strelok89 commented on PR #3855:
URL: https://github.com/apache/celeborn/pull/3855#issuecomment-5748446938
Thanks — all three Copilot findings were valid and are fixed in `df4aa1e6a`,
each reproduced first and each with a regression test.
They turned out to be one bug class: Go template truthiness treating an
explicit zero value as unset. The same class was fixed earlier in this branch
for `minReplicaCount: 0`; these were its siblings.
| finding | before | now |
|---|---|---|
| `zones[].replicas: 0` | rendered the ceil-split count, so a zone could not
be parked | renders `0`; keyed on `hasKey` |
| `zones[].affinity: {}` | kept the chart's `podAntiAffinity`, contradicting
the documented replace | clears it for that zone; keyed on `hasKey` |
| `celeborn.metrics.extraLabels` | interpolated into the double-quoted
`/bin/sh` command, so `$()` in a valid label would be evaluated at startup |
passed via the `CELEBORN_ZONE_METRIC_LABELS` env var and referenced from the
command |
Adding that env var shifted the positional `env[N]` assertions in three
existing tests, so those moved to index-independent `contains` matchers.
`helm unittest --strict` is 121 → 124 passing, `helm lint` clean, and the
default-values render is still byte-identical to `main` — these options remain
opt-in and change nothing for existing releases.
The two `flink2` job failures are unrelated infrastructure flakes: sbt never
launched, failing to fetch `com.swoval:file-tree-views` from Maven Central with
a `SocketException (Connection reset)` on a cold cache. The same matrix cells
passed on their other run (26/28 flink jobs green). A re-run of the two failed
jobs should clear them.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
