This change changes the ABCI socket client to allow goroutines to block writing to the internal queue. This has the effect ensuring that callers of the ABCI methods do not error on a full internal queue at the expense of allowing the number of goroutines waiting on this internal queue to grow in an unbounded fashion. This tradeoff seems preferable since it allows callers of the ABCI methods to be certain that a request that was made will reach the application if it is available.
This change was initially implemented here: e13b4386ff and never landed on v0.34, only v0.35+
The main effect of this change is to flush the socket client and server message
encoding buffers immediately once the message is fully and correctly encoded.
This allows us to remove the timer and some other special cases, without
changing the observed behaviour of the system.
-- Background
The socket protocol client and server each use a buffered writer to encode
request and response messages onto the underlying connection. This reduces the
possibility of a single message being split across multiple writes, but has the
side-effect that a request may remain buffered for some time.
The implementation worked around this by keeping a ticker that occasionally
triggers a flush, and by flushing the writer in response to an explicit request
baked into the client/server protocol (see also #6994).
These workarounds are both unnecessary: Once a message has been dequeued for
sending and fully encoded in wire format, there is no real use keeping all or
part of it buffered locally. Moreover, using an asynchronous process to flush
the buffer makes the round-trip performance of the request unpredictable.
-- Benchmarks
Code: https://play.golang.org/p/0ChUOxJOiHt
I found no pre-existing performance benchmarks to justify the flush pattern,
but a natural question is whether this will significantly harm client/server
performance. To test this, I implemented a simple benchmark that transfers
randomly-sized byte buffers from a no-op "client" to a no-op "server" over a
Unix-domain socket, using a buffered writer, both with and without explicit
flushes after each write.
As the following data show, flushing every time (FLUSH=true) does reduce raw
throughput, but not by a significant amount except for very small request
sizes, where the transfer time is already trivial (1.9μs). Given that the
client is calibrated for 1MiB transactions, the overhead is not meaningful.
The percentage in each section is the speedup for flushing only when the buffer
is full, relative to flushing every block. The benchmark uses the default
buffer size (4096 bytes), which is the same value used by the socket client and
server implementation:
FLUSH NBLOCKS MAX AVG TOTAL ELAPSED TIME/BLOCK
false 3957471 512 255 1011165416 2.00018873s 505ns
true 1068568 512 255 273064368 2.000217051s 1.871µs
(73%)
false 536096 4096 2048 1098066401 2.000229108s 3.731µs
true 477911 4096 2047 978746731 2.000177825s 4.185µs
(10.8%)
false 124595 16384 8181 1019340160 2.000235086s 16.053µs
true 120995 16384 8179 989703064 2.000329349s 16.532µs
(2.9%)
false 2114 1048576 525693 1111316541 2.000479928s 946.3µs
true 2083 1048576 526379 1096449173 2.001817137s 961.025µs
(1.5%)
Note also that the FLUSH=false baseline is actually faster than the production
code, which flushes more often than is required by the buffer filling up.
Moreover, the timer slows down the overall transaction rate of the client and
server, indepenedent of how fast the socket transfer is, so the loss on a real
workload is probably much less.
## Description
Internalize some libs. This reduces the amount ot public API tendermint is supporting. The moved libraries are mainly ones that are used within Tendermint-core.
Fixes the race condition between a callback being set and called during ReCheckTx. Note, I do not see equivalent logic in the gRPC client (anymore) as #5439 suggests, so only the socket client was updated.
