zeroshade commented on PR #1301:
URL: https://github.com/apache/arrow-go/pull/1301#issuecomment-5590016746
CI results, two runs of this branch (both green on every macOS job), against
the
post-#1275 baseline on `main`:
| macOS ARM64 job | `main` avg | `main` max | this PR #1 | this PR #2 | this
PR mean | cap |
| --- | --- | --- | --- | --- | --- | --- |
| Go 1.25 | 16.4m | 20.1m | 17m05s | 7m51s | **12.5m** | 20 |
| Go 1.26 | 16.0m | 20.4m | 10m35s | 10m47s | **10.7m** | 20 |
| Go 1.27 | 16.5m | 19.3m | 8m44s | 11m08s | **9.9m** | 20 |
| Go 1.25 - CGO | 19.7m | 24.1m | 21m43s | 18m30s | **20.1m** | 25 |
| Go 1.26 - CGO | 18.5m | 24.8m | 15m07s | 15m14s | **15.2m** | 25 |
| Go 1.27 - CGO | 17.1m | 24.2m | 18m54s | 16m02s | **17.5m** | 25 |
The non-CGO jobs — the ones that were actually being killed at the cap —
average
**11.0m over six samples, down from 16.3m**, which is back to the pre-#1275
baseline of ~11.5m. Worst observed is 17m05s against a 20-minute cap, versus
20.4m before.
Two caveats worth stating rather than burying:
- The Go 1.25 non-CGO job read 17m05s on the first run and 7m51s on the
second.
A 9-minute spread on identical code is runner variance, not signal; it is
why
I took a second sample before claiming anything. It also means the "worst
observed" figure above should be read as noisy.
- The CGO jobs improve only modestly (18.4m → 17.6m mean). They were never
the
ones timing out, and they spend a large fraction of their wall time on
Homebrew setup that this change does not touch.
`parquet/file` still gets its own invocation on Darwin, so the memory fix
from
#1275 is intact; nothing else about the OOM conditions changed.
--
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]
