From 716cd84cd1f12d093b6b3a67b6b460c748fe7dff Mon Sep 17 00:00:00 2001 From: sageweil Date: Wed, 3 Oct 2007 23:54:00 +0000 Subject: [PATCH] rename fixup plans.. git-svn-id: https://ceph.svn.sf.net/svnroot/ceph@1883 29311d96-e01e-0410-9327-a35deaab8ce9 --- branches/sage/mds/TODO | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+) diff --git a/branches/sage/mds/TODO b/branches/sage/mds/TODO index ee87154c2511a..644bbec8e882b 100644 --- a/branches/sage/mds/TODO +++ b/branches/sage/mds/TODO @@ -51,6 +51,32 @@ sage doc mdsmon - per-mds, shared standby queues +fix rename +if (0) { +- who cares if dir items are dirtied out of order? + - need to make sure slave_prepare isn't trimmed before teh commit commits + - put xlist_item in MDRequest, add to LogSegment for duration of delay before change is applied to + the cache + - dirtying has to be attached to teh _original_ logsegment.. mdr->ls +} else { +- witnesses + - rename_apply after _prepare + - srci auth do versionlock on inode to avoid mention in journal prior to commit|rollback + - add in-memory rollback in the _commit (on failure) +} +- srci auth needs to freeze inode before exporting it + - hrm, what about misdirected lock requests.. singleauth won't capture a rename export! +- make sure handle_slave_rename_prep can only happen once + (e.g. if witness journals then fails, and initiator re-requests prepare) + +fix link/unlink +- prepare should be sure that the handle_slave_link or whatever only happens _once_ on the slave + + + +mds segments +- reverse_export is fuggered by the log segment business. need to delay auth/clean step until export is confirmed (i.e. make second pass over the subtree?). retest migrator! pay close attention to cache_expire checks on auth ordering... ,.dkjasdfjdajfdkjl + - upshot: once it's done, it'll be much cleaner! mds bugs - fix server unlink .. needs to use slave_requests to clean up any failures during the resolve stage -- 2.39.5