I learned about dynamic debugging in the kernel driver this afternoon which gets me closer to something like an access log on the client side: echo -n 'file fs/ceph/dir.c +p' >/sys/kernel/debug/dynamic_debug/control echo -n 'file fs/ceph/file.c +p' >/sys/kernel/debug/dynamic_debug/control A bit more verbose then an access log though. What I'm trying to do is track down a specific client that is seeing significantly more read IOPS than other clients. From there I want to know exactly what that client is up to. Having something like an access log would be pretty helpful in tracking this behavior down. Bryan > On Apr 21, 2022, at 12:29 PM, Bryan Stillwell <bstillwell@xxxxxxxxxxx> wrote: > > I'm wondering if there's a way to get access_log like logging for CephFS metadata requests? > > I've found setting debug_mds to 4 on the mds is close, but that doesn't include cached requests. It's also kind of weird how it presents the path as the inode of a folder (which isn't even the folder which holds the file, but one that is up a few levels in my testing) and then the filename. > > It probably makes more sense for logging to be on the client side so we can see cache hits/misses and also latency numbers, but that could be weird with a kernel module. > > Any help is appreciated! > > Thanks, > Bryan _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |