 {
go p.reap()
2013-12-19 23:42:17 +00:00
}
2013-12-09 20:09:57 +00:00
}
// Shutdown is used to close the connection pool
func (p *ConnPool) Shutdown() error {
p.once.Do(p.init)
2013-12-09 20:09:57 +00:00
p.Lock()
defer p.Unlock()
for _, conn := range p.pool {
conn.Close()
2013-12-09 20:09:57 +00:00
}
p.pool = make(map[string]*Conn)
2013-12-09 20:09:57 +00:00
if p.shutdown {
return nil
}
p.shutdown = true
close(p.shutdownCh)
2013-12-09 20:09:57 +00:00
return nil
}
// acquire will return a pooled connection, if available. Otherwise it will
// wait for an existing connection attempt to finish, if one if in progress,
// and will return that one if it succeeds. If all else fails, it will return a
// newly-created connection and add it to the pool.
func (p *ConnPool) acquire(dc string, nodeName string, addr net.Addr) (*Conn, error) {
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
if nodeName == "" {
return nil, fmt.Errorf("pool: ConnPool.acquire requires a node name")
}
addrStr := addr.String()
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
poolKey := nodeName + ":" + addrStr
// Check to see if there's a pooled connection available. This is up
// here since it should the vastly more common case than the rest
// of the code here.
p.Lock()
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
c := p.pool[poolKey]
if c != nil {
c.markForUse()
p.Unlock()
return c, nil
2013-12-09 20:09:57 +00:00
}
// If not (while we are still locked), set up the throttling structure
// for this address, which will make everyone else wait until our
// attempt is done.
var wait chan struct{}
var ok bool
if wait, ok = p.limiter[addrStr]; !ok {
wait = make(chan struct{})
p.limiter[addrStr] = wait
}
isLeadThread := !ok
p.Unlock()
// If we are the lead thread, make the new connection and then wake
// everybody else up to see if we got it.
if isLeadThread {
c, err := p.getNewConn(dc, nodeName, addr)
p.Lock()
delete(p.limiter, addrStr)
close(wait)
if err != nil {
p.Unlock()
return nil, err
}
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
p.pool[poolKey] = c
p.Unlock()
return c, nil
}
// Otherwise, wait for the lead thread to attempt the connection
// and use what's in the pool at that point.
select {
case <-p.shutdownCh:
return nil, fmt.Errorf("rpc error: shutdown")
case <-wait:
}
// See if the lead thread was able to get us a connection.
p.Lock()
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
if c := p.pool[poolKey]; c != nil {
c.markForUse()
p.Unlock()
return c, nil
2013-12-09 20:09:57 +00:00
}
p.Unlock()
return nil, fmt.Errorf("rpc error: lead thread didn't get connection")
2013-12-09 20:09:57 +00:00
}
// HalfCloser is an interface that exposes a TCP half-close without exposing
// the underlying TLS or raw TCP connection.
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
type HalfCloser interface {
CloseWrite() error
}
// DialTimeout is used to establish a raw connection to the given server, with
// given connection timeout. It also writes RPCTLS as the first byte.
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
func (p *ConnPool) DialTimeout(
dc string,
nodeName string,
addr net.Addr,
actualRPCType RPCType,
) (net.Conn, HalfCloser, error) {
p.once.Do(p.init)
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
if p.Server && p.GatewayResolver != nil && p.TLSConfigurator != nil && dc != p.Datacenter {
// NOTE: TLS is required on this branch.
