ogayotHi, we have a package (kitty) that calls mlock(2) multiple times as part of the build-time test suite. On ppc64el builders, the pagesize is 64k and the RLIMIT_MEMLOCK is set to 64k too. So only a single call to mlock() can succeed. Would it be realistic to bump the RLIMIT_MEMLOCK value? Thanks!12:00
ogayot(for now, I'll just disable the test if the limit is too low)12:27
cjwatsonogayot: In the past I've resisted making this sort of tweak specifically to the builder environment.  My reasoning is that if a build-time test requires some property of the system, and we tweak that just for builders, then it means that the tests are a less accurate reflection of what will actually work when the package is installed on a real system.12:50
cjwatsonogayot: If we need different limits, we should consider either (a) changing systemd or whatever to have different defaults, or (b) changing the package to have requirements that are more consistent with the systems we actually ship.12:51
cjwatson(IMO anyway12:52
ogayotcjwatson: yes that makes sense. I like the idea of having systemd set more sensible values by default. I'll keep the test disabled for now. Thanks!12:53
sarnoldcjwatson: how about upgrading the builders to 22.04 or newer? I think the limits were raised19:29
sarnoldmax locked memory           (kbytes, -l) 3300230019:29
sarnoldmax locked memory           (kbytes, -l) 793982419:29
sarnoldI suspect it's now a fraction of installed memory size19:29
cjwatsonsarnold: actually in progress :)19:30
cjwatsonjust kept getting bumped by other things19:30
sarnoldcjwatson: yay! :D19:30
sarnoldI'm familiar with that :(19:30

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!