consul/agent/proxycfg/manager_test.go
R.B. Boyer 8e22d80e35
connect: fix failover through a mesh gateway to a remote datacenter (#6259)
Failover is pushed entirely down to the data plane by creating envoy
clusters and putting each successive destination in a different load
assignment priority band. For example this shows that normally requests
go to 1.2.3.4:8080 but when that fails they go to 6.7.8.9:8080:

- name: foo
  load_assignment:
    cluster_name: foo
    policy:
      overprovisioning_factor: 100000
    endpoints:
    - priority: 0
      lb_endpoints:
      - endpoint:
          address:
            socket_address:
              address: 1.2.3.4
              port_value: 8080
    - priority: 1
      lb_endpoints:
      - endpoint:
          address:
            socket_address:
              address: 6.7.8.9
              port_value: 8080

Mesh gateways route requests based solely on the SNI header tacked onto
the TLS layer. Envoy currently only lets you configure the outbound SNI
header at the cluster layer.

If you try to failover through a mesh gateway you ideally would
configure the SNI value per endpoint, but that's not possible in envoy
today.

This PR introduces a simpler way around the problem for now:

1. We identify any target of failover that will use mesh gateway mode local or
   remote and then further isolate any resolver node in the compiled discovery
   chain that has a failover destination set to one of those targets.

2. For each of these resolvers we will perform a small measurement of
   comparative healths of the endpoints that come back from the health API for the
   set of primary target and serial failover targets. We walk the list of targets
   in order and if any endpoint is healthy we return that target, otherwise we
   move on to the next target.

3. The CDS and EDS endpoints both perform the measurements in (2) for the
   affected resolver nodes.

4. For CDS this measurement selects which TLS SNI field to use for the cluster
   (note the cluster is always going to be named for the primary target)

5. For EDS this measurement selects which set of endpoints will populate the
   cluster. Priority tiered failover is ignored.

One of the big downsides to this approach to failover is that the failover
detection and correction is going to be controlled by consul rather than
deferring that entirely to the data plane as with the prior version. This also
means that we are bound to only failover using official health signals and
cannot make use of data plane signals like outlier detection to affect
failover.

In this specific scenario the lack of data plane signals is ok because the
effectiveness is already muted by the fact that the ultimate destination
endpoints will have their data plane signals scrambled when they pass through
the mesh gateway wrapper anyway so we're not losing much.

Another related fix is that we now use the endpoint health from the
underlying service, not the health of the gateway (regardless of
failover mode).
2019-08-05 13:30:35 -05:00