closes: #5439
`abci.Client`:
- Sync and Async methods now accept a context for cancellation
* grpc client uses context to cancel both Sync and Async requests
* local client ignores context parameter
* socket client uses context to cancel Sync requests and to drop Async requests before sending them if context was cancelled prior to that
- Async methods return an error
* socket client returns an error immediately if queue is full for Async requests
* local client always returns nil error
* grpc client returns an error if context was cancelled before we got response or the receiving queue had a space for response (do not confuse with the sending queue from the socket client)
- specify clients semantics in [doc.go](https://raw.githubusercontent.com/tendermint/tendermint/27112fffa62276bc016d56741f686f0f77931748/abci/client/doc.go)
`mempool.TxInfo`
- add optional `Context` to `TxInfo`, which can be used to cancel `CheckTx` request
Closes#5190
* libs/bits: inline defer and change order of mutexes
Closes#3217
* abci/client: unexpose StopForError func
a) it's not called outside
b) the reason for exposing it in the first place is unclear
c) Stop already exist if someone from outside wants to stop the client
## Description
This PR wraps the stdlib sync.(RW)Mutex & godeadlock.(RW)Mutex. This enables using go-deadlock via a build flag instead of using sed to replace sync with godeadlock in all files
Closes: #3242
* libs/common: Refactor libs/common 5
- move mathematical functions and types out of `libs/common` to math pkg
- move net functions out of `libs/common` to net pkg
- move string functions out of `libs/common` to strings pkg
- move async functions out of `libs/common` to async pkg
- move bit functions out of `libs/common` to bits pkg
- move cmap functions out of `libs/common` to cmap pkg
- move os functions out of `libs/common` to os pkg
Signed-off-by: Marko Baricevic <marbar3778@yahoo.com>
* fix testing issues
* fix tests
closes#41417
woooooooooooooooooo kill the cmn pkg
Signed-off-by: Marko Baricevic <marbar3778@yahoo.com>
* add changelog entry
* fix goimport issues
* run gofmt
* libs/common: Refactor libs/common 4
- move byte function out of cmn to its own pkg
- move tempfile out of cmn to its own pkg
- move throttletimer to its own pkg
ref #4147
Signed-off-by: Marko Baricevic <marbar3778@yahoo.com>
* add changelog entry
* fix linting issues
* libs/common: refactor libs common 3
- move nil.go into types folder and make private
- move service & baseservice out of common into service pkg
ref #4147
Signed-off-by: Marko Baricevic <marbar3778@yahoo.com>
* add changelog entry
Follow up from #3512
Specifically:
cli.conn.Close() need not be under the mutex (#3512 (comment))
call the reqRes callback after the resCb so they always happen in the same order (#3512)
Fixes#3513
As per #2127, this refactors the RequestCheckTx ProtoBuf struct to allow for a flag indicating whether a query is a recheck or not (and allows for possible future, more nuanced states).
In order to pass this extended information through to the ABCI app, the proxy.AppConnMempool (and, for consistency, the proxy.AppConnConsensus) interface seems to need to be refactored along with abcicli.Client.
And, as per this comment, I've made the following modification to the protobuf definition for the RequestCheckTx structure:
enum CheckTxType {
New = 0;
Recheck = 1;
}
message RequestCheckTx {
bytes tx = 1;
CheckTxType type = 2;
}
* Refactor ABCI CheckTx to notify of recheck
As per #2127, this refactors the `RequestCheckTx` ProtoBuf struct to allow for:
1. a flag indicating whether a query is a recheck or not (and allows for
possible future, more nuanced states)
2. an `additional_data` bytes array to provide information for those more
nuanced states.
In order to pass this extended information through to the ABCI app, the
`proxy.AppConnMempool` (and, for consistency, the
`proxy.AppConnConsensus`) interface seems to need to be refactored.
