Bug #1178
closed
cfuse doesn't shut down -- Fw cap doesn't flush on one inode
Added by Greg Farnum almost 13 years ago.
Updated over 7 years ago.
Description
Logs in kai:~gregf/logs/fsstress/hung_client2. This is on my mds_rename branch with 4 MDSes in fsstress. I suspect it's an MDS bug but am not sure yet.
- Status changed from New to In Progress
Oh, so it's an inode where the auth got changed, and it only has the cap for the new auth but it doesn't think it is the auth for some reason...
oh, interesting. in that case it won't get an IMPORT event because it wasn't exported.. and the client doesn't find out about the auth change. we may want to force an IMPORT message to get sent anyway, or something like that.
I'm pretty sure the client is getting the right import/export messages, there's just something going wrong in the twiddling somewhere. Specifically what I see is that it goes through the check_caps function on the inode in question, and calls send_caps but doesn't go through the auth_cap branch and add the dirty_caps to the cap message.
- Status changed from In Progress to 7
Will be testing locally, push later. We had to make sure to check the mseq when setting a new auth_cap based on the MDS' encoded auth state, or we could get an old encoding after an auth move and improperly revert back to the old state.
- Target version changed from v0.30 to v0.31
- Translation missing: en.field_position set to 700
- Translation missing: en.field_story_points set to 5
- Translation missing: en.field_position deleted (
703)
- Translation missing: en.field_position set to 703
- Status changed from 7 to Resolved
Pushed to master in commit:787f03cbd7489cb31f5feeec08f84d0b68aa5c0a.
- Project changed from Ceph to CephFS
- Category deleted (
1)
- Target version deleted (
v0.31)
Bulk updating project=ceph category=mds bugs so that I can remove the MDS category from the Ceph project to avoid confusion.
Also available in: Atom
PDF