516 lines
15 KiB
Go

package proxycfg
import (
"log"
"os"
"path"
"testing"
"time"
"github.com/mitchellh/copystructure"
"github.com/stretchr/testify/require"
"github.com/hashicorp/consul/agent/cache"
cachetype "github.com/hashicorp/consul/agent/cache-types"
"github.com/hashicorp/consul/agent/consul/discoverychain"
"github.com/hashicorp/consul/agent/local"
"github.com/hashicorp/consul/agent/structs"
"github.com/hashicorp/consul/agent/token"
)
// assertLastReqArgs verifies that each request type had the correct source
// parameters (e.g. Datacenter name) and token.
func assertLastReqArgs(t *testing.T, types *TestCacheTypes, token string, source *structs.QuerySource) {
t.Helper()
// Roots needs correct DC and token
rootReq := types.roots.lastReq.Load()
require.IsType(t, rootReq, &structs.DCSpecificRequest{})
require.Equal(t, token, rootReq.(*structs.DCSpecificRequest).Token)
require.Equal(t, source.Datacenter, rootReq.(*structs.DCSpecificRequest).Datacenter)
// Leaf needs correct DC and token
leafReq := types.leaf.lastReq.Load()
require.IsType(t, leafReq, &cachetype.ConnectCALeafRequest{})
require.Equal(t, token, leafReq.(*cachetype.ConnectCALeafRequest).Token)
require.Equal(t, source.Datacenter, leafReq.(*cachetype.ConnectCALeafRequest).Datacenter)
// Intentions needs correct DC and token
intReq := types.intentions.lastReq.Load()
require.IsType(t, intReq, &structs.IntentionQueryRequest{})
require.Equal(t, token, intReq.(*structs.IntentionQueryRequest).Token)
require.Equal(t, source.Datacenter, intReq.(*structs.IntentionQueryRequest).Datacenter)
}
func TestManager_BasicLifecycle(t *testing.T) {
// Create a bunch of common data for the various test cases.
roots, leaf := TestCerts(t)
dbDefaultChain := func() *structs.CompiledDiscoveryChain {
return discoverychain.TestCompileConfigEntries(t, "db", "default", "dc1", "dc1",
func(req *discoverychain.CompileRequest) {
// This is because structs.TestUpstreams uses an opaque config
// to override connect timeouts.
req.OverrideConnectTimeout = 1 * time.Second
},
&structs.ServiceResolverConfigEntry{
Kind: structs.ServiceResolver,
Name: "db",
},
)
}
dbSplitChain := func() *structs.CompiledDiscoveryChain {
return discoverychain.TestCompileConfigEntries(t, "db", "default", "dc1", "dc1",
func(req *discoverychain.CompileRequest) {
// This is because structs.TestUpstreams uses an opaque config
// to override connect timeouts.
req.OverrideConnectTimeout = 1 * time.Second
},
&structs.ProxyConfigEntry{
Kind: structs.ProxyDefaults,
Name: structs.ProxyConfigGlobal,
Config: map[string]interface{}{
"protocol": "http",
},
},
&structs.ServiceResolverConfigEntry{
Kind: structs.ServiceResolver,
Name: "db",
Subsets: map[string]structs.ServiceResolverSubset{
"v1": {
Filter: "Service.Meta.version == v1",
},
"v2": {
Filter: "Service.Meta.version == v2",
},
},
},
&structs.ServiceSplitterConfigEntry{
Kind: structs.ServiceSplitter,
Name: "db",
Splits: []structs.ServiceSplit{
{Weight: 60, ServiceSubset: "v1"},
{Weight: 40, ServiceSubset: "v2"},
},
},
)
}
webProxy := &structs.NodeService{
Kind: structs.ServiceKindConnectProxy,
ID: "web-sidecar-proxy",
Service: "web-sidecar-proxy",
Port: 9999,
Proxy: structs.ConnectProxyConfig{
DestinationServiceID: "web",
DestinationServiceName: "web",
LocalServiceAddress: "127.0.0.1",
LocalServicePort: 8080,
Config: map[string]interface{}{
"foo": "bar",
},
Upstreams: structs.TestUpstreams(t),
},
}
rootsCacheKey := testGenCacheKey(&structs.DCSpecificRequest{
