]>
git.apps.os.sepia.ceph.com Git - ceph.git/log
Sage Weil [Wed, 5 Aug 2009 21:12:00 +0000 (14:12 -0700)]
v0.12
Sage Weil [Wed, 5 Aug 2009 21:11:17 +0000 (14:11 -0700)]
Makefile: add missing header to noinst_HEADERS
Sage Weil [Wed, 5 Aug 2009 21:09:59 +0000 (14:09 -0700)]
Makefile: remove java hadoop _SOURCES dep
Sage Weil [Wed, 5 Aug 2009 18:41:45 +0000 (11:41 -0700)]
osdmap: kill unused/useless get_num_osds()
Sage Weil [Wed, 5 Aug 2009 18:40:05 +0000 (11:40 -0700)]
mds: show journal entry lengths in debug output
Sage Weil [Wed, 5 Aug 2009 18:38:42 +0000 (11:38 -0700)]
kclient: cleanup unused label
Sage Weil [Wed, 5 Aug 2009 18:38:32 +0000 (11:38 -0700)]
kclient: revamp fsync
Be smarter about when we write back caps on fsync, and when we
wait. Also, wait only for those caps to write back, not for all
caps to be clean, avoiding starvation.
Sage Weil [Tue, 4 Aug 2009 23:41:36 +0000 (16:41 -0700)]
mds: include mds # in table names
Silly us, we were using the same object names for multiple mds's.
That doesn't work so well.
Renamed like so:
mdstable_snaptable -> mds_snaptable
mdstable_inotable -> mds0_inotable
sessionmap -> mds0_sessionmap
mdstable_anchortable -> mds_anchortable
Sage Weil [Tue, 4 Aug 2009 23:08:32 +0000 (16:08 -0700)]
mds: wait for rejoin_gather_finish to complete before finishing
To do that we add ourselves to the rejoin_ack_gather. Otherwise
we end up in up:active before we've even finished our
parallel_fetch or finished up our caps!
Sage Weil [Tue, 4 Aug 2009 23:07:24 +0000 (16:07 -0700)]
mds: set primary lock state to LOCK from strong replica when appropriate
This is needed only because we identify_files_to_recover() before
sending the rejoin acks, and that may twiddle the lock state, so
we need to be in a compatible state.
Sage Weil [Tue, 4 Aug 2009 21:16:03 +0000 (14:16 -0700)]
Makefile: reorg
Sage Weil [Tue, 4 Aug 2009 21:10:50 +0000 (14:10 -0700)]
mds: fix lock state assert on sync->excl
Sage Weil [Tue, 4 Aug 2009 21:10:33 +0000 (14:10 -0700)]
mds: set up root inode properly on replay
Sage Weil [Tue, 4 Aug 2009 20:19:41 +0000 (13:19 -0700)]
mds: fix reversed conditional on simplelock ->excl transition
Sage Weil [Tue, 4 Aug 2009 20:09:32 +0000 (13:09 -0700)]
Makefile: no fakefuse
Sage Weil [Tue, 4 Aug 2009 19:53:22 +0000 (12:53 -0700)]
psim: calc stddev
Sage Weil [Tue, 4 Aug 2009 19:52:43 +0000 (12:52 -0700)]
osdmap: only use domains if ndom > 1
Sage Weil [Tue, 4 Aug 2009 17:22:39 +0000 (10:22 -0700)]
crush: fix grammar so we can recompile what we decompiled
Sage Weil [Tue, 4 Aug 2009 04:34:15 +0000 (21:34 -0700)]
mds: start resolve with root as UNKNOWN (if it's not ours)
Anything that's not ours should be unknown, including the root dir frag,
which normally starts out as mds0.
If we leave it as 0, then when mds0 claims a subset of /, its bounds are
left as 0 as well instead of being set to unknown. Which leads to
incorrect resolve stage results.
Sage Weil [Tue, 4 Aug 2009 03:00:08 +0000 (20:00 -0700)]
mds: fix typo (C_Locker_Eval crash)
Sage Weil [Tue, 4 Aug 2009 00:29:39 +0000 (17:29 -0700)]
kclient: fix locking snafu in ESTALE handling
Sage Weil [Tue, 4 Aug 2009 00:28:03 +0000 (17:28 -0700)]
kclient: choose mds based on dentry parent, if needed
Sage Weil [Tue, 4 Aug 2009 00:19:49 +0000 (17:19 -0700)]
mds: adjust mds state asserts to include clientreplay
Sage Weil [Mon, 3 Aug 2009 23:46:20 +0000 (16:46 -0700)]
mds: twiddle missing locks on export
Sage Weil [Mon, 3 Aug 2009 21:25:35 +0000 (14:25 -0700)]
mds: open_root only after up:active
Otherwise our lock states get totally out of whack.