return DialTimeoutWithRPCTypeViaMeshGateway(
dc,
nodeName,
addr,
p.SrcAddr,
p.TLSConfigurator.OutgoingALPNRPCWrapper(),
actualRPCType,
RPCTLS,
// gateway stuff
p.Server,
p.TLSConfigurator,
p.GatewayResolver,
p.Datacenter,
)
}
return p.dial(
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
dc,
nodeName,
addr,
actualRPCType,
RPCTLS,
)
}
func (p *ConnPool) dial(
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
dc string,
nodeName string,
addr net.Addr,
actualRPCType RPCType,
tlsRPCType RPCType,
) (net.Conn, HalfCloser, error) {
2013-12-09 20:09:57 +00:00
// Try to dial the conn
d := &net.Dialer{LocalAddr: p.SrcAddr, Timeout: defaultDialTimeout}
conn, err := d.Dial("tcp", addr.String())
2013-12-09 20:09:57 +00:00
if err != nil {
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
return nil, nil, err
2013-12-09 20:09:57 +00:00
}
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
var hc HalfCloser
if tcp, ok := conn.(*net.TCPConn); ok {
tcp.SetKeepAlive(true)
tcp.SetNoDelay(true)
// Expose TCPConn CloseWrite method on HalfCloser
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
hc = tcp
}
// Check if TLS is enabled
if p.TLSConfigurator.UseTLS(dc) {
wrapper := p.TLSConfigurator.OutgoingRPCWrapper()
// Switch the connection into TLS mode
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
if _, err := conn.Write([]byte{byte(tlsRPCType)}); err != nil {
conn.Close()
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
return nil, nil, err
}
2013-12-09 20:09:57 +00:00
// Wrap the connection in a TLS client
tlsConn, err := wrapper(dc, conn)
if err != nil {
conn.Close()
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
return nil, nil, err
}
2014-10-02 17:26:25 +00:00
conn = tlsConn
// If this is a tls.Conn, expose HalfCloser to caller
if tlsConn, ok := conn.(*tls.Conn); ok {
hc = tlsConn
}
}
2013-12-09 20:09:57 +00:00
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
// Send the type-byte for the protocol if one is required.
//
// When using insecure TLS there is no inner type-byte as these connections
// aren't wrapped like the standard TLS ones are.
if tlsRPCType != RPCTLSInsecure {
if _, err := conn.Write([]byte{byte(actualRPCType)}); err != nil {
conn.Close()
return nil, nil, err
}
}
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
return conn, hc, nil
}
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
// DialTimeoutWithRPCTypeViaMeshGateway dials the destination node and sets up
// the connection to be the correct RPC type using ALPN. This currently is
// exclusively used to dial other servers in foreign datacenters via mesh
// gateways.
//
// NOTE: There is a close mirror of this method in agent/consul/wanfed/wanfed.go:dial
func DialTimeoutWithRPCTypeViaMeshGateway(
dc string,
nodeName string,
addr net.Addr,
src *net.TCPAddr,
wrapper tlsutil.ALPNWrapper,
actualRPCType RPCType,
tlsRPCType RPCType,
// gateway stuff
dialingFromServer bool,
tlsConfigurator *tlsutil.Configurator,
gatewayResolver func(string) string,
thisDatacenter string,
) (net.Conn, HalfCloser, error) {
if !dialingFromServer {
return nil, nil, fmt.Errorf("must dial via mesh gateways from a server agent")
} else if gatewayResolver == nil {
return nil, nil, fmt.Errorf("gatewayResolver is nil")
} else if tlsConfigurator == nil {
return nil, nil, fmt.Errorf("tlsConfigurator is nil")
} else if dc == thisDatacenter {
return nil, nil, fmt.Errorf("cannot dial servers in the same datacenter via a mesh gateway")
} else if wrapper == nil {
return nil, nil, fmt.Errorf("cannot dial via a mesh gateway when outgoing TLS is disabled")
}
nextProto := actualRPCType.ALPNString()
if nextProto == "" {
return nil, nil, fmt.Errorf("rpc type %d cannot be routed through a mesh gateway", actualRPCType)
}
gwAddr := gatewayResolver(dc)
if gwAddr == "" {
return nil, nil, structs.ErrDCNotAvailable
}
dialer := &net.Dialer{LocalAddr: src, Timeout: defaultDialTimeout}
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
rawConn, err := dialer.Dial("tcp", gwAddr)
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
if err != nil {
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
return nil, nil, err
}
if tcp, ok := rawConn.(*net.TCPConn); ok {
_ = tcp.SetKeepAlive(true)
_ = tcp.SetNoDelay(true)
}
// NOTE: now we wrap the connection in a TLS client.