Datacenter: "dc1",
QueryOptions: structs.QueryOptions{Token: "my-token"},
})
leafCacheKey := testGenCacheKey(&cachetype.ConnectCALeafRequest{
Datacenter: "dc1",
Token: "my-token",
Service: "web",
})
intentionCacheKey := testGenCacheKey(&structs.IntentionQueryRequest{
Datacenter: "dc1",
QueryOptions: structs.QueryOptions{Token: "my-token"},
Match: &structs.IntentionQueryMatch{
Type: structs.IntentionMatchDestination,
Entries: []structs.IntentionMatchEntry{
{
Namespace: structs.IntentionDefaultNamespace,
Name: "web",
},
},
},
})
dbChainCacheKey := testGenCacheKey(&structs.DiscoveryChainRequest{
Name: "db",
EvaluateInDatacenter: "dc1",
EvaluateInNamespace: "default",
// This is because structs.TestUpstreams uses an opaque config
// to override connect timeouts.
OverrideConnectTimeout: 1 * time.Second,
Datacenter: "dc1",
QueryOptions: structs.QueryOptions{Token: "my-token"},
})
dbHealthCacheKey := testGenCacheKey(&structs.ServiceSpecificRequest{
Datacenter: "dc1",
QueryOptions: structs.QueryOptions{Token: "my-token", Filter: ""},
ServiceName: "db",
Connect: true,
})
db_v1_HealthCacheKey := testGenCacheKey(&structs.ServiceSpecificRequest{
Datacenter: "dc1",
QueryOptions: structs.QueryOptions{Token: "my-token",
Filter: "Service.Meta.version == v1",
},
ServiceName: "db",
Connect: true,
})
db_v2_HealthCacheKey := testGenCacheKey(&structs.ServiceSpecificRequest{
Datacenter: "dc1",
QueryOptions: structs.QueryOptions{Token: "my-token",
Filter: "Service.Meta.version == v2",
},
ServiceName: "db",
Connect: true,
})
// Create test cases using some of the common data above.
tests := []*testcase_BasicLifecycle{
{
name: "simple-default-resolver",
setup: func(t *testing.T, types *TestCacheTypes) {
// Note that we deliberately leave the 'geo-cache' prepared query to time out
types.health.Set(dbHealthCacheKey, &structs.IndexedCheckServiceNodes{
Nodes: TestUpstreamNodes(t),
})
types.compiledChain.Set(dbChainCacheKey, &structs.DiscoveryChainResponse{
Chain: dbDefaultChain(),
})
},
expectSnap: &ConfigSnapshot{
Kind: structs.ServiceKindConnectProxy,
Service: webProxy.Service,
ProxyID: webProxy.ID,
Address: webProxy.Address,
Port: webProxy.Port,
Proxy: webProxy.Proxy,
TaggedAddresses: make(map[string]structs.ServiceAddress),
Roots: roots,
ConnectProxy: configSnapshotConnectProxy{
Leaf: leaf,
DiscoveryChain: map[string]*structs.CompiledDiscoveryChain{
"db": dbDefaultChain(),
},
WatchedUpstreams: nil, // Clone() clears this out
WatchedUpstreamEndpoints: map[string]map[string]structs.CheckServiceNodes{
"db": {
"db.default.dc1": TestUpstreamNodes(t),
},
},
WatchedGateways: nil, // Clone() clears this out
WatchedGatewayEndpoints: map[string]map[string]structs.CheckServiceNodes{
"db": {},
},
UpstreamEndpoints: map[string]structs.CheckServiceNodes{},
},
Datacenter: "dc1",
},
},
{
name: "chain-resolver-with-version-split",
setup: func(t *testing.T, types *TestCacheTypes) {
// Note that we deliberately leave the 'geo-cache' prepared query to time out
types.health.Set(db_v1_HealthCacheKey, &structs.IndexedCheckServiceNodes{
Nodes: TestUpstreamNodes(t),
})
types.health.Set(db_v2_HealthCacheKey, &structs.IndexedCheckServiceNodes{
Nodes: TestUpstreamNodesAlternate(t),
})
types.compiledChain.Set(dbChainCacheKey, &structs.DiscoveryChainResponse{
Chain: dbSplitChain(),
})
},
expectSnap: &ConfigSnapshot{
Kind: structs.ServiceKindConnectProxy,
Service: webProxy.Service,
ProxyID: webProxy.ID,
Address: webProxy.Address,
Port: webProxy.Port,
