I called victory way too fast, still can happen, just not as often. The
fact that the slave can flood the master is a bit annoying, it should wait
and continue later. Right now the behavior is if this happen, the
slave/master hang (web gui is still accessible, but impossible to cancel
the jobs)
Finally find the problem. Seem like my mount drive under Linux was under
heavy usage (the polling was occurring at the same time the IT was doing
big backup) and therefor the Master Disk IO could not follow to write the
jobs and the RAM vm.dirty_ was flooded. Making the build to hang.
So I chan
More info, it seem like if I use the Windows slave computer and the poll
occur, it work just fine. So when the user is lock (sleep is disabled, only
the screen is allowed to go to sleep on this computer and maximum
performance profile is used) and a poll on the master request the slave to
do so
Seem like I put the blame on WarningsPublisher too soon, yesterday it hang
on the compilation, running a batch command!
Any way to debug this? Right now this make the whole setup useless, it hang
everyday when start by polling. Starting it manually doesn't reproduce the
effect.
--
You received
I have narrow it down again to the WarningsPublisher MSBuild only when the
build is started by a poll, starting the build manually doesn't seem to
show the hang behavior on the Windows slave (Linux Master)
Anybody have an idea of why or how to fix this?
--
You received this message because you
It hang again today, seem like the MSBuild Warning plugin is the pipeline
step that hang under Windows 10 if the user is in lock (not log off) state.
When user that run the slave client is active on the machine it work fine.
step([$class: 'WarningsPublisher', canRunOnFailed: true, consoleParsers