Sage Weil [Mon, 3 Aug 2009 21:25:16 +0000 (14:25 -0700)]
mds: issue caps on non-auth inodes
Not sure how that && is_auth() snuck in there.
Sage Weil [Mon, 3 Aug 2009 18:39:27 +0000 (11:39 -0700)]
kclient: use caps, fragtree only to choose mds (not hierarchy)
Since we require caps for all inodes in our cache, no need to consider
parents when identifying where to sent a request. Just look at fragtree
(for fragmented dirs) or caps.
Sage Weil [Mon, 3 Aug 2009 19:10:29 +0000 (12:10 -0700)]
kclient: remove stray ; in dout macro
Greg Farnum [Tue, 4 Aug 2009 21:34:17 +0000 (14:34 -0700)]
Hadoop: Replaced some unclean file references.
Hey cool, it can do a map-reduce now!
Greg Farnum [Tue, 4 Aug 2009 21:11:59 +0000 (14:11 -0700)]
uClient: chmod now protects unchangeable bits.
Greg Farnum [Tue, 4 Aug 2009 19:43:34 +0000 (12:43 -0700)]
uClient: _rename doesn't remove nodes from cache.
Sage says this is unnecessary.
Greg Farnum [Tue, 4 Aug 2009 07:54:21 +0000 (00:54 -0700)]
Hadoop: Added a setPermissions method.
Greg Farnum [Tue, 4 Aug 2009 07:18:49 +0000 (00:18 -0700)]
Hadoop: CephFSInterface cleans up memory better on errors.
Greg Farnum [Tue, 4 Aug 2009 01:38:05 +0000 (18:38 -0700)]
Hadoop: Is much more careful with filepath sanitizing in BOTH directions.
Also has a minor interface change allowing it to build in Hadoop .20
Greg Farnum [Tue, 4 Aug 2009 00:36:35 +0000 (17:36 -0700)]
uClient: Unmounting now waits on mds_requests.
Greg Farnum [Fri, 31 Jul 2009 21:17:29 +0000 (14:17 -0700)]
Hadoop: All configuration is now done through Hadoop's config mechanism.
This includes debug on CephFileSystem/n, passing it to ceph's Client,
and the monitor address.
Greg Farnum [Fri, 31 Jul 2009 20:03:31 +0000 (13:03 -0700)]
Hadoop: Now does a proper stat.
Sage Weil [Mon, 3 Aug 2009 16:37:50 +0000 (09:37 -0700)]
kclient: less noisy about mds session timeouts
In particular, we don't want to print this multiple times. Use
state to note when session appears to be hung, and pr_info on
transitions.
Sage Weil [Fri, 31 Jul 2009 22:45:18 +0000 (15:45 -0700)]
msgr: avoid closing fd twice
Set fd to -1 after so we don't close it again later. Otherwise
we get EBADF in random other places and are unhappy.
Sage Weil [Fri, 31 Jul 2009 23:08:25 +0000 (16:08 -0700)]
todo
Sage Weil [Thu, 30 Jul 2009 22:39:41 +0000 (15:39 -0700)]
kclient: debug peer_reset thing
I saw ceph_peer_reset with a bad entity_name.. wtf.
Sage Weil [Thu, 30 Jul 2009 22:08:44 +0000 (15:08 -0700)]
kclient: maintain per-cap-bit tid map to fix pipelined cap updates
We want to allow pipelined cap updates, like
client->mds writeback Fw 1
dirty FwAx
client->mds writeback FwAx 2
dirty Ax
client->mds writeback Ax 3
mds->client ack 1
mds->client ack 2
mds->client ack 3
We need to make sure that the Fw bit is only cleaned after ack 2,
and Ax after ack 3. A single tid for the inode isn't sufficient,
since that would e.g. ignore ack 2... we need a tid per cap bit so
we can pipeline writeback of different caps.
Note that we can't simply write back dirty | flushing caps every
time, since the write may also be releasing the cap. And it would
gum up the MDS locking.
Move the last_tid to the inode, and only pay attention to 16 bits
per cap bit.. that's 17*2 bytes, vs the old 16. Could be worse.
An 8 bit tid is probably also sufficient (that's 256 pipelined
writes) if we're concerned about inode size down the road.
Sage Weil [Wed, 29 Jul 2009 22:52:41 +0000 (15:52 -0700)]
kclient: maintain list of flushing snaps; reflush after reconnect
Sage Weil [Wed, 29 Jul 2009 22:51:54 +0000 (15:51 -0700)]
kclient: fix queue_cap_snap refs, calls in handle_snap
Call on correct ci. Skip past dropping inodes without dropping
our spinlock. Hold ref on prior inode until we traverse to the
next one. (We can't iput while holding our spinlock.)
