Previously I wrote about the "Broken Pipe" error sporadically causing our gerrit tests to fail for unknown reasons. To try and avoid those (successfully as it seems), I reduced the nr of executors for our OsmocomBuild1 debian slave to 2; used to be 6 IIRC. My argument was that 'make -j 9' would anyway utlize the resources, except during the python testing phase.
But the build slave queue is getting awfully long sometimes and builds are just sitting and waiting for a long time. But when I 'top' on the build slave, I see the CPU idling between 50% and 90% ... assuming that the CPU activity in docker is also shown on the host OS. So now I increased the nr of executors to 3, seraching the optimum that still avoids the 'Broken Pipe' error. Would of course be much nicer to fix this particular curious failure and just hammer the nr of executors up to the roof again.
~N
On 16 Jan 2017, at 15:05, Neels Hofmeyr nhofmeyr@sysmocom.de wrote:
But the build slave queue is getting awfully long sometimes and builds are just sitting and waiting for a long time. But when I 'top' on the build slave, I see the CPU idling between 50% and 90% ... assuming that the CPU activity in docker is also shown on the host OS. So now I increased the nr of executors to 3, seraching the optimum that still avoids the 'Broken Pipe' error. Would of course be much nicer to fix this particular curious failure and just hammer the nr of executors up to the roof again.
thank you. I assume we run into a user limit but I don't know how to make the kernel report it (or which mib file to read). So another dimension is to use sysctl to tune things...
holger
-
Holger Freyther -
Neels Hofmeyr