This is a little coarse, but the idea is that we'll send information
about the channels a peer has upon the peer-up event that we send to
reactors that we can then use to reject peers (if neeeded) from reactors.
This solves the problem where statesync would hang in test networks
(and presumably real) where we would attempt to statesync from seed
nodes, thereby hanging silently forever.
This change has two main effects:
1. Remove most of the Async methods from the abci.Client interface.
Remaining are FlushAsync, CommitTxAsync, and DeliverTxAsync.
2. Rename the synchronous methods to remove the "Sync" suffix.
The rest of the change is updating the implementations, subsets, and mocks of
the interface, along with the call sites that point to them.
* Fix stringly-typed mock stubs.
* Rename helper method.
* p2p: migrate to use new interface for channel errors
* Update internal/p2p/p2ptest/require.go
Co-authored-by: M. J. Fromberger <michael.j.fromberger@gmail.com>
* rename
* feedback
Co-authored-by: M. J. Fromberger <michael.j.fromberger@gmail.com>
This continues the push of plumbing contexts through tendermint. I
attempted to find all goroutines in the production code (non-test) and
made sure that these threads would exit when their contexts were
canceled, and I believe this PR does that.
This is, perhaps, the trival final piece of #7075 that I've been
working on.
There's more work to be done:
- push more of the setup into the pacakges themselves
- move channel-based sending/filtering out of the
- simplify the buffering throuhgout the p2p stack.
This is intended to fix a test failure that occurs in the p2p state provider. The issue presents as the state provider timing out waiting for the consensus params response.
The reason that this can occur is because the statesync reactor has the possibility of attempting to respond to the params request before the state provider is ready to read it. This results in the reactor hitting the `default` case seen here and then never sending on the channel. The stateprovider will then block waiting for a response and never receive one because the reactor opted not to send it.
When statesync is stopped during shutdown, it has the possibility of deadlocking. A dump of goroutines reveals that this is related to the peerUpdates channel not returning anything on its `Done()` channel when `OnStop` is called. As this is occuring, `processPeerUpdate` is attempting to acquire the reactor lock. It appears that this lock can never be acquired. I looked for the places where the lock may remain locked accidentally and cleaned them up in hopes to eradicate the issue. Dumps of the relevant goroutines may be found below. Note that the line numbers below are relative to the code in the `v0.35.0-rc1` tag.
```
goroutine 36 [chan receive]:
github.com/tendermint/tendermint/internal/statesync.(*Reactor).OnStop(0xc00058f200)
github.com/tendermint/tendermint/internal/statesync/reactor.go:243 +0x117
github.com/tendermint/tendermint/libs/service.(*BaseService).Stop(0xc00058f200, 0x0, 0x0)
github.com/tendermint/tendermint/libs/service/service.go:171 +0x323
github.com/tendermint/tendermint/node.(*nodeImpl).OnStop(0xc0001ea240)
github.com/tendermint/tendermint/node/node.go:769 +0x132
github.com/tendermint/tendermint/libs/service.(*BaseService).Stop(0xc0001ea240, 0x0, 0x0)
github.com/tendermint/tendermint/libs/service/service.go:171 +0x323
github.com/tendermint/tendermint/cmd/tendermint/commands.NewRunNodeCmd.func1.1()
github.com/tendermint/tendermint/cmd/tendermint/commands/run_node.go:143 +0x62
github.com/tendermint/tendermint/libs/os.TrapSignal.func1(0xc000629500, 0x7fdb52f96358, 0xc0002b5030, 0xc00000daa0)
github.com/tendermint/tendermint/libs/os/os.go:26 +0x102
created by github.com/tendermint/tendermint/libs/os.TrapSignal
github.com/tendermint/tendermint/libs/os/os.go:22 +0xe6
goroutine 188 [semacquire]:
sync.runtime_SemacquireMutex(0xc00026b1cc, 0x0, 0x1)
runtime/sema.go:71 +0x47
sync.(*Mutex).lockSlow(0xc00026b1c8)
sync/mutex.go:138 +0x105
sync.(*Mutex).Lock(...)
sync/mutex.go:81
sync.(*RWMutex).Lock(0xc00026b1c8)
sync/rwmutex.go:111 +0x90
github.com/tendermint/tendermint/internal/statesync.(*Reactor).processPeerUpdate(0xc00026b080, 0xc000650008, 0x28, 0x124de90, 0x4)
github.com/tendermint/tendermint/internal/statesync/reactor.go:849 +0x1a5
github.com/tendermint/tendermint/internal/statesync.(*Reactor).processPeerUpdates(0xc00026b080)
github.com/tendermint/tendermint/internal/statesync/reactor.go:883 +0xab
created by github.com/tendermint/tendermint/internal/statesync.(*Reactor.OnStart
github.com/tendermint/tendermint/internal/statesync/reactor.go:219 +0xcd)
```
This test reliably gets hung up on network configuration, (which may
be a real issue,) but it's network setup is handcranked and we should
ensure that the test focuses on it's core assertions and doesn't fail for
test architecture reasons.
I've been noticing that there are a number of situations where the
statesync reactor blocks waiting for peers (or similar,) I've moved
things around to improve outcomes in local tests.
This PR make some tweaks to backfill after running e2e tests:
- Separates sync and backfill as two distinct processes that the node calls. The reason is because if sync fails then the node should fail but if backfill fails it is still possible to proceed.
- Removes peers who don't have the block at a height from the local peer list. As the process goes backwards if a node doesn't have a block at a height they're likely pruning blocks and thus they won't have any prior ones either.
- Sleep when we've run out of peers, then try again.
Sam KleinmanandGitHub
M. J. FrombergerandGitHub
William BanfieldandGitHub
JayT106andGitHub
Callum WatersandGitHub
Aleksandr BezobchukandGitHub