Move from StatePause->StateReplicate on heartbeat response when possible #52
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
tbg
force-pushed
the
probe-state-heartbeat
branch
from
4aad73b to
5460636
Compare
pav-kv
reviewed
May 17, 2023
raft.go
Outdated
Contributor
There was a problem hiding this comment.
Is it fine to allow the StateSnapshot here? Maybe test that too?
Contributor
Author
There was a problem hiding this comment.
I made this check for StateProbe explicitly, and added this comment:
// If the follower is fully caught up but also in StateProbe (as can happen
// if ReportUnreachable was called), we also want to send an append (it will
// be empty) to allow the follower to transition back to StateReplicate once
// it responds.
//
// Note that StateSnapshot typically satisfies pr.Match < lastIndex, but
// `pr.Paused()` is always true for StateSnapshot, so sendAppend is a
// no-op.
2c96d75 to
7d1ecfd
Compare
tbg
commented
May 19, 2023
Contributor
Author
tbg
left a comment
tbg
left a comment
There was a problem hiding this comment.
Updated, PTAL. There is a new commit touching up TestLeaderAppResp (in preparation for changing the outcome of one test case in the main commit).
The main commit has been changed to only change behavior for StateProbe (not StateSnapshot) with commentary in both cases on why StateSnapshot is not affected.
raft.go
Outdated
Contributor
Author
There was a problem hiding this comment.
I made this check for StateProbe explicitly, and added this comment:
// If the follower is fully caught up but also in StateProbe (as can happen
// if ReportUnreachable was called), we also want to send an append (it will
// be empty) to allow the follower to transition back to StateReplicate once
// it responds.
//
// Note that StateSnapshot typically satisfies pr.Match < lastIndex, but
// `pr.Paused()` is always true for StateSnapshot, so sendAppend is a
// no-op.
pav-kv
approved these changes
May 24, 2023
ahrtr
reviewed
May 24, 2023
tbg
force-pushed
the
probe-state-heartbeat
branch
from
183b700 to
c4e2300
Compare
This makes `go test -rewrite ./...` stable. Signed-off-by: Tobias Grieger <[email protected]>
It will be touched in this PR. Signed-off-by: Tobias Grieger <[email protected]>
After a call to `ReportUnreachable`, a fully caught up follower would end up in StateReplicate and not leave it despite responding to heartbeats. This is a bug which is going to be fixed in a follow-up commit. Signed-off-by: Tobias Grieger <[email protected]>
See individual commits. Essentially, when a fully caught-up follower was reported unreachable, it'd transition to `StateProbe` but then wouldn't recover from that via heartbeats (once they resumed). This caused some issues in CRDB because we rely on the reported status to reason about the safety of leadership changes, etc. This PR makes it such that StateProbe resolves on its own: when the leader hears back from the follower via a heartbeat, it sends an empty MsgApp, and as response to this moves the follower back into StateProbe. Signed-off-by: Tobias Grieger <[email protected]>
tbg
force-pushed
the
probe-state-heartbeat
branch
from
c4e2300 to
49bac48
Compare
Contributor
Author
|
TFTRs! |
tbg
added a commit
to tbg/cockroach
that referenced
this pull request
May 24, 2023
This picks up etcd-io/raft#52, which addresses a bug discovered in cockroachdb#99191 that could keep a follower in StateProbe even though it was fully caught up. Since we bumped raft just a few days ago, this pretty much only picks up that PR and nothing else: etcd-io/raft@4967cff...eb88ac5 Epic: none Release note: None
craig bot
pushed a commit
to cockroachdb/cockroach
that referenced
this pull request
May 24, 2023
103826: go.mod: bump raft to pick up etcd-io/raft#52 r=pavelkalinnikov a=tbg This picks up etcd-io/raft#52, which addresses a bug discovered in #99191 that could keep a follower in StateProbe even though it was fully caught up. Since we bumped raft just a few days ago, this pretty much only picks up that PR and nothing else: etcd-io/raft@4967cff...eb88ac5 Epic: none Release note: None Co-authored-by: Tobias Grieger <[email protected]>
This was referenced Jul 6, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
See individual commits. Essentially, when a fully caught-up follower was
reported unreachable, it'd transition to
StateProbebut then wouldn'trecover from that via heartbeats (once they resumed).
This caused some issues in CRDB because we rely on the reported status
to reason about the safety of leadership changes, etc.
This PR makes it such that StateProbe resolves on its own: when the
leader hears back from the follower via a heartbeat, it sends an
empty MsgApp, and as response to this moves the follower back into
StateProbe.