]>
git.apps.os.sepia.ceph.com Git - ceph.git/log
Sage Weil [Thu, 9 Oct 2008 05:06:24 +0000 (22:06 -0700)]
dstart.sh: replace dstartnew.sh
-n will create new fs, otherwise it will start up using existing data.
Sage Weil [Thu, 9 Oct 2008 05:05:35 +0000 (22:05 -0700)]
filestore: fix collection attrs
Need to prepend user.ceph. to attr names...
Sage Weil [Wed, 8 Oct 2008 23:45:31 +0000 (16:45 -0700)]
todos
Sage Weil [Wed, 8 Oct 2008 23:31:50 +0000 (16:31 -0700)]
mon: include pg usage stats, header in pg dump
Sage Weil [Wed, 8 Oct 2008 23:18:10 +0000 (16:18 -0700)]
debian: move map tools and crun to ceph-common
Sage Weil [Wed, 8 Oct 2008 19:10:16 +0000 (12:10 -0700)]
osd: account for used kb separately
Account for used separately from available, as the underlying fs may do tricks
reserving space for root and such.
Sage Weil [Wed, 8 Oct 2008 18:43:08 +0000 (11:43 -0700)]
osd: use kb instead of blocks for pg, osd utilization
Sage Weil [Wed, 8 Oct 2008 17:52:46 +0000 (10:52 -0700)]
osd: prioritize heartbeat messages
Sage Weil [Wed, 8 Oct 2008 17:49:12 +0000 (10:49 -0700)]
msgr: include priority in msg header, make dispatch a priority queue
Generalizes previous hack that put messages from the monitor at the front of the
dispatch queue. Monitor now just sets a (non-default) default_send_priority
of CEPH_MSG_PRIO_HIGH. That value is used only if the message priority isn't
set explicitly by set_priority() before being queued for send.
Sage Weil [Wed, 8 Oct 2008 16:47:37 +0000 (09:47 -0700)]
journal: protect journal access (namely, completions) with mutex
The commit_waiters map was getting corrupted occasionally.
Sage Weil [Wed, 8 Oct 2008 16:31:23 +0000 (09:31 -0700)]
debian: include crun in osd, mds, mon packages
Sage Weil [Wed, 8 Oct 2008 16:28:58 +0000 (09:28 -0700)]
osdmaptool: fix silly num_dom bug
Erwin, Brock A [Wed, 8 Oct 2008 16:28:32 +0000 (09:28 -0700)]
ceph.spec.in: include crun
Yehuda Sadeh [Tue, 7 Oct 2008 20:18:52 +0000 (13:18 -0700)]
kclient: initializing kaddr (merge fix)
Yehuda Sadeh [Tue, 7 Oct 2008 20:04:32 +0000 (13:04 -0700)]
Merge branch 'writepage' into unstable
Conflicts:
src/kernel/messenger.c
Yehuda Sadeh [Tue, 7 Oct 2008 18:25:09 +0000 (11:25 -0700)]
kclient: use both writepage and crc
Yehuda Sadeh [Tue, 7 Oct 2008 00:43:42 +0000 (17:43 -0700)]
kclient: disabling the readpage
Yehuda Sadeh [Tue, 7 Oct 2008 00:37:41 +0000 (17:37 -0700)]
kclient: disabling readpage
Yehuda Sadeh [Tue, 7 Oct 2008 00:32:00 +0000 (17:32 -0700)]
kclient: use writepage instead of sendmsg for some of the data
Yehuda Sadeh [Tue, 7 Oct 2008 19:35:27 +0000 (12:35 -0700)]
kclient: missing include file
Yehuda Sadeh [Tue, 7 Oct 2008 18:27:46 +0000 (11:27 -0700)]
Revert "kclient: use writepage instead of sendmsg for some of the data"
This reverts commit
406fc33a6bb6120925edda0db9fc5f624d43efb9 .