Commits:
* Fix linting issue
* Add CHANGELOG_PENDING entry
* Remove extraneous explicit initialization
* Update ABCI spec doc to include new CheckTx params
* Rename method param for consistency
* Rename CheckTxType enum values and remove additional_data param
* use READ lock/unlock in ConsensusState#GetLastHeight
Refs #2721
* do not use defers when there's no need
* fix peer formatting (output its address instead of the pointer)
```
[54310]: E[11-02|11:59:39.851] Connection failed @ sendRoutine module=p2p peer=0xb78f00 conn=MConn{74.207.236.148:26656} err="pong timeout"
```
https://github.com/tendermint/tendermint/issues/2721#issuecomment-435326581
* panic if peer has no state
https://github.com/tendermint/tendermint/issues/2721#issuecomment-435347165
It's confusing that sometimes we check if peer has a state, but most of
the times we expect it to be there
1. add79700b5/mempool/reactor.go (L138)
2. add79700b5/rpc/core/consensus.go (L196) (edited)
I will change everything to always assume peer has a state and panic
otherwise
that should help identify issues earlier
* abci/localclient: extend lock on app callback
App callback should be protected by lock as well (note this was already
done for InitChainAsync, why not for others???). Otherwise, when we
execute the block, tx might come in and call the callback in the same
time we're updating it in execBlockOnProxyApp => DATA RACE
Fixes#2721
Consensus state is locked
```
goroutine 113333 [semacquire, 309 minutes]:
sync.runtime_SemacquireMutex(0xc00180009c, 0xc0000c7e00)
/usr/local/go/src/runtime/sema.go:71 +0x3d
sync.(*RWMutex).RLock(0xc001800090)
/usr/local/go/src/sync/rwmutex.go:50 +0x4e
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusState).GetRoundState(0xc001800000, 0x0)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/state.go:218 +0x46
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusReactor).queryMaj23Routine(0xc0017def80, 0x11104a0, 0xc0072488f0, 0xc007248
9c0)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/reactor.go:735 +0x16d
created by github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusReactor).AddPeer
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/reactor.go:172 +0x236
```
because localClient is locked
```
goroutine 1899 [semacquire, 309 minutes]:
sync.runtime_SemacquireMutex(0xc00003363c, 0xc0000cb500)
/usr/local/go/src/runtime/sema.go:71 +0x3d
sync.(*Mutex).Lock(0xc000033638)
/usr/local/go/src/sync/mutex.go:134 +0xff
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/abci/client.(*localClient).SetResponseCallback(0xc0001fb560, 0xc007868540)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/abci/client/local_client.go:32 +0x33
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/proxy.(*appConnConsensus).SetResponseCallback(0xc00002f750, 0xc007868540)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/proxy/app_conn.go:57 +0x40
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/state.execBlockOnProxyApp(0x1104e20, 0xc002ca0ba0, 0x11092a0, 0xc00002f750, 0xc0001fe960, 0xc000bfc660, 0x110cfe0, 0xc000090330, 0xc9d12, 0xc000d9d5a0, ...)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/state/execution.go:230 +0x1fd
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/state.(*BlockExecutor).ApplyBlock(0xc002c2a230, 0x7, 0x0, 0xc000eae880, 0x6, 0xc002e52c60, 0x16, 0x1f927, 0xc9d12, 0xc000d9d5a0, ...)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/state/execution.go:96 +0x142
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusState).finalizeCommit(0xc001800000, 0x1f928)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/state.go:1339 +0xa3e
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusState).tryFinalizeCommit(0xc001800000, 0x1f928)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/state.go:1270 +0x451
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusState).enterCommit.func1(0xc001800000, 0x0, 0x1f928)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/state.go:1218 +0x90
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusState).enterCommit(0xc001800000, 0x1f928, 0x0)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/state.go:1247 +0x6b8
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusState).addVote(0xc001800000, 0xc003d8dea0, 0xc000cf4cc0, 0x28, 0xf1, 0xc003bc7ad0, 0xc003bc7b10)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/state.go:1659 +0xbad
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusState).tryAddVote(0xc001800000, 0xc003d8dea0, 0xc000cf4cc0, 0x28, 0xf1, 0xf1, 0xf1)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/state.go:1517 +0x59
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusState).handleMsg(0xc001800000, 0xd98200, 0xc0070dbed0, 0xc000cf4cc0, 0x28)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/state.go:660 +0x64b
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusState).receiveRoutine(0xc001800000, 0x0)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/state.go:617 +0x670
created by github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus.(*ConsensusState).OnStart
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/consensus/state.go:311 +0x132
```
tx comes in and CheckTx is executed right when we execute the block
```
goroutine 111044 [semacquire, 309 minutes]:
sync.runtime_SemacquireMutex(0xc00003363c, 0x0)
/usr/local/go/src/runtime/sema.go:71 +0x3d
sync.(*Mutex).Lock(0xc000033638)
/usr/local/go/src/sync/mutex.go:134 +0xff
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/abci/client.(*localClient).CheckTxAsync(0xc0001fb0e0, 0xc002d94500, 0x13f, 0x280, 0x0)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/abci/client/local_client.go:85 +0x47
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/proxy.(*appConnMempool).CheckTxAsync(0xc00002f720, 0xc002d94500, 0x13f, 0x280, 0x1)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/proxy/app_conn.go:114 +0x51
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/mempool.(*Mempool).CheckTx(0xc002d3a320, 0xc002d94500, 0x13f, 0x280, 0xc0072355f0, 0x0, 0x0)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/mempool/mempool.go:316 +0x17b
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/rpc/core.BroadcastTxSync(0xc002d94500, 0x13f, 0x280, 0x0, 0x0, 0x0)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/rpc/core/mempool.go:93 +0xb8
reflect.Value.call(0xd85560, 0x10326c0, 0x13, 0xec7b8b, 0x4, 0xc00663f180, 0x1, 0x1, 0xc00663f180, 0xc00663f188, ...)
