Closed o-smirnov closed 4 years ago
OK, timing it, I'm seeing regimes where memory_full_info()
takes up to a second. No idea what causes this, but clearly some unexpected behaviour is afoot. And because this is on the log filter not the handler, this happens all the time (even with disabled log messages).
Will tweak as follows:
make three levels of memory logging: 0/1/2. Level 2 uses full_info. Level 1 will use the fast but less accurate memory_info()
change cadence of full info reports to ~3 sec or so
add USR1/USR2 handler to change memory logging on the fly
Rerunning a test for https://github.com/ratt-ru/CubiCal/issues/388... with a time chunk of 2 and a freq chunk of 7500. Everything started off briskly, but slowed down to molasses towards the end
CPUs are all red in top. Things move along, but at a snail's pace. Gdb shows the damn thing is mostly stuck in...
Yes,
psutils.memory_full_info()
!!! Well the docs do warn it's slower thanmemory_info()
... who knew we were talking about hours and hours slower!