test: reduce GitLab Puma memory use - #629
Conversation
|
Let's do a few rounds of CI checks before merging to confirm it's the right fix. |
Puma reconfiguration cause temporary issues, documented in [1]. We do not need reconfiguration during CI. Closes: Roche#628 [1]: https://federal-support.gitlab.com/hc/en-us/articles/40212946732692-Frequent-502-Errors-and-Restarts-Due-to-Puma-Single-Mode-Configuration Signed-off-by: Jean-Philippe Evrard <jean-philippe.evrard+rochepub@external.roche.com>
|
First pass OK. |
|
Second pass failed on 502 (x1) |
|
Third pass OK |
|
Fourth pass fail (x1) |
|
Fifth pass OK |
|
6th pass: Fail (x1) |
|
it seems CI is now b0rked. |
|
7th pass working. |
|
8th pass: Fail (x1) |
|
9th pass: Success. |
|
10th pass: Fail. This makes it a 50% success chance to merge, which is an improvement, but not by a large margin. |
This is an unrelated issue. |
|
Another issue appeared: Seems like an issue with lightningcss.linux-arm64-musl.node, which triggers my question: Do we care about arm right now? |
|
Rebuilding in case it was a flake (it used to work at the beginning of this initiative) |
|
@o1da even though this is not fully improving, it's better than without it. I will work on two other fixes: 1) resilience to 502 errors 2) improving test stack to produce less 502 errors (moving to multinode). This patch is still worth it, imo. |
|
I am still investigating something that might improve further, and had different settings (going multi node mode). But as I am doing this on the side, it might take time. Hence why I am thinking this is worth it already. |
Puma reconfiguration cause temporary issues, documented in 1. We do not need reconfiguration during CI.
Closes: #628