Yehuda Sadeh [Tue, 7 Oct 2008 00:32:00 +0000 (17:32 -0700)]
kclient: use writepage instead of sendmsg for some of the data
Sage Weil [Tue, 7 Oct 2008 19:32:40 +0000 (12:32 -0700)]
msgr: expand footer flags, include nocrc, kclient nocrc mount option
Sage Weil [Mon, 6 Oct 2008 23:36:56 +0000 (16:36 -0700)]
ebofs: -ENOSPC in apply_transaction
Yehuda Sadeh [Tue, 7 Oct 2008 18:44:10 +0000 (11:44 -0700)]
kclient: change default rsize to 128k
Yehuda Sadeh [Tue, 7 Oct 2008 18:25:09 +0000 (11:25 -0700)]
kclient: use both writepage and crc
Yehuda Sadeh [Tue, 7 Oct 2008 00:43:42 +0000 (17:43 -0700)]
kclient: disabling the readpage
Yehuda Sadeh [Tue, 7 Oct 2008 00:37:41 +0000 (17:37 -0700)]
kclient: disabling readpage
Yehuda Sadeh [Tue, 7 Oct 2008 00:32:00 +0000 (17:32 -0700)]
kclient: use writepage instead of sendmsg for some of the data
Yehuda Sadeh [Mon, 6 Oct 2008 23:10:40 +0000 (16:10 -0700)]
kclient: small fix take two
Yehuda Sadeh [Mon, 6 Oct 2008 23:03:10 +0000 (16:03 -0700)]
kclient: small fix for bdi
Yehuda Sadeh [Mon, 6 Oct 2008 23:00:34 +0000 (16:00 -0700)]
kclient: use a private bdi
Sage Weil [Mon, 6 Oct 2008 22:48:03 +0000 (15:48 -0700)]
kclient: fix file to object mapping
We ignore cases where the stripe_unit is small, for now, as that will require
passing a stride size back to the caller so that the correctly spaced
pages are used.
Sage Weil [Mon, 6 Oct 2008 21:38:03 +0000 (14:38 -0700)]
crushtool: verify explicit item pos
Sage Weil [Mon, 6 Oct 2008 21:35:45 +0000 (14:35 -0700)]
cm.txt: fix up metaroot pool
Sage Weil [Mon, 6 Oct 2008 18:07:18 +0000 (11:07 -0700)]
crushtool: verify item pos
Sage Weil [Mon, 6 Oct 2008 18:07:13 +0000 (11:07 -0700)]
dstartnew.sh: use custom crush map with separate metadata osd pool
Sage Weil [Mon, 6 Oct 2008 18:04:20 +0000 (11:04 -0700)]
crush: fix grammar
Yehuda Sadeh [Fri, 3 Oct 2008 23:34:38 +0000 (16:34 -0700)]
crun won't be removed in make clean
Sage Weil [Fri, 3 Oct 2008 23:44:34 +0000 (16:44 -0700)]
kclient: set first,last snap interval in lease release messages
Sage Weil [Fri, 3 Oct 2008 23:09:15 +0000 (16:09 -0700)]
dstartnew.sh: keep core dumps
Sage Weil [Fri, 3 Oct 2008 21:51:51 +0000 (14:51 -0700)]
mon: tell osd we manually mark "down"
Sage Weil [Fri, 3 Oct 2008 23:11:48 +0000 (16:11 -0700)]
kclient: fix mds session leak
Drop ref after handling lease callback, and during shutdown.
Sage Weil [Fri, 3 Oct 2008 22:30:33 +0000 (15:30 -0700)]
client: fix objectcacher flush/commit callback
There is now an ObjectCacher callback for all-acked and all-committed. Client uses the latter.
Sage Weil [Fri, 3 Oct 2008 21:42:59 +0000 (14:42 -0700)]
kclient: fix mds timeout waiting for map
Unwind locking when wait_for_new_map returns (timeout)
error.
Sage Weil [Fri, 3 Oct 2008 21:17:56 +0000 (14:17 -0700)]
update dstartnew to do btrfs
Sage Weil [Fri, 3 Oct 2008 21:17:08 +0000 (14:17 -0700)]
filestore: fix completion callbacks when journaling is off
Need to do completions for all older op_seqs, not just the most recent.
Sage Weil [Thu, 2 Oct 2008 21:17:27 +0000 (14:17 -0700)]
osd: increase osd_heartbeat_grace
Yehuda Sadeh [Fri, 3 Oct 2008 20:57:12 +0000 (13:57 -0700)]
fix for cmon startup
Sage Weil [Fri, 3 Oct 2008 20:59:49 +0000 (13:59 -0700)]
kclient: fix readpages when not all pages are found
Adjust the request len, page count when fewer than the desired
number of pages are found.
