757692 - run_signscript should never return success for a killed worker

Status: RESOLVED FIXED

(Release Engineering :: General, defect)

Description

[[e-mail] (:bhearsum)]

We currently rely on proc.poll()'s return value to decide what we return. However, workers that don't handle SIGINT well or otherwise return 0 when killed can result in unwanted behaviour.

Comment 1

[Erick Dransch [:edransch]]

Attachment: Set error rc when child process is killed

Set a flag when the timer runs out. If the flag is set, we should ignore the child's return code and use -1 instead.

There may be a more elegant way to refactor the block and avoid using the flag, but it's not coming to me right now.

Comment 2

[[e-mail] (:bhearsum)]

Comment on attachment 626296 [details] [diff] [review]
Set error rc when child process is killed

Will try to look at this soon, Catlee should probably look, too.

Comment 3

[[e-mail] (:bhearsum)]

Looks like this just needs to be checked in. I'll see if I can do that later this week.

Comment 4

[[e-mail] (:bhearsum)]

Comment on attachment 626296 [details] [diff] [review]
Set error rc when child process is killed

This patch applied cleanly to lib/python/signing/server.py and passes all tests. It looks sane and safe still, so I went ahead and landed it. Planning to restart the signing servers today.

Comment 5

[[e-mail] (:bhearsum)]

All signing servers have been updated for this change.