Proxy: webProxy.Proxy,
TaggedAddresses: make(map[string]structs.ServiceAddress),
Roots: roots,
ConnectProxy: configSnapshotConnectProxy{
Leaf: leaf,
DiscoveryChain: map[string]*structs.CompiledDiscoveryChain{
"db": dbSplitChain(),
},
WatchedUpstreams: nil, // Clone() clears this out
WatchedUpstreamEndpoints: map[string]map[string]structs.CheckServiceNodes{
"db": {
"v1.db.default.dc1": TestUpstreamNodes(t),
"v2.db.default.dc1": TestUpstreamNodesAlternate(t),
},
},
WatchedGateways: nil, // Clone() clears this out
WatchedGatewayEndpoints: map[string]map[string]structs.CheckServiceNodes{
"db": {},
},
UpstreamEndpoints: map[string]structs.CheckServiceNodes{},
},
Datacenter: "dc1",
},
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
require.NotNil(t, tt.setup)
require.NotNil(t, tt.expectSnap)
// Use a mocked cache to make life simpler
types := NewTestCacheTypes(t)
// Setup initial values
types.roots.Set(rootsCacheKey, roots)
types.leaf.Set(leafCacheKey, leaf)
types.intentions.Set(intentionCacheKey, TestIntentions(t))
tt.setup(t, types)
expectSnapCopy, err := copystructure.Copy(tt.expectSnap)
require.NoError(t, err)
webProxyCopy, err := copystructure.Copy(webProxy)
require.NoError(t, err)
testManager_BasicLifecycle(t, tt, types,
rootsCacheKey, leafCacheKey,
roots, leaf,
webProxyCopy.(*structs.NodeService),
expectSnapCopy.(*ConfigSnapshot),
)
})
}
}
type testcase_BasicLifecycle struct {
name string
setup func(t *testing.T, types *TestCacheTypes)
webProxy *structs.NodeService
expectSnap *ConfigSnapshot
}
func testManager_BasicLifecycle(
t *testing.T,
tt *testcase_BasicLifecycle,
types *TestCacheTypes,
rootsCacheKey, leafCacheKey string,
roots *structs.IndexedCARoots,
leaf *structs.IssuedCert,
webProxy *structs.NodeService,
expectSnap *ConfigSnapshot,
) {
c := TestCacheWithTypes(t, types)
require := require.New(t)
logger := log.New(os.Stderr, "", log.LstdFlags)
state := local.NewState(local.Config{}, logger, &token.Store{})
source := &structs.QuerySource{
Node: "node1",
Datacenter: "dc1",
}
// Stub state syncing
state.TriggerSyncChanges = func() {}
// Create manager
m, err := NewManager(ManagerConfig{c, state, source, logger})
require.NoError(err)
// And run it
go func() {
err := m.Run()
require.NoError(err)
}()
// BEFORE we register, we should be able to get a watch channel
wCh, cancel := m.Watch(webProxy.ID)
defer cancel()
// And it should block with nothing sent on it yet
assertWatchChanBlocks(t, wCh)
require.NoError(state.AddService(webProxy, "my-token"))
// We should see the initial config delivered but not until after the
// coalesce timeout
start := time.Now()
assertWatchChanRecvs(t, wCh, expectSnap)
require.True(time.Since(start) >= coalesceTimeout)
assertLastReqArgs(t, types, "my-token", source)
// Update NodeConfig
webProxy.Port = 7777
require.NoError(state.AddService(webProxy, "my-token"))
expectSnap.Port = 7777
assertWatchChanRecvs(t, wCh, expectSnap)
// Register a second watcher
wCh2, cancel2 := m.Watch(webProxy.ID)
defer cancel2()
// New watcher should immediately receive the current state
assertWatchChanRecvs(t, wCh2, expectSnap)
// Change token
require.NoError(state.AddService(webProxy, "other-token"))
assertWatchChanRecvs(t, wCh, expectSnap)
assertWatchChanRecvs(t, wCh2, expectSnap)
// This is actually sort of timing dependent - the cache background fetcher
// will still be fetching with the old token, but we rely on the fact that our
// mock type will have been blocked on those for a while.
