nim-libp2p/libp2p/protocols/pubsub/pubsubpeer.nim

250 lines
8.0 KiB
Nim
Raw Normal View History

2019-09-10 02:15:52 +00:00
## Nim-LibP2P
2019-09-24 17:48:23 +00:00
## Copyright (c) 2019 Status Research & Development GmbH
2019-09-10 02:15:52 +00:00
## Licensed under either of
## * Apache License, version 2.0, ([LICENSE-APACHE](LICENSE-APACHE))
## * MIT license ([LICENSE-MIT](LICENSE-MIT))
## at your option.
## This file may not be copied, modified, or distributed except according to
## those terms.
import std/[hashes, options, strutils, tables]
2020-06-11 18:09:34 +00:00
import chronos, chronicles, nimcrypto/sha2, metrics
import rpc/[messages, message, protobuf],
2019-10-03 21:34:12 +00:00
timedcache,
../../switch,
../../peerid,
../../peerinfo,
../../stream/connection,
2019-09-12 05:46:08 +00:00
../../crypto/crypto,
../../protobuf/minprotobuf,
../../utility
2019-09-10 02:15:52 +00:00
logScope:
topics = "pubsubpeer"
2019-09-10 02:15:52 +00:00
when defined(libp2p_expensive_metrics):
declareCounter(libp2p_pubsub_sent_messages, "number of messages sent", labels = ["id", "topic"])
declareCounter(libp2p_pubsub_received_messages, "number of messages received", labels = ["id", "topic"])
declareCounter(libp2p_pubsub_skipped_received_messages, "number of received skipped messages", labels = ["id"])
declareCounter(libp2p_pubsub_skipped_sent_messages, "number of sent skipped messages", labels = ["id"])
const
DefaultSendTimeout* = 10.seconds
2019-09-10 02:15:52 +00:00
type
PubSubObserver* = ref object
onRecv*: proc(peer: PubSubPeer; msgs: var RPCMsg) {.gcsafe, raises: [Defect].}
onSend*: proc(peer: PubSubPeer; msgs: var RPCMsg) {.gcsafe, raises: [Defect].}
2019-09-10 02:15:52 +00:00
PubSubPeer* = ref object of RootObj
switch*: Switch # switch instance to dial peers
codec*: string # the protocol that this peer joined from
sendConn: Connection # cached send connection
peerId*: PeerID
handler*: RPCHandler
sentRpcCache: TimedCache[string] # cache for already sent messages
recvdRpcCache: TimedCache[string] # cache for already received messages
observers*: ref seq[PubSubObserver] # ref as in smart_ptr
subscribed*: bool # are we subscribed to this peer
dialLock: AsyncLock
RPCHandler* = proc(peer: PubSubPeer, msg: seq[RPCMsg]): Future[void] {.gcsafe.}
2019-09-10 02:15:52 +00:00
func hash*(p: PubSubPeer): Hash =
# int is either 32/64, so intptr basically, pubsubpeer is a ref
cast[pointer](p).hash
proc id*(p: PubSubPeer): string =
doAssert(not p.isNil, "nil pubsubpeer")
p.peerId.pretty
PubSub (Gossip & Flood) Implementation (#36) This adds gossipsub and floodsub, as well as basic interop testing with the go libp2p daemon. * add close event * wip: gossipsub * splitting rpc message * making message handling more consistent * initial gossipsub implementation * feat: nim 1.0 cleanup * wip: gossipsub protobuf * adding encoding/decoding of gossipsub messages * add disconnect handler * add proper gossipsub msg handling * misc: cleanup for nim 1.0 * splitting floodsub and gossipsub tests * feat: add mesh rebalansing * test pubsub * add mesh rebalansing tests * testing mesh maintenance * finishing mcache implementatin * wip: commenting out broken tests * wip: don't run heartbeat for now * switchout debug for trace logging * testing gossip peer selection algorithm * test stream piping * more work around message amplification * get the peerid from message * use timed cache as backing store * allow setting timeout in constructor * several changes to improve performance * more through testing of msg amplification * prevent gc issues * allow piping to self and prevent deadlocks * improove floodsub * allow running hook on cache eviction * prevent race conditions * prevent race conditions and improove tests * use hashes as cache keys * removing useless file * don't create a new seq * re-enable pubsub tests * fix imports * reduce number of runs to speed up tests * break out control message processing * normalize sleeps between steps * implement proper transport filtering * initial interop testing * clean up floodsub publish logic * allow dialing without a protocol * adding multiple reads/writes * use protobuf varint in mplex * don't loose conn's peerInfo * initial interop pubsub tests * don't duplicate connections/peers * bring back interop tests * wip: interop * re-enable interop and daemon tests * add multiple read write tests from handlers * don't cleanup channel prematurely * use correct channel to send/receive msgs * adjust tests with latest changes * include interop tests * remove temp logging output * fix ci * use correct public key serialization * additional tests for pubsub interop
2019-12-06 02:16:18 +00:00
proc connected*(p: PubSubPeer): bool =
not p.sendConn.isNil and not
(p.sendConn.closed or p.sendConn.atEof)
proc recvObservers(p: PubSubPeer, msg: var RPCMsg) =
# trigger hooks
if not(isNil(p.observers)) and p.observers[].len > 0:
for obs in p.observers[]:
if not(isNil(obs)): # TODO: should never be nil, but...
