Feature #3541
closed
mds: robust ino lookup using file backpointers
Added by Sage Weil over 11 years ago.
Updated over 10 years ago.
Description
Following #3540 and #3863, implement the lookup-by-ino algorithm described in http://www.spinics.net/lists/ceph-devel/msg11872.html. This ticket covers updating the lookup request to include the base pool and implementing the MDS logic to try and recursively identify who is authoritative for the inode. Once completed we should be able to look up any inode at any time. (ie, this also covers the rename race fix, in which we query all active MDSes for an inode we should have and can't find, then do another backtrace lookup if we fail. It should be a minimal amount of new code since we already have all-MDS cache queries.)
- Translation missing: en.field_position deleted (
1)
- Translation missing: en.field_position set to 16
- Translation missing: en.field_story_points set to 13
- Translation missing: en.field_position deleted (
17)
- Translation missing: en.field_position set to 17
Is this bug supposed to encompass the anchor table replacement work as well? I wouldn't expect so, but the presence of #3542 without a separate set of "replace anchor table" bugs implies yes...
Also, do we have an algorithm or design for this?
Greg Farnum wrote:
Is this bug supposed to encompass the anchor table replacement work as well? I wouldn't expect so, but the presence of #3542 without a separate set of "replace anchor table" bugs implies yes...
Just the ino lookup using backpointer piece, nothing to do with anchor table yet.
Also, do we have an algorithm or design for this?
Not yet.
- Description updated (diff)
- Translation missing: en.field_story_points deleted (
13)
- Translation missing: en.field_position deleted (
18)
- Translation missing: en.field_position set to 18
- Translation missing: en.field_story_points set to 13
- Translation missing: en.field_position deleted (
19)
- Translation missing: en.field_position set to 8
- Status changed from New to In Progress
- Assignee set to Zheng Yan
A bunch of this got done, but Sage isn't sure if the client -> LOOKUPINO messages are wired up to that infrastructure or not. Yan? :)
(Or somebody else should look at this carefully when we've got time, and we'll want tests regardless.)
- Assignee deleted (
Zheng Yan)
- Status changed from In Progress to Resolved
Also available in: Atom
PDF