Tasks #1039
closed
Ceph - Bug #910: Multi-MDS Ceph does not pass fsstress
cfuse: requests max_size from non-auth MDS
Added by Greg Farnum about 13 years ago.
Updated over 7 years ago.
Description
See kai:/home/gregf/logs/fsstress/cfuse_caps
The export of caps seems to go well, and the client drops them, but the next time the MDS encodes inodestats/caps for a request on a child inode it reissues caps to the client, and those seem to take over the ones from the auth. Not sure if the problem is that the client is misdirecting requests or that the MDS is doing it wrong (well, it's obviously doing it wrong, but maybe that's because it's not supposed to be doing it at all).
- Status changed from New to In Progress
- Assignee set to Greg Farnum
Okay, the replica MDS is behaving appropriately -- it's issuing only shared caps. It looks like the problem is that for the client these shared caps then prevent it from appropriately updating the master MDS with its new max_size. Trying to figure out exactly why now.
- Subject changed from non-auth MDS reissues client caps to cfuse: requests max_size from non-auth MDS
So the replica mds0 caps are examined before the auth mds1 caps are, and then the client requests a new max_size from mds0, which updates the requested_max_size et al so it doesn't request changes from mds1. Bah, working out a fix.
Testing out a fix now based on only updating max_size on the auth_cap but it'll take a few runs to verify and I want to see if there are any similar problems that might crop up.
- Category changed from 1 to 11
- Status changed from In Progress to Resolved
Looks good. Pushed to master in commit:c9859c2fc8005566ad6159a3d66e2ba10f82a986.
- Project changed from Ceph to CephFS
- Category deleted (
11)
- Target version deleted (
v0.28)
Bulk updating project=ceph category=ceph-fuse issues to move to fs project so that we can remove the ceph-fuse category from the ceph project
Also available in: Atom
PDF