tlsConn, err := wrapper(dc, nodeName, nextProto, rawConn)
if err != nil {
return nil, nil, err
}
var conn net.Conn = tlsConn
var hc HalfCloser
if tlsConn, ok := conn.(*tls.Conn); ok {
// Expose *tls.Conn CloseWrite method on HalfCloser
hc = tlsConn
}
return conn, hc, nil
}
// getNewConn is used to return a new connection
func (p *ConnPool) getNewConn(dc string, nodeName string, addr net.Addr) (*Conn, error) {
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
if nodeName == "" {
return nil, fmt.Errorf("pool: ConnPool.getNewConn requires a node name")
Adds support for snapshots and restores. (#2396) * Updates Raft library to get new snapshot/restore API. * Basic backup and restore working, but need some cleanup. * Breaks out a snapshot module and adds a SHA256 integrity check. * Adds snapshot ACL and fills in some missing comments. * Require a consistent read for snapshots. * Make sure snapshot works if ACLs aren't enabled. * Adds a bit of package documentation. * Returns an empty response from restore to avoid EOF errors. * Adds API client support for snapshots. * Makes internal file names match on-disk file snapshots. * Adds DC and token coverage for snapshot API test. * Adds missing documentation. * Adds a unit test for the snapshot client endpoint. * Moves the connection pool out of the client for easier testing. * Fixes an incidental issue in the prepared query unit test. I realized I had two servers in bootstrap mode so this wasn't a good setup. * Adds a half close to the TCP stream and fixes panic on error. * Adds client and endpoint tests for snapshots. * Moves the pool back into the snapshot RPC client. * Adds a TLS test and fixes half-closes for TLS connections. * Tweaks some comments. * Adds a low-level snapshot test. This is independent of Consul so we can pull this out into a library later if we want to. * Cleans up snapshot and archive and completes archive tests. * Sends a clear error for snapshot operations in dev mode. Snapshots require the Raft snapshots to be readable, which isn't supported in dev mode. Send a clear error instead of a deep-down Raft one. * Adds docs for the snapshot endpoint. * Adds a stale mode and index feedback for snapshot saves. This gives folks a way to extract data even if the cluster has no leader. * Changes the internal format of a snapshot from zip to tgz. * Pulls in Raft fix to cancel inflight before a restore. * Pulls in new Raft restore interface. * Adds metadata to snapshot saves and a verify function. * Adds basic save and restore snapshot CLI commands. * Gets rid of tarball extensions and adds restore message. * Fixes an incidental bad link in the KV docs. * Adds documentation for the snapshot CLI commands. * Scuttle any request body when a snapshot is saved. * Fixes archive unit test error message check. * Allows for nil output writers in snapshot RPC handlers. * Renames hash list Decode to DecodeAndVerify. * Closes the client connection for snapshot ops. * Lowers timeout for restore ops. * Updates Raft vendor to get new Restore signature and integrates with Consul. * Bounces the leader's internal state when we do a restore.
2016-10-26 02:20:24 +00:00
}
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
// Get a new, raw connection and write the Consul multiplex byte to set the mode
conn, _, err := p.DialTimeout(dc, nodeName, addr, RPCMultiplexV2)
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
if err != nil {
return nil, err
}
2013-12-09 22:52:22 +00:00
// Setup the logger
conf := yamux.DefaultConfig()
conf.LogOutput = p.LogOutput
// Create a multiplexed session
session, err := yamux.Client(conn, conf)
if err != nil {
conn.Close()
return nil, fmt.Errorf("Failed to create yamux client: %w", err)
}
2013-12-09 20:09:57 +00:00
// Wrap the connection
c := &Conn{
refCount: 1,
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
nodeName: nodeName,
addr: addr,
session: session,
clients: list.New(),
lastUsed: time.Now(),
pool: p,
}
return c, nil
}
2015-09-15 12:22:08 +00:00
// clearConn is used to clear any cached connection, potentially in response to an error
func (p *ConnPool) clearConn(conn *Conn) {
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
if conn.nodeName == "" {
panic("pool: ConnPool.acquire requires a node name")
}
// Ensure returned streams are closed
atomic.StoreInt32(&conn.shouldClose, 1)
// Clear from the cache
addrStr := conn.addr.String()
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
poolKey := conn.nodeName + ":" + addrStr
p.Lock()
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
if c, ok := p.pool[poolKey]; ok && c == conn {
delete(p.pool, poolKey)
}
p.Unlock()
// Close down immediately if idle
if refCount := atomic.LoadInt32(&conn.refCount); refCount == 0 {
conn.Close()
}
}
2013-12-09 20:09:57 +00:00
// releaseConn is invoked when we are done with a conn to reduce the ref count
func (p *ConnPool) releaseConn(conn *Conn) {
refCount := atomic.AddInt32(&conn.refCount, -1)
if refCount == 0 && atomic.LoadInt32(&conn.shouldClose) == 1 {
conn.Close()
}
2013-12-09 20:09:57 +00:00
}
2013-12-09 22:58:49 +00:00
// getClient is used to get a usable client for an address
func (p *ConnPool) getClient(dc string, nodeName string, addr net.Addr) (*Conn, *StreamClient, error) {