Greg Farnum [Thu, 30 Jul 2009 21:39:02 +0000 (14:39 -0700)]
Hadoop: Moved ceph debug level set into Java code; disabled CephIOStream debugs.
Greg Farnum [Thu, 30 Jul 2009 21:37:21 +0000 (14:37 -0700)]
uClient: Whoops! Made Inode get/put dout at a lower importance.
Greg Farnum [Thu, 30 Jul 2009 20:20:00 +0000 (13:20 -0700)]
Hadoop: Much cleanup of methods.
Many no-longer-required methods are now private and renamed; overrides and deprecated methods are clearly marked and mostly justified.
Greg Farnum [Thu, 30 Jul 2009 19:36:09 +0000 (12:36 -0700)]
Hadoop: This method no longer exists in the FileSystem API.
Greg Farnum [Thu, 30 Jul 2009 19:35:27 +0000 (12:35 -0700)]
Hadoop: No good reason to overwrite these copy/move functions that I can see
Greg Farnum [Thu, 30 Jul 2009 19:33:25 +0000 (12:33 -0700)]
Hadoop: Since Client is behaving, put back in the close.
Greg Farnum [Thu, 30 Jul 2009 19:18:15 +0000 (12:18 -0700)]
uclient: Fixes refcounting bug when you chdir to cwd.
Greg Farnum [Thu, 30 Jul 2009 03:11:15 +0000 (20:11 -0700)]
Hadoop: More work with deprecated and inappropriately-overridden functions.
Greg Farnum [Thu, 30 Jul 2009 02:29:36 +0000 (19:29 -0700)]
Hadoop: Removed deprecated functions
Greg Farnum [Thu, 30 Jul 2009 02:27:02 +0000 (19:27 -0700)]
Hadoop: moveFromLocalFile uses copyFromLocalFile; is more robust.
Greg Farnum [Thu, 30 Jul 2009 00:01:54 +0000 (17:01 -0700)]
Hadoop: More debugging output changes.
Greg Farnum [Wed, 29 Jul 2009 21:47:51 +0000 (14:47 -0700)]
Hadoop: getFileStatus behaves better with folders.
Greg Farnum [Wed, 29 Jul 2009 21:23:21 +0000 (14:23 -0700)]
uclient: removed cout debug statement
Sage Weil [Thu, 30 Jul 2009 20:42:45 +0000 (13:42 -0700)]
kclient: don't croak on mds request will NULL r_request
Sage Weil [Thu, 30 Jul 2009 20:37:52 +0000 (13:37 -0700)]
mds: process session msgs during clientreplay
Sage Weil [Thu, 30 Jul 2009 20:00:49 +0000 (13:00 -0700)]
mds: no need to take mds_lock in truncate callback
It's already held when the Objecter message handler is called.
Sage Weil [Thu, 30 Jul 2009 19:58:04 +0000 (12:58 -0700)]
mds: defer caps msgs until clientreplay or active
Sage Weil [Wed, 29 Jul 2009 20:50:54 +0000 (13:50 -0700)]
osd: adjust debug output
Sage Weil [Wed, 29 Jul 2009 17:39:58 +0000 (10:39 -0700)]
mds: tolerate a missing directory object
Including error in log, mark dir complete. Some metadata may
be missing!
Sage Weil [Wed, 29 Jul 2009 23:11:04 +0000 (16:11 -0700)]
kclient: release unused caps, but keep some reserve
Release unused caps. Keep some reserve so we can do a readdir
(which reserves lots of caps) quickly, without reallocating a
zillion new caps.
Sage Weil [Wed, 29 Jul 2009 20:28:23 +0000 (13:28 -0700)]
kclient: nofail mode for osd writes
If nofail is specified, allocate request from mempool, and do not
return error on message send failure. Instead, mark the request,
and periodically retry.
This isn't perfect: we can still starve indefinitely trying to
send the write, but it'll do until we have a better way to reserve
resources for writeback messages.
Sage Weil [Tue, 28 Jul 2009 20:01:49 +0000 (13:01 -0700)]
kclient: use mempool for osd req in writeback paths
Only in writeback path. For reads and sync writes, avoid the
mempool.
Greg Farnum [Wed, 29 Jul 2009 01:26:59 +0000 (18:26 -0700)]
Hadoop: Removed unnecessary reload of hadoopcephfs.
Greg Farnum [Wed, 29 Jul 2009 01:24:16 +0000 (18:24 -0700)]
Hadoop: More debugging statements, converted to use a method we can flag off.
Greg Farnum [Wed, 29 Jul 2009 01:22:00 +0000 (18:22 -0700)]
libceph: add a check in ceph_deinitialize to prevent badness on multiple calls.
