Fix Socket telemetry outerloop test failures#45170
Conversation
Based purely on code inspection, since I couldn't repro the failures happening in the lab, I believe what's happening is we're not outputting the right events if the connect ends up completing so fast that it's treated as a synchronous completion. The fix is to move the relevant tracing to be done when the work completes, regardless of the completion mode. I was able to simulate at least one set of failures by delaying the calling thread before it reaches a particular point, and this fixes that issue, so even if it's not fixing all known problems (hopefully it is), it's at least fixing some.
|
Tagging subscribers to this area: @dotnet/ncl Issue DetailsBased purely on code inspection, since I couldn't repro the failures happening in the lab, I believe what's happening is we're not outputting the right events if the connect ends up completing so fast that it's treated as a synchronous completion. The fix is to move the relevant tracing to be done when the work completes, regardless of the completion mode. I was able to simulate at least one set of failures by delaying the calling thread before it reaches a particular point, and this fixes that issue, so even if it's not fixing all known problems (hopefully it is), it's at least fixing some. Fixes #43931
|
wfurt
left a comment
There was a problem hiding this comment.
LGTM.
We should know in few day after this get submitted what impact it has.
|
/azp list |
|
/azp run runtime-libraries-coreclr outerloop-windows |
|
Azure Pipelines successfully started running 1 pipeline(s). |
Based purely on code inspection, since I couldn't repro the failures happening in the lab, I believe what's happening is we're not outputting the right events if the connect ends up completing so fast that it's treated as a synchronous completion. The fix is to move the relevant tracing to be done when the work completes, regardless of the completion mode.
I was able to simulate at least one set of failures by delaying the calling thread before it reaches a particular point, and this fixes that issue, so even if it's not fixing all known problems (hopefully it is), it's at least fixing some.
Fixes #43931
cc: @MihaZupan, @wfurt