obs.onRecv(p, msg)
proc sendObservers(p: PubSubPeer, msg: var RPCMsg) =
# trigger hooks
if not(isNil(p.observers)) and p.observers[].len > 0:
for obs in p.observers[]:
if not(isNil(obs)): # TODO: should never be nil, but...
obs.onSend(p, msg)
proc handle*(p: PubSubPeer, conn: Connection) {.async.} =
logScope:
peer = p.id
debug "starting pubsub read loop for peer", closed = conn.closed
2019-09-10 02:15:52 +00:00
try:
2020-05-25 14:33:24 +00:00
try:
while not conn.atEof:
trace "waiting for data", closed = conn.closed
let data = await conn.readLp(64 * 1024)
2020-05-25 14:33:24 +00:00
let digest = $(sha256.digest(data))
trace "read data from peer", data = data.shortLog
2020-05-25 14:33:24 +00:00
if digest in p.recvdRpcCache:
when defined(libp2p_expensive_metrics):
libp2p_pubsub_skipped_received_messages.inc(labelValues = [p.id])
trace "message already received, skipping"
2020-05-25 14:33:24 +00:00
continue
var rmsg = decodeRpcMsg(data)
if rmsg.isErr():
notice "failed to decode msg from peer"
break
var msg = rmsg.get()
trace "decoded msg from peer", msg = msg.shortLog
2020-05-25 14:33:24 +00:00
# trigger hooks
p.recvObservers(msg)
when defined(libp2p_expensive_metrics):
for m in msg.messages:
for t in m.topicIDs:
# metrics
libp2p_pubsub_received_messages.inc(labelValues = [p.id, t])
2020-06-11 18:09:34 +00:00
2020-05-25 14:33:24 +00:00
await p.handler(p, @[msg])
p.recvdRpcCache.put(digest)
finally:
debug "exiting pubsub peer read loop"
2020-05-25 14:33:24 +00:00
await conn.close()
if p.sendConn == conn:
p.sendConn = nil
except CancelledError as exc:
raise exc
PubSub (Gossip & Flood) Implementation (#36) This adds gossipsub and floodsub, as well as basic interop testing with the go libp2p daemon. * add close event * wip: gossipsub * splitting rpc message * making message handling more consistent * initial gossipsub implementation * feat: nim 1.0 cleanup * wip: gossipsub protobuf * adding encoding/decoding of gossipsub messages * add disconnect handler * add proper gossipsub msg handling * misc: cleanup for nim 1.0 * splitting floodsub and gossipsub tests * feat: add mesh rebalansing * test pubsub * add mesh rebalansing tests * testing mesh maintenance * finishing mcache implementatin * wip: commenting out broken tests * wip: don't run heartbeat for now * switchout debug for trace logging * testing gossip peer selection algorithm * test stream piping * more work around message amplification * get the peerid from message * use timed cache as backing store * allow setting timeout in constructor * several changes to improve performance * more through testing of msg amplification * prevent gc issues * allow piping to self and prevent deadlocks * improove floodsub * allow running hook on cache eviction * prevent race conditions * prevent race conditions and improove tests * use hashes as cache keys * removing useless file * don't create a new seq * re-enable pubsub tests * fix imports * reduce number of runs to speed up tests * break out control message processing * normalize sleeps between steps * implement proper transport filtering * initial interop testing * clean up floodsub publish logic * allow dialing without a protocol * adding multiple reads/writes * use protobuf varint in mplex * don't loose conn's peerInfo * initial interop pubsub tests * don't duplicate connections/peers * bring back interop tests * wip: interop * re-enable interop and daemon tests * add multiple read write tests from handlers * don't cleanup channel prematurely * use correct channel to send/receive msgs * adjust tests with latest changes * include interop tests * remove temp logging output * fix ci * use correct public key serialization * additional tests for pubsub interop
2019-12-06 02:16:18 +00:00
except CatchableError as exc:
trace "Exception occurred in PubSubPeer.handle", exc = exc.msg
2019-09-10 02:15:52 +00:00
proc getSendConn(p: PubSubPeer): Future[Connection] {.async.} =
# get a cached send connection or create a new one
block: # check if there's an existing connection that can be reused
let current = p.sendConn
if not current.isNil:
if not (current.closed() or current.atEof):
