cmd/devp2p: fix flakey tests in CI#22757
Conversation
renaynay
force-pushed
the
ethtest-fix-flakeyOldAnnounce
branch
from
598fa5f to
ef36620
Compare
renaynay
changed the title
Done, but now it needs a rebase |
…test chain only after block is received
renaynay
force-pushed
the
ethtest-fix-flakeyOldAnnounce
branch
from
ef36620 to
79e82fa
Compare
holiman
left a comment
holiman
left a comment
There was a problem hiding this comment.
Generally I'm ok with it as it is, if it solves the problems, but would prefer if we could get rid of the random timeout
| t.Fatal() | ||
| } | ||
| }) | ||
| time.Sleep(100 * time.Millisecond) |
There was a problem hiding this comment.
This is a bit indicative of it still being flaky. I'm guessing you added it to make the spurious packets from earlier tests to interfere with later tests, but is it still needed?
If it's only once in a while, maybe we can either figure out what spurious packets those are, and just 'continue' on them, or maybe somehow drain the deliver buffer between test executions?
Just having a random 100ms timout like that means it'll work "more often", but still spuriously fail on e.g. appveyor.
Not a blocker, but I would prefer if we solved it without having that in
|
LGTM! |
|
Unfortunately, still flaky on Appveyor: |
| // update test suite chain | ||
| s.chain.blocks = append(s.chain.blocks, s.fullChain.blocks[nextBlock]) | ||
| // wait for client to update its chain | ||
| if err := receiveConn.waitForBlock66(s.chain.Head()); err != nil { |
There was a problem hiding this comment.
Shouldn't this be ?
| if err := receiveConn.waitForBlock66(s.chain.Head()); err != nil { | |
| if err := receiveConn.waitForBlock66(s.fullchain.blocks[nextBlock]); err != nil { |
I really don't know, just a thought..?
|
i think you also need to apply this: https://github.com/renaynay/go-ethereum/blob/b5eb4d1195849a8ca75fbdeb5976fe7234d0e142/cmd/devp2p/internal/ethtest/types.go#L345 here: Or, well, maybe it doesn't matter, since we have request ids. And it won't make it less flaky, so feel free to ignore it |
|
Ah, so the error is in the status exchange, where the 'other side' has a lower block than we expected. Isn't it because the previous broacast test already progressed 'us' one block? |
…o-ethereum into ethtest-fix-flakeyOldAnnounce
This PR fixes a couple of issues in the eth test suite that caused flakiness when run in the CI.
2 participants
This PR fixes a couple of issues in the eth test suite that caused flakiness when run in the CI.
It depends on #22754 getting merged.