fix: stop the dropped-channel sentinel from poisoning vchannel-health (v1.2.x) - #528
Merged
Conversation
DataCoord marks a dropped channel by writing funcutil.DroppedChannelCheckpointTimestamp (math.MaxUint64) into its channel checkpoint, which decodes to the year 4199. `show vchannel-health` took the newest checkpoint anywhere in the cluster as the reference for how far behind a vchannel is, so one dropped channel dragged the reference to the year 4199 and every vchannel reported a lag of ~1.14e9 minutes. On a cluster with any drop in flight the command flagged every healthy vchannel FROZEN, which is the opposite of what it is for. Exclude the sentinel, zero, and any future-dated value from the reference, render a vchannel that carries the sentinel as `dropped` under its own DROPPED_CHECKPOINT flag, and keep it out of the FROZEN verdict. Two smaller changes come with it: Label the plain-text output as UTC. The table header already said so; the one-line-command output, which is how this is usually run against a customer cluster, did not, leaving the reader no way to know the timezone. Report the checkpoint's message ID as a seek position. DataCoord hands this position to the flusher through GetChannelRecoveryInfo, and the flusher starts its WAL scanner from the smallest one across the vchannels it serves, so a position that has aged out of the message queue's retention keeps the flusher from starting. Whoever is looking at a frozen vchannel needs that value to compare against the topic's earliest available offset. An 8-byte ID is a Kafka offset and is printed as the integer it is; anything else is printed as hex. Signed-off-by: zhenshan.cao <zhenshan.cao@zilliz.com>
Collaborator
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: czs007 The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
|
Tick the box to add this pull request to the merge queue (same as
|
czs007
added a commit
that referenced
this pull request
Aug 26, 2026
v1.2.0 shipped `show vchannel-health` with a defect that made it flag every healthy vchannel as FROZEN on any cluster with a dropped channel (#528). Bump the version so a build from this branch reports `Birdwatcher Version 1.2.1`. Signed-off-by: zhenshan.cao <zhenshan.cao@zilliz.com>
czs007
added a commit
that referenced
this pull request
Aug 26, 2026
Version bump so a build from `v1.2.x` reports `Birdwatcher Version 1.2.1`, ahead of tagging v1.2.1. v1.2.0 shipped `show vchannel-health` with the defect fixed in #528: a dropped channel's reserved checkpoint value became the reference the command measured every other vchannel against, so every healthy vchannel was reported FROZEN. One line, same shape as #526.
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
Cherry-pick of #527 onto
v1.2.x, so the fix can ship as v1.2.1.show vchannel-healthin the released v1.2.0 flags every healthy vchannel as FROZEN on any cluster that has a dropped channel: DataCoord writesfuncutil.DroppedChannelCheckpointTimestamp(math.MaxUint64, decoding to the year 4199) into a dropped channel's checkpoint, and that value became the reference the command measured every other vchannel against.Seen on a live 2.6 cluster: 196 reported FROZEN, only 19 had actually stopped.
Same commit as #527. Verified against a local 2.6.17 cluster: v1.2.0 reports
lag=1143016227minand flags healthy vchannels; this build reportslag=0minand none. gofmt, build and golangci-lint clean.