# The existing send connection looks like it might work - reuse it
2020-08-17 10:39:25 +00:00
trace "Reusing existing connection", oid = $current.oid
return current
# Send connection is set but broken - get rid of it
p.sendConn = nil
# Careful, p.sendConn might change after here!
await current.close() # TODO this might be unnecessary
try:
# Testing has demonstrated that when we perform concurrent meshsub dials
# and later close one of them, other implementations such as rust-libp2p
# become deaf to our messages (potentially due to the clean-up associated
# with closing connections). To prevent this, we use a lock that ensures
# that only a single dial will be performed for each peer:
await p.dialLock.acquire()
# Another concurrent dial may have populated p.sendConn
if p.sendConn != nil:
let current = p.sendConn
if not current.isNil:
if not (current.closed() or current.atEof):
# The existing send connection looks like it might work - reuse it
trace "Reusing existing connection", oid = $current.oid
return current
# Grab a new send connection
let newConn = await p.switch.dial(p.peerId, p.codec) # ...and here
if newConn.isNil:
return nil
trace "Caching new send connection", oid = $newConn.oid
p.sendConn = newConn
asyncCheck p.handle(newConn) # start a read loop on the new connection
return newConn
finally:
if p.dialLock.locked:
p.dialLock.release()
proc send*(
p: PubSubPeer,
msg: RPCMsg,
timeout: Duration = DefaultSendTimeout) {.async.} =
doAssert(not isNil(p), "pubsubpeer nil!")
logScope:
peer = p.id
rpcMsg = shortLog(msg)
trace "sending msg to peer"
# trigger send hooks
var mm = msg # hooks can modify the message
p.sendObservers(mm)
let encoded = encodeRpcMsg(mm)
if encoded.len <= 0:
info "empty message, skipping"
return
PubSub (Gossip & Flood) Implementation (#36) This adds gossipsub and floodsub, as well as basic interop testing with the go libp2p daemon. * add close event * wip: gossipsub * splitting rpc message * making message handling more consistent * initial gossipsub implementation * feat: nim 1.0 cleanup * wip: gossipsub protobuf * adding encoding/decoding of gossipsub messages * add disconnect handler * add proper gossipsub msg handling * misc: cleanup for nim 1.0 * splitting floodsub and gossipsub tests * feat: add mesh rebalansing * test pubsub * add mesh rebalansing tests * testing mesh maintenance * finishing mcache implementatin * wip: commenting out broken tests * wip: don't run heartbeat for now * switchout debug for trace logging * testing gossip peer selection algorithm * test stream piping * more work around message amplification * get the peerid from message * use timed cache as backing store * allow setting timeout in constructor * several changes to improve performance * more through testing of msg amplification * prevent gc issues * allow piping to self and prevent deadlocks * improove floodsub * allow running hook on cache eviction * prevent race conditions * prevent race conditions and improove tests * use hashes as cache keys * removing useless file * don't create a new seq * re-enable pubsub tests * fix imports * reduce number of runs to speed up tests * break out control message processing * normalize sleeps between steps * implement proper transport filtering * initial interop testing * clean up floodsub publish logic * allow dialing without a protocol * adding multiple reads/writes * use protobuf varint in mplex * don't loose conn's peerInfo * initial interop pubsub tests * don't duplicate connections/peers * bring back interop tests * wip: interop * re-enable interop and daemon tests * add multiple read write tests from handlers * don't cleanup channel prematurely * use correct channel to send/receive msgs * adjust tests with latest changes * include interop tests * remove temp logging output * fix ci * use correct public key serialization * additional tests for pubsub interop
2019-12-06 02:16:18 +00:00
logScope:
encoded = shortLog(encoded)
let digest = $(sha256.digest(encoded))
if digest in p.sentRpcCache:
trace "message already sent to peer, skipping"
when defined(libp2p_expensive_metrics):
libp2p_pubsub_skipped_sent_messages.inc(labelValues = [p.id])
return
2020-05-23 16:50:29 +00:00
var conn: Connection
try:
trace "about to send message"
conn = await p.getSendConn()
if conn == nil:
debug "Couldn't get send connection, dropping message"
return
trace "sending encoded msgs to peer", connId = $conn.oid
await conn.writeLp(encoded).wait(timeout)
p.sentRpcCache.put(digest)
trace "sent pubsub message to remote", connId = $conn.oid
when defined(libp2p_expensive_metrics):
for x in mm.messages:
for t in x.topicIDs:
# metrics
libp2p_pubsub_sent_messages.inc(labelValues = [p.id, t])
except CatchableError as exc:
trace "unable to send to remote", exc = exc.msg