assertLastReqArgs(t, types, "other-token", source)
// Update roots
newRoots, newLeaf := TestCerts(t)
newRoots.Roots = append(newRoots.Roots, roots.Roots...)
types.roots.Set(rootsCacheKey, newRoots)
// Expect new roots in snapshot
expectSnap.Roots = newRoots
assertWatchChanRecvs(t, wCh, expectSnap)
assertWatchChanRecvs(t, wCh2, expectSnap)
// Update leaf
types.leaf.Set(leafCacheKey, newLeaf)
// Expect new roots in snapshot
expectSnap.ConnectProxy.Leaf = newLeaf
assertWatchChanRecvs(t, wCh, expectSnap)
assertWatchChanRecvs(t, wCh2, expectSnap)
// Remove the proxy
state.RemoveService(webProxy.ID)
// Chan should NOT close
assertWatchChanBlocks(t, wCh)
assertWatchChanBlocks(t, wCh2)
// Re-add the proxy with another new port
webProxy.Port = 3333
require.NoError(state.AddService(webProxy, "other-token"))
// Same watch chan should be notified again
expectSnap.Port = 3333
assertWatchChanRecvs(t, wCh, expectSnap)
assertWatchChanRecvs(t, wCh2, expectSnap)
// Cancel watch
cancel()
// Watch chan should be closed
assertWatchChanRecvs(t, wCh, nil)
// We specifically don't remove the proxy or cancel the second watcher to
// ensure both are cleaned up by close.
require.NoError(m.Close())
// Sanity check the state is clean
m.mu.Lock()
defer m.mu.Unlock()
require.Len(m.proxies, 0)
require.Len(m.watchers, 0)
}
func assertWatchChanBlocks(t *testing.T, ch <-chan *ConfigSnapshot) {
t.Helper()
select {
case <-ch:
t.Fatal("Should be nothing sent on watch chan yet")
default:
}
}
func assertWatchChanRecvs(t *testing.T, ch <-chan *ConfigSnapshot, expect *ConfigSnapshot) {
t.Helper()
select {
case got, ok := <-ch:
require.Equal(t, expect, got)
if expect == nil {
require.False(t, ok, "watch chan should be closed")
}
case <-time.After(100*time.Millisecond + coalesceTimeout):
t.Fatal("recv timeout")
}
}
func TestManager_deliverLatest(t *testing.T) {
// None of these need to do anything to test this method just be valid
logger := log.New(os.Stderr, "", log.LstdFlags)
cfg := ManagerConfig{
Cache: cache.New(nil),
State: local.NewState(local.Config{}, logger, &token.Store{}),
Source: &structs.QuerySource{
Node: "node1",
Datacenter: "dc1",
},
Logger: logger,
}
require := require.New(t)
m, err := NewManager(cfg)
require.NoError(err)
snap1 := &ConfigSnapshot{
ProxyID: "test-proxy",
Port: 1111,
}
snap2 := &ConfigSnapshot{
ProxyID: "test-proxy",
Port: 2222,
}
// Put an overall time limit on this test case so we don't have to guard every
// call to ensure the whole test doesn't deadlock.
time.AfterFunc(100*time.Millisecond, func() {
t.Fatal("test timed out")
})
// test 1 buffered chan
ch1 := make(chan *ConfigSnapshot, 1)
// Sending to an unblocked chan should work
m.deliverLatest(snap1, ch1)
// Check it was delivered
require.Equal(snap1, <-ch1)
// Now send both without reading simulating a slow client
m.deliverLatest(snap1, ch1)
m.deliverLatest(snap2, ch1)
// Check we got the _second_ one
require.Equal(snap2, <-ch1)
// Same again for 5-buffered chan
ch5 := make(chan *ConfigSnapshot, 5)
// Sending to an unblocked chan should work
m.deliverLatest(snap1, ch5)
// Check it was delivered
require.Equal(snap1, <-ch5)
// Now send enough to fill the chan simulating a slow client
for i := 0; i < 5; i++ {
m.deliverLatest(snap1, ch5)
}
m.deliverLatest(snap2, ch5)
// Check we got the _second_ one
require.Equal(snap2, <-ch5)
}
func testGenCacheKey(req cache.Request) string {
info := req.CacheInfo()
return path.Join(info.Key, info.Datacenter)
}