Sage Weil [Fri, 3 Oct 2008 20:34:31 +0000 (13:34 -0700)]
mds: create snaprealm for stray dentries resulting from rename
This was done in the unlink path, but overlooked by rename(), resulting in
problems down the line when the stray nlink=0 item was purged.
Yehuda Sadeh [Fri, 3 Oct 2008 20:45:58 +0000 (13:45 -0700)]
cmon will run as a daemon in any case
Yehuda Sadeh [Fri, 3 Oct 2008 20:43:35 +0000 (13:43 -0700)]
combined vstartnew.sh and startnew.sh into vstartnew.sh
Yehuda Sadeh [Fri, 3 Oct 2008 19:41:16 +0000 (12:41 -0700)]
startnew.sh uses crun in order to execute modules
Yehuda Sadeh [Fri, 3 Oct 2008 19:33:26 +0000 (12:33 -0700)]
crun will not run infintely if -d specified on the command line
Yehuda Sadeh [Fri, 3 Oct 2008 19:08:22 +0000 (12:08 -0700)]
Merge branch 'unstable' of ssh://ceph.newdream.net/git/ceph into unstable
Sage Weil [Fri, 3 Oct 2008 19:09:05 +0000 (12:09 -0700)]
msgr: clean up crc handling
Sage Weil [Fri, 3 Oct 2008 19:08:32 +0000 (12:08 -0700)]
mds: don't issue leases on dentries in stray dir
Also, assert in remove_dentry() that the dentry has no more leases.
Sage Weil [Fri, 3 Oct 2008 17:56:45 +0000 (10:56 -0700)]
msgr: don't host msg bufferlists in write_message
payload and data need to be preserved in case the connection drops and the messages
need to be resent.
Yehuda Sadeh [Fri, 3 Oct 2008 19:07:07 +0000 (12:07 -0700)]
also kill crun when stopping the system
Yehuda Sadeh [Fri, 3 Oct 2008 19:06:28 +0000 (12:06 -0700)]
added crun wrapper
Yehuda Sadeh [Fri, 3 Oct 2008 18:54:53 +0000 (11:54 -0700)]
add -f flag for foreground execution of modules (with logs in the log files)
Sage Weil [Fri, 3 Oct 2008 05:23:26 +0000 (22:23 -0700)]
msgr: include crc in msg header. verify crc for received messages.
Sage Weil [Fri, 3 Oct 2008 04:56:58 +0000 (21:56 -0700)]
clean up crc32c
Sage Weil [Fri, 3 Oct 2008 00:01:10 +0000 (17:01 -0700)]
add crc to messages. protocol change.
Sage Weil [Thu, 2 Oct 2008 16:57:48 +0000 (09:57 -0700)]
todos
Yehuda Sadeh [Fri, 3 Oct 2008 00:34:04 +0000 (17:34 -0700)]
kclient: can use names of modules in order to set debug mask
Yehuda Sadeh [Thu, 2 Oct 2008 23:36:20 +0000 (16:36 -0700)]
kclient: set asynchronous error cleanup
Yehuda Sadeh [Thu, 2 Oct 2008 23:00:26 +0000 (16:00 -0700)]
kclient: writepages async error handling and some cleanup
Yehuda Sadeh [Thu, 2 Oct 2008 19:06:48 +0000 (12:06 -0700)]
kclient: set page error bit on readpages error
Sage Weil [Thu, 2 Oct 2008 21:10:24 +0000 (14:10 -0700)]
kclient: fix incremental osdmap decoding bug
Sage Weil [Thu, 2 Oct 2008 20:39:36 +0000 (13:39 -0700)]
kclient: some osdmap cleanups
mostly cosmetic
Sage Weil [Thu, 2 Oct 2008 20:39:02 +0000 (13:39 -0700)]
mon: reweight osds via cmonctl
Allow reweight (offload adjustment) via cmonctl using float value.