retries := 0
START:
2013-12-09 22:58:49 +00:00
// Try to get a conn first
conn, err := p.acquire(dc, nodeName, addr)
2013-12-09 22:58:49 +00:00
if err != nil {
return nil, nil, fmt.Errorf("failed to get conn: %v", err)
2013-12-09 22:58:49 +00:00
}
// Get a client
client, err := conn.getClient()
if err != nil {
p.clearConn(conn)
p.releaseConn(conn)
// Try to redial, possible that the TCP session closed due to timeout
if retries == 0 {
retries++
goto START
}
return nil, nil, fmt.Errorf("failed to start stream: %v", err)
}
return conn, client, nil
}
// RPC is used to make an RPC call to a remote host
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
func (p *ConnPool) RPC(
dc string,
nodeName string,
addr net.Addr,
method string,
args interface{},
reply interface{},
) error {
if nodeName == "" {
return fmt.Errorf("pool: ConnPool.RPC requires a node name")
}
// TODO (autoconf) probably will want to have a way to invoke the
// secure or insecure variant depending on whether its an ongoing
// or first time config request. For now though this is fine until
// those ongoing requests are implemented.
if method == "AutoEncrypt.Sign" || method == "AutoConfig.InitialConfiguration" {
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
return p.rpcInsecure(dc, nodeName, addr, method, args, reply)
} else {
return p.rpc(dc, nodeName, addr, method, args, reply)
}
}
// rpcInsecure is used to make an RPC call to a remote host.
// It doesn't actually use any of the pooling, it is here so that it is
// transparent for the consumer. The pool cannot be used because
// AutoEncrypt.Sign is a one-off call and it doesn't make sense to pool that
// connection if it is not being reused.
wan federation via mesh gateways (#6884) This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch: There are several distinct chunks of code that are affected: * new flags and config options for the server * retry join WAN is slightly different * retry join code is shared to discover primary mesh gateways from secondary datacenters * because retry join logic runs in the *agent* and the results of that operation for primary mesh gateways are needed in the *server* there are some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur at multiple layers of abstraction just to pass the data down to the right layer. * new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers * the function signature for RPC dialing picked up a new required field (the node name of the destination) * several new RPCs for manipulating a FederationState object: `FederationState:{Apply,Get,List,ListMeshGateways}` * 3 read-only internal APIs for debugging use to invoke those RPCs from curl * raft and fsm changes to persist these FederationStates * replication for FederationStates as they are canonically stored in the Primary and replicated to the Secondaries. * a special derivative of anti-entropy that runs in secondaries to snapshot their local mesh gateway `CheckServiceNodes` and sync them into their upstream FederationState in the primary (this works in conjunction with the replication to distribute addresses for all mesh gateways in all DCs to all other DCs) * a "gateway locator" convenience object to make use of this data to choose the addresses of gateways to use for any given RPC or gossip operation to a remote DC. This gets data from the "retry join" logic in the agent and also directly calls into the FSM. * RPC (`:8300`) on the server sniffs the first byte of a new connection to determine if it's actually doing native TLS. If so it checks the ALPN header for protocol determination (just like how the existing system uses the type-byte marker). * 2 new kinds of protocols are exclusively decoded via this native TLS mechanism: one for ferrying "packet" operations (udp-like) from the gossip layer and one for "stream" operations (tcp-like). The packet operations re-use sockets (using length-prefixing) to cut down on TLS re-negotiation overhead. * the server instances specially wrap the `memberlist.NetTransport` when running with gateway federation enabled (in a `wanfed.Transport`). The general gist is that if it tries to dial a node in the SAME datacenter (deduced by looking at the suffix of the node name) there is no change. If dialing a DIFFERENT datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh gateways to eventually end up in a server's :8300 port. * a new flag when launching a mesh gateway via `consul connect envoy` to indicate that the servers are to be exposed. This sets a special service meta when registering the gateway into the catalog. * `proxycfg/xds` notice this metadata blob to activate additional watches for the FederationState objects as well as the location of all of the consul servers in that datacenter. * `xds:` if the extra metadata is in place additional clusters are defined in a DC to bulk sink all traffic to another DC's gateways. For the current datacenter we listen on a wildcard name (`server.<dc>.consul`) that load balances all servers as well as one mini-cluster per node (`<node>.server.<dc>.consul`) * the `consul tls cert create` command got a new flag (`-node`) to help create an additional SAN in certs that can be used with this flavor of federation.