/usr/local/go/src/reflect/value.go:447 +0x449
reflect.Value.Call(0xd85560, 0x10326c0, 0x13, 0xc00663f180, 0x1, 0x1, 0x0, 0x0, 0xc005cc9344)
/usr/local/go/src/reflect/value.go:308 +0xa4
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/rpc/lib/server.makeHTTPHandler.func2(0x1102060, 0xc00663f100, 0xc0082d7900)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/rpc/lib/server/handlers.go:269 +0x188
net/http.HandlerFunc.ServeHTTP(0xc002c81f20, 0x1102060, 0xc00663f100, 0xc0082d7900)
/usr/local/go/src/net/http/server.go:1964 +0x44
net/http.(*ServeMux).ServeHTTP(0xc002c81b60, 0x1102060, 0xc00663f100, 0xc0082d7900)
/usr/local/go/src/net/http/server.go:2361 +0x127
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/rpc/lib/server.maxBytesHandler.ServeHTTP(0x10f8a40, 0xc002c81b60, 0xf4240, 0x1102060, 0xc00663f100, 0xc0082d7900)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/rpc/lib/server/http_server.go:219 +0xcf
github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/rpc/lib/server.RecoverAndLogHandler.func1(0x1103220, 0xc00121e620, 0xc0082d7900)
/root/go/src/github.com/MinterTeam/minter-go-node/vendor/github.com/tendermint/tendermint/rpc/lib/server/http_server.go:192 +0x394
net/http.HandlerFunc.ServeHTTP(0xc002c06ea0, 0x1103220, 0xc00121e620, 0xc0082d7900)
/usr/local/go/src/net/http/server.go:1964 +0x44
net/http.serverHandler.ServeHTTP(0xc001a1aa90, 0x1103220, 0xc00121e620, 0xc0082d7900)
/usr/local/go/src/net/http/server.go:2741 +0xab
net/http.(*conn).serve(0xc00785a3c0, 0x11041a0, 0xc000f844c0)
/usr/local/go/src/net/http/server.go:1847 +0x646
created by net/http.(*Server).Serve
/usr/local/go/src/net/http/server.go:2851 +0x2f5
```
* consensus: use read lock in Receive#VoteMessage
* use defer to unlock mutex because application might panic
* use defer in every method of the localClient
* add a changelog entry
* drain channels before Unsubscribe(All)
Read 55362ed766/libs/pubsub/pubsub.go (L13)
for the detailed explanation of the issue.
We'll need to fix it someday. Make sure to keep an eye on
https://github.com/tendermint/tendermint/blob/master/docs/architecture/adr-033-pubsub.md
* retry instead of panic when peer has no state in reactors other than consensus
in /dump_consensus_state RPC endpoint, skip a peer with no state
* rpc/core/mempool: simplify error messages
* rpc/core/mempool: use time.After instead of timer
also, do not log DeliverTx result (to be consistent with other memthods)
* unlock before calling the callback in reqRes#SetCallback
* Add test showing normal Sync call
* Show hanging response if server dies during BeginBlock
* Flush all in-flight requests on shutdown, not just pending requests
* Add bugfix to CHANGELOG
