> Well, I've been unsuccessful so far on creating a standalone test. > > I have put some scripting in place to capture some additional information on > the server with the out of memory issues. I have a script which just > periodically dumps the output of free -m to a text file. > > So, the output of free -m immediately before and after the out of memory > error looks like this: > > Just before: > total used free shared buff/cache > available > Mem: 14877 2978 132 3553 11766 > 7943 > Swap: 0 0 0 > > Just after: > total used free shared buff/cache > available > Mem: 14877 2946 649 3548 11280 > 7982 > Swap: 0 0 0 > > If I have nearly 8 GB of memory left, why am I getting out of memory errors?
Doesn't the default NUMA setup mean that it can't actually allocate all the available memory to a single NUMA zone (or whatever it's called)?
Or am I talking ancient Linux history with that?
-- Bill Moran
By the way we have similar issues running in our smaller stage environment. 9.5.6 on CentOS 7.2
Only ever my stage environment on smaller AWS t2 boxen. Memory looks fine but Postgres says it has no access to any. And my settings are really low.
Not sure if it's the same issue, but we run into it a few times a year in my stage environment. Also running postgres in unlimited mode for ulimit.