Sage Weil [Thu, 2 Oct 2008 20:01:40 +0000 (13:01 -0700)]
kclient: make standby an allowable connection race state
Sage Weil [Thu, 2 Oct 2008 19:58:06 +0000 (12:58 -0700)]
msgr: make standby an allowable connection race state
Sage Weil [Thu, 2 Oct 2008 18:49:42 +0000 (11:49 -0700)]
kclient: no .snap in root dir
Sage Weil [Thu, 2 Oct 2008 18:36:43 +0000 (11:36 -0700)]
osd: fix reversed logic on push
Sage Weil [Thu, 2 Oct 2008 18:15:21 +0000 (11:15 -0700)]
mds: remove stale proc_message()
Sage Weil [Thu, 2 Oct 2008 18:15:08 +0000 (11:15 -0700)]
mon: allow new arguments to be sent to running mon, mds, osd via cmonctl
Sage Weil [Thu, 2 Oct 2008 16:57:48 +0000 (09:57 -0700)]
todos
Sage Weil [Wed, 1 Oct 2008 22:17:27 +0000 (15:17 -0700)]
kclient: mds reset stub
Get rid of compile warning. Add todo.
Sage Weil [Wed, 1 Oct 2008 22:34:16 +0000 (15:34 -0700)]
msgr: fixed bug in accept
banner[] isn't null terminated
Sage Weil [Wed, 1 Oct 2008 22:03:41 +0000 (15:03 -0700)]
messenger protocol changes
There is now a plaintext banner that is exchanged so that both ends can
verify they are talking to the right software. This string should be
changed each time the wire protocol changes.
There are also mon, osd, and mds specific protocol versions defined, but
these are not yet used.
Yehuda Sadeh [Wed, 1 Oct 2008 20:41:05 +0000 (13:41 -0700)]
kclient: whitespaces
Yehuda Sadeh [Wed, 1 Oct 2008 20:38:26 +0000 (13:38 -0700)]
kclient: can supply multiple monitor ips on mount
Yehuda Sadeh [Mon, 29 Sep 2008 20:33:02 +0000 (13:33 -0700)]
kclient: adding unmaskable debug flag
Sage Weil [Wed, 1 Oct 2008 19:00:53 +0000 (12:00 -0700)]
makefile update
Sage Weil [Wed, 1 Oct 2008 18:53:38 +0000 (11:53 -0700)]
v0.4, todos
Sage Weil [Wed, 1 Oct 2008 18:41:00 +0000 (11:41 -0700)]
Merge branch 'master' into unstable
Sage Weil [Wed, 1 Oct 2008 18:40:48 +0000 (11:40 -0700)]
config: comment
Sage Weil [Wed, 1 Oct 2008 18:39:42 +0000 (11:39 -0700)]
mds: fix file recovery
Make state bits distinct. Duh.
Also, if an inode is queued for recovery, and doesn't need it later, unqueue it
(clear the bit, drop auth pin).
Sage Weil [Wed, 1 Oct 2008 17:50:51 +0000 (10:50 -0700)]
kclient: fix pending cap_snap checks, queuing
Add new cap_snaps to the _tail_ of the list.
Also, use a helper to check for pending (write in progress) check, and be
sure to check the list tail in that case.
Sage Weil [Wed, 1 Oct 2008 17:50:46 +0000 (10:50 -0700)]
client: fix queued cap_snap write-in-progress checks
Sage Weil [Tue, 30 Sep 2008 19:59:27 +0000 (12:59 -0700)]
kclient: be more paranoid in prepare_pages
It's theoretically possible to get racing incoming osd_op_replies. Make sure we only give the pages away once.
Sage Weil [Tue, 30 Sep 2008 19:58:17 +0000 (12:58 -0700)]
kclient: fix bad pointer arithmetic in messenger
This was causing intermittent slab corruption when TCP fragmented the stream in unusual ways.
Sage Weil [Mon, 29 Sep 2008 23:30:54 +0000 (16:30 -0700)]
kclient: don't include pages in read requests; some osdc cleanup
Sage Weil [Mon, 29 Sep 2008 22:25:07 +0000 (15:25 -0700)]
kclient: switch to sync writes when NEARFULL osdmap flag set; ENOSPC on FULL flag
Sage Weil [Mon, 29 Sep 2008 20:52:28 +0000 (13:52 -0700)]
todos
Sage Weil [Mon, 29 Sep 2008 20:31:24 +0000 (13:31 -0700)]
kclient: default to safe writeback
Now that async writepages is working, we can wait for data to be fully committed without the higher latency affecting writeback throughput.
Sage Weil [Mon, 29 Sep 2008 20:30:35 +0000 (13:30 -0700)]
mds: choose new mon only if beacons are delayed, and switch to mon that sends up beacons
This streamlines the mds <-> mon beacon interaction a bit.