Hi all, I've been bitten several times by using memcached on a NUMA node in Linux, specifying less than the available RAM on that node for memcached's maximum memory argument, and then running for a little while until the OOM killer starts to eat my system processes and crash the machine.
I don't have any time to fix this issue, but I do have a proposal. When passed a "max ram" flag, could memcached please fault in all those pages by writing something nonzero to each page? By doing so, it will alert the OS kernel that this memory is needed and cannot be overcommitted by the kernel. This will have the desirable behaviour of memcached either crashing on startup with "out of memory", or working perfectly fine until the end of time. Right now, memcached doesn't crash on startup, it just crashes sometime later. Thoughts? Thanks, Michael. -- --- You received this message because you are subscribed to the Google Groups "memcached" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/groups/opt_out.