Greg Farnum [Tue, 28 Jul 2009 18:45:15 +0000 (11:45 -0700)]
uclient: getcwd fixes
Sage Weil [Wed, 29 Jul 2009 00:02:22 +0000 (17:02 -0700)]
kclient: bug on back kmap
Sage Weil [Tue, 28 Jul 2009 21:44:48 +0000 (14:44 -0700)]
osd: fail gracefully when problems reading /proc/loadavg
Sage Weil [Tue, 28 Jul 2009 18:31:43 +0000 (11:31 -0700)]
auth: compat with old struct
Sage Weil [Tue, 28 Jul 2009 18:18:17 +0000 (11:18 -0700)]
initscript: skip if auto start = no/false/0 and nothing or type specified
Sage Weil [Tue, 28 Jul 2009 18:12:48 +0000 (11:12 -0700)]
kclient: avoid overflow in mds reconnect msg size calc
Sage Weil [Mon, 27 Jul 2009 23:14:09 +0000 (16:14 -0700)]
mon: allow osd down->out interval to be 0 (i.e., manual)
Sage Weil [Mon, 27 Jul 2009 23:13:34 +0000 (16:13 -0700)]
osd: remove bad osdmap access
osd may not exist
Sage Weil [Tue, 28 Jul 2009 18:05:14 +0000 (11:05 -0700)]
auth: add ClientTicket
Sage Weil [Tue, 28 Jul 2009 17:37:59 +0000 (10:37 -0700)]
uclient: use lookup parent in getcwd()
Sage Weil [Tue, 28 Jul 2009 17:37:51 +0000 (10:37 -0700)]
mds: add lookup parent op
Greg Farnum [Tue, 28 Jul 2009 16:27:53 +0000 (09:27 -0700)]
Hadoop: Many silly debug statements.
Greg Farnum [Tue, 28 Jul 2009 00:20:21 +0000 (17:20 -0700)]
Hadoop: Minor fixes to the still-unworking Hadoop code.
Greg Farnum [Tue, 28 Jul 2009 00:19:39 +0000 (17:19 -0700)]
Hadoop: hadoopcephfs is now self-contained instead of requiring other libs.
Greg Farnum [Tue, 28 Jul 2009 00:17:47 +0000 (17:17 -0700)]
uclient/libceph: add a getcwd function
Sage Weil [Mon, 27 Jul 2009 22:05:56 +0000 (15:05 -0700)]
kclient: replace osdc request_tree with rbtree
This avoids a possible memory allocation in writeout path.
Sage Weil [Mon, 27 Jul 2009 21:12:33 +0000 (14:12 -0700)]
auth: remove ceph_client_ticket struct
Ticket is generally opaque to client anyway.
Sage Weil [Mon, 27 Jul 2009 18:09:39 +0000 (11:09 -0700)]
osd: generate primary backlog independent of master log
We may have the master log (i.e. know the correct pg state), but
still need a backlog to recover some peers.
Sage Weil [Mon, 27 Jul 2009 18:57:07 +0000 (11:57 -0700)]
kclient: set mapping error on writepage error
Sage Weil [Mon, 27 Jul 2009 17:11:36 +0000 (10:11 -0700)]
kclient: debugfs mounts on /sys/kernel/debug
Sage Weil [Thu, 23 Jul 2009 23:17:41 +0000 (16:17 -0700)]
Makefile: skip hadoop client if jni.h not present
Sage Weil [Thu, 23 Jul 2009 19:50:34 +0000 (12:50 -0700)]
kclient: use a mempool for page vector in writepages
Sage Weil [Thu, 23 Jul 2009 19:30:59 +0000 (12:30 -0700)]
kclient: put writepages pagevec on the stack
Sage Weil [Thu, 23 Jul 2009 18:45:16 +0000 (11:45 -0700)]
debian: upload packages as sage
Sage Weil [Thu, 23 Jul 2009 17:39:32 +0000 (10:39 -0700)]
kclient: fix bad kmalloc
Sage Weil [Thu, 23 Jul 2009 17:39:17 +0000 (10:39 -0700)]
kclient: rewrite layout ioctls
Sage Weil [Thu, 23 Jul 2009 18:49:44 +0000 (11:49 -0700)]
mds: only set striping component of file layout, for now.
Sage Weil [Thu, 23 Jul 2009 17:39:06 +0000 (10:39 -0700)]
fix compat
Sage Weil [Wed, 22 Jul 2009 23:41:03 +0000 (16:41 -0700)]
kclient: define compat_ioctl too
Sage Weil [Wed, 22 Jul 2009 23:22:45 +0000 (16:22 -0700)]
osd: actually adjust rd/wr accounting values
Sage Weil [Wed, 22 Jul 2009 23:15:38 +0000 (16:15 -0700)]
osd: add rd/wr ops/bytes stats to pgs, pools