From: Sage Weil <sage@newdream.net>
Date: Tue, 27 May 2008 14:32:00 +0000 (-0700)
Subject: todos
X-Git-Tag: v0.3~170^2~45
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=353f41cca01a62cdc9b5c2605d2eb63394295876;p=ceph.git

todos
---

diff --git a/src/TODO b/src/TODO
index e0a9be2cfe7..d30ebe82e1c 100644
--- a/src/TODO
+++ b/src/TODO
@@ -73,10 +73,15 @@ mds mustfix
 - rollback vs accounting is just broken!  can we avoid it.
   - even with _just_ parent dir mtime, the rollback metablob isn't sufficient during replay.. we need special handling regardless.
 - when renaming a dir, how do we know if we need to update the anchor table?
+- anchor records need to include the dentry hash
+  - adjust conditional guard on anchor table updates accordingly (i.e., update table even if in same dir)
 
-- replay of dir fragmentation  (dont want dir frozen, pins, etc.?)
 - fix rm -r vs mds exports
+  - use new dir size values in dirstat
+
+- replay of dir fragmentation  (dont want dir frozen, pins, etc.?)
 - rename slave in-memory rollback on failure
+
 - proper handling of cache expire messages during rejoin phase?
   -> i think cache expires are fine; the rejoin_ack handler just has to behave if rejoining items go missing
 - try_remove_unlinked_dn thing