You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We only update this metric when we start emitting resolved events for changefeeds. Say one changefeed gets stuck during the initial scan (we emit no resolved events until the initial scan finishes), we won't update the max_behind_nanos metric and customers won't know the changefeed is stuck.
We could initialize changefeed jobs running with an initial scan with some seed highwater / resolved event.
The text was updated successfully, but these errors were encountered:
jayshrivastava
added
C-bug
Code not up to spec/doc, specs & docs deemed correct. Solution expected to change code/behavior.
A-cdc
Change Data Capture
labels
Feb 13, 2023
jayshrivastava
changed the title
cdc: max_behind_nanos does not update without checkpoints
cdc: max_behind_nanos does not update when jobs are stuck
Feb 13, 2023
jayshrivastava
changed the title
cdc: max_behind_nanos does not update when jobs are stuck
cdc: max_behind_nanos does not update when jobs are stuck during initial scans
Feb 15, 2023
Previously, this node-level metric would measure the maximum time between
the present and the oldest checkpoint seen by a change aggregator. Since
this metric was updated by in-memory checkpoints, it was prone to odd
behavior. For example:
- When a node restarts and a changefeed immediately begins a catchup scan,
there are no checkpoints for this changefeed available to calculate the value
of this metric. It's possible that an "inifinite" catchup scan could trigger
where the metric would never get updated (
had the description "Largest commit-to-emit duration of any running feed",
Informs: cockroachdb#97931Closes: cockroachdb#97043Closes: cockroachdb#99409
<what was there before: Previously, ...>
<why it needed to change: This was inadequate because ...>
<what you did about it: To address this, this patch ...>
See https://github.com/cockroachlabs/support/issues/2053#issuecomment-1427839384.
We only update this metric when we start emitting resolved events for changefeeds. Say one changefeed gets stuck during the initial scan (we emit no resolved events until the initial scan finishes), we won't update the
max_behind_nanos
metric and customers won't know the changefeed is stuck.We could initialize changefeed jobs running with an initial scan with some seed highwater / resolved event.
Also related: #93919
Jira issue: CRDB-24482
Epic CRDB-8669
The text was updated successfully, but these errors were encountered: