Hello List, I'm here to ask about memcg, low-memory conditions, OOM, and getting malloc to be able to return NULL in low-memory situations. The context follows: I've been experimenting with some very-tiny PostgreSQL databases inside of memory control groups (~hundreds of megabytes) and at the lower end of my testing I noticed that out-of-memory conditions in memcg seem to cause the OOM killer to trigger. This was surprising because the non-memcg containerized versions have overcommit turned off, and when not using memcg that results in malloc returning NULL when the system is out of memory, which allows PostgreSQL to abort the transaction and give an error message. This is a desirable behavior. By comparison, OOM killing any child process will cause the database to enter crash recovery and restart. That is not so desirable. After doing some more digging, I learned from some mailing list posts in around 2008-2009 that disabling overcommit doesn't have the analogous effect in a cgroup as it does normally in Linux (in fact, it would appear to do nothing). However, I couldn't find any evidence that this line of discussion had continued in a substantial way after those postings, so here I am. I'd like to know: * Is there some mechanism by which I can make malloc return NULL in low-memory situations * If not, what would someone need to do to make this happen and get this merged? Cheers. -- fdr -- To unsubscribe from this list: send the line "unsubscribe cgroups" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
