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
{{ message }}
Repository navigation
Helix machines should reboot post-workitem when exit code is STATUS_DLL_INIT_FAILED (-1073741502) #10013
As noted in dotnet/runtime#72142, sometimes Windows ARM64 machines get into a state where every process started by the Helix client gets STATUS_DLL_INIT_FAILED (-1073741502)
To make things better for our users we should
Introduce a final "exit code" message by default in the work item
If the machine is a windows machine and the exit code is -1073741502, we should use existing functionality to cause a machine reboot post test- this seems to unblock the problem immediately.
Changes went in but I missed something; the exit code is treated as unsigned int inside the helix client (and magically becomes signed somewhere downstream) so I've changed:
Print code that puts exit code at end of log now converts to signed (to prevent confusion)
Use unsigned value (added comment) when checking for infra retry / reboot behavior
Activity
MattGal commentedon Jul 19, 2022
I'll do this work next week when I'm on FR, assigning to self for dibs.
MattGal commentedon Jul 25, 2022
Merged https://dnceng.visualstudio.com/internal/_git/dotnet-helix-machines/pullrequest/24367, will do some testing when this finishes deploying to staging.
MattGal commentedon Jul 26, 2022
Changes went in but I missed something; the exit code is treated as unsigned int inside the helix client (and magically becomes signed somewhere downstream) so I've changed:
New PR: https://dnceng.visualstudio.com/internal/_git/dotnet-helix-machines/pullrequest/24389?_a=files
MattGal commentedon Jul 26, 2022
Latest changes solve the problem and make the logs look good, moving to rollout.
See this job in staging for an example (the work item hard-codes always returning a bad value, so deadletter is inevitable)
MattGal commentedon Jul 27, 2022
These changes are now live and haven't caused trouble, so closing this issue.