2020-03-09 20:59:02 +00:00
func (p *ConnPool) rpcInsecure(dc string, nodeName string, addr net.Addr, method string, args interface{}, reply interface{}) error {
if dc != p.Datacenter {
return fmt.Errorf("insecure dialing prohibited between datacenters")
}
var codec rpc.ClientCodec
conn, _, err := p.dial(dc, nodeName, addr, 0, RPCTLSInsecure)
if err != nil {
return fmt.Errorf("rpcinsecure error establishing connection: %v", err)
}
codec = msgpackrpc.NewCodecFromHandle(true, true, conn, structs.MsgpackHandle)
// Make the RPC call
err = msgpackrpc.CallWithCodec(codec, method, args, reply)
if err != nil {
return fmt.Errorf("rpcinsecure error making call: %v", err)
}
return nil
}
func (p *ConnPool) rpc(dc string, nodeName string, addr net.Addr, method string, args interface{}, reply interface{}) error {
p.once.Do(p.init)
// Get a usable client
conn, sc, err := p.getClient(dc, nodeName, addr)
if err != nil {
return fmt.Errorf("rpc error getting client: %v", err)
}
2013-12-09 22:58:49 +00:00
// Make the RPC call
err = msgpackrpc.CallWithCodec(sc.codec, method, args, reply)
if err != nil {
sc.Close()
// See the comment in leader_test.go TestLeader_ChangeServerID
// about how we found this. The tldr is that if we see this
// error, we know this connection is toast, so we should clear
// it and make a new one on the next attempt.
if lib.IsErrEOF(err) {
p.clearConn(conn)
}
p.releaseConn(conn)
return fmt.Errorf("rpc error making call: %v", err)
2013-12-09 22:58:49 +00:00
}
// Done with the connection
conn.returnClient(sc)
p.releaseConn(conn)
return nil
2013-12-09 22:58:49 +00:00
}
2013-12-19 23:42:17 +00:00
// Ping sends a Status.Ping message to the specified server and
// returns true if healthy, false if an error occurred
func (p *ConnPool) Ping(dc string, nodeName string, addr net.Addr) (bool, error) {
var out struct{}
err := p.RPC(dc, nodeName, addr, "Status.Ping", struct{}{}, &out)
return err == nil, err
}
2013-12-19 23:42:17 +00:00
// Reap is used to close conns open over maxTime
func (p *ConnPool) reap() {
for {
2013-12-19 23:42:17 +00:00
// Sleep for a while
select {
case <-p.shutdownCh:
return
case <-time.After(time.Second):
}
2013-12-19 23:42:17 +00:00
// Reap all old conns
p.Lock()
var removed []string
2013-12-19 23:42:17 +00:00
now := time.Now()
for host, conn := range p.pool {
// Skip recently used connections
if now.Sub(conn.lastUsed) < p.MaxTime {
continue
2013-12-19 23:42:17 +00:00
}
// Skip connections with active streams
if atomic.LoadInt32(&conn.refCount) > 0 {
continue
}
// Close the conn
conn.Close()
// Remove from pool
removed = append(removed, host)
}
for _, host := range removed {
delete(p.pool, host)
2013-12-19 23:42:17 +00:00
}
p.Unlock()
}
}