]> git.apps.os.sepia.ceph.com Git - ceph.git/log
ceph.git
15 years agolibceph: Add check before umount in deinitialize
Paul Chiang [Thu, 10 Jun 2010 02:18:53 +0000 (10:18 +0800)]
libceph: Add check before umount in deinitialize

Signed-off-by: Paul Chiang <paul_chiang@tcloudcomputing.com>
Signed-off-by: Sage Weil <sage@newdream.net>
15 years agohostname override option for ceph init script
CC Lien [Tue, 22 Jun 2010 09:28:32 +0000 (17:28 +0800)]
hostname override option for ceph init script

15 years agobuffer: fix logic in write_fd()
Henry C Chang [Thu, 1 Jul 2010 08:03:43 +0000 (16:03 +0800)]
buffer: fix logic in write_fd()

Reset iovec after finishing writev operation.
This bug occurs when buffer list size is larger than IOV_MAX.

15 years agomsgr: fix dispatch throttler release
Sage Weil [Thu, 1 Jul 2010 19:45:16 +0000 (12:45 -0700)]
msgr: fix dispatch throttler release

We need to release the same amount back to the throttler as we originally
reserved.  Store that amount in the Message, and catch all the error
paths.  This fixes the case where messages get fed back into dispatch
locally (i.e. not via read_message()).

15 years agocfuse: daemonize by default
Sage Weil [Thu, 1 Jul 2010 18:15:39 +0000 (11:15 -0700)]
cfuse: daemonize by default

We have to fork, then start client, to avoid killing our pthreads.  Use
a socketpair to tell the parent process about mount success/failure.

15 years agoconfig: -f just means foreground; do not change log output
Sage Weil [Thu, 1 Jul 2010 18:01:18 +0000 (11:01 -0700)]
config: -f just means foreground; do not change log output

Before it would set log_to_stdout = false, but the default might be true.

15 years agoclient: start/stop objectcacher thread in init/shutdown, not constructor
Sage Weil [Thu, 1 Jul 2010 17:57:21 +0000 (10:57 -0700)]
client: start/stop objectcacher thread in init/shutdown, not constructor

15 years agoconfig: disallow -d|--daemonize
Sage Weil [Thu, 1 Jul 2010 16:33:02 +0000 (09:33 -0700)]
config: disallow -d|--daemonize

Either we can daemonize, and do that by default, or we can't, and
shouldn't, even if you pass -d.

15 years agocommon: break init into set_defaults and init steps
Sage Weil [Thu, 1 Jul 2