# Next time sendConn is used, it will be have its close flag set and thus
# will be recycled
if not isNil(conn):
await conn.close()
raise exc
PubSub (Gossip & Flood) Implementation (#36) This adds gossipsub and floodsub, as well as basic interop testing with the go libp2p daemon. * add close event * wip: gossipsub * splitting rpc message * making message handling more consistent * initial gossipsub implementation * feat: nim 1.0 cleanup * wip: gossipsub protobuf * adding encoding/decoding of gossipsub messages * add disconnect handler * add proper gossipsub msg handling * misc: cleanup for nim 1.0 * splitting floodsub and gossipsub tests * feat: add mesh rebalansing * test pubsub * add mesh rebalansing tests * testing mesh maintenance * finishing mcache implementatin * wip: commenting out broken tests * wip: don't run heartbeat for now * switchout debug for trace logging * testing gossip peer selection algorithm * test stream piping * more work around message amplification * get the peerid from message * use timed cache as backing store * allow setting timeout in constructor * several changes to improve performance * more through testing of msg amplification * prevent gc issues * allow piping to self and prevent deadlocks * improove floodsub * allow running hook on cache eviction * prevent race conditions * prevent race conditions and improove tests * use hashes as cache keys * removing useless file * don't create a new seq * re-enable pubsub tests * fix imports * reduce number of runs to speed up tests * break out control message processing * normalize sleeps between steps * implement proper transport filtering * initial interop testing * clean up floodsub publish logic * allow dialing without a protocol * adding multiple reads/writes * use protobuf varint in mplex * don't loose conn's peerInfo * initial interop pubsub tests * don't duplicate connections/peers * bring back interop tests * wip: interop * re-enable interop and daemon tests * add multiple read write tests from handlers * don't cleanup channel prematurely * use correct channel to send/receive msgs * adjust tests with latest changes * include interop tests * remove temp logging output * fix ci * use correct public key serialization * additional tests for pubsub interop
2019-12-06 02:16:18 +00:00
proc `$`*(p: PubSubPeer): string =
p.id
proc newPubSubPeer*(peerId: PeerID,
switch: Switch,
codec: string): PubSubPeer =
2019-09-10 02:15:52 +00:00
new result
result.switch = switch
result.codec = codec
result.peerId = peerId
PubSub (Gossip & Flood) Implementation (#36) This adds gossipsub and floodsub, as well as basic interop testing with the go libp2p daemon. * add close event * wip: gossipsub * splitting rpc message * making message handling more consistent * initial gossipsub implementation * feat: nim 1.0 cleanup * wip: gossipsub protobuf * adding encoding/decoding of gossipsub messages * add disconnect handler * add proper gossipsub msg handling * misc: cleanup for nim 1.0 * splitting floodsub and gossipsub tests * feat: add mesh rebalansing * test pubsub * add mesh rebalansing tests * testing mesh maintenance * finishing mcache implementatin * wip: commenting out broken tests * wip: don't run heartbeat for now * switchout debug for trace logging * testing gossip peer selection algorithm * test stream piping * more work around message amplification * get the peerid from message * use timed cache as backing store * allow setting timeout in constructor * several changes to improve performance * more through testing of msg amplification * prevent gc issues * allow piping to self and prevent deadlocks * improove floodsub * allow running hook on cache eviction * prevent race conditions * prevent race conditions and improove tests * use hashes as cache keys * removing useless file * don't create a new seq * re-enable pubsub tests * fix imports * reduce number of runs to speed up tests * break out control message processing * normalize sleeps between steps * implement proper transport filtering * initial interop testing * clean up floodsub publish logic * allow dialing without a protocol * adding multiple reads/writes * use protobuf varint in mplex * don't loose conn's peerInfo * initial interop pubsub tests * don't duplicate connections/peers * bring back interop tests * wip: interop * re-enable interop and daemon tests * add multiple read write tests from handlers * don't cleanup channel prematurely * use correct channel to send/receive msgs * adjust tests with latest changes * include interop tests * remove temp logging output * fix ci * use correct public key serialization * additional tests for pubsub interop
2019-12-06 02:16:18 +00:00
result.sentRpcCache = newTimedCache[string](2.minutes)
result.recvdRpcCache = newTimedCache[string](2.minutes)
result.dialLock = newAsyncLock()