From aed420ace502f4827c8528f5fc530d39ac07d4a0 Mon Sep 17 00:00:00 2001 From: sage Date: Wed, 13 Sep 2006 21:00:01 +0000 Subject: [PATCH] post fast. git-svn-id: https://ceph.svn.sf.net/svnroot/ceph@855 29311d96-e01e-0410-9327-a35deaab8ce9 --- ceph/TODO | 32 ++++++++++++++++++-------------- 1 file changed, 18 insertions(+), 14 deletions(-) diff --git a/ceph/TODO b/ceph/TODO index 5eda56c1ee32a..6a3590b77233c 100644 --- a/ceph/TODO +++ b/ceph/TODO @@ -4,34 +4,38 @@ - how to get usage feedback to monitor? -messenger -- lookup upcall, -- distributed namer (eg send to MSG_ADDR_MON_ANY) -- failure reporting -- generalize out a transport layer? - - eg reliable tcp for most things, connectionless unreliable datagrams for monitors? - - or, aggressive connection closing on monitors? or just max_connections and an lru? -- share same tcp socket for sender and receiver? -- close idle connections? -- osds: forget idle client addrs - osd/rados - clean up writeahead logs - fix heartbeat wrt new replication - figure out new rep failure cases - same_primary_since -> same_tail_since -- mark residual pgs obsolete *** +- mark residual pgs obsolete ??? +- read vs write threads - deal with divergent disconnected primaries - rdlocks - pg_bit changes - use pg->info.same_role_since wrt replication ops. - report crashed pgs? + +messenger +- lookup upcall, +- distributed namer (eg send to MSG_ADDR_MON_ANY) +- failure reporting +- generalize out a transport layer? + - eg reliable tcp for most things, connectionless unreliable datagrams for monitors? + - or, aggressive connection closing on monitors? or just max_connections and an lru? +- share same tcp socket for sender and receiver? +- close idle connections? +- osds: forget idle client addrs + monitor - monitor needs to monitor some osds... - monitor pgs, notify on out - watch osd utilization; adjust overload in cluster map + + objecter - handle new rep mode failure modes... head, tail, middle @@ -40,17 +44,17 @@ objectcacher - ocacher caps transitions vs locks - test read locks - reliability - heartbeat vs ping - how to choose peer sets - osdmonitor, filter ebofs +- nonblocking write on missing onodes? - reallocate if dirty is cancelable. specifically, so that pg log writes are efficient. -- fix buffercache::rx_finish uninitialized pstart bug - fix sync() - fix bug in node rotation on insert (and reenable) +- snapshots - allow btree sets instead of maps - verify LRU behavior sensible: writes go to mid, not top! - fix NEAR_LAST_FWD (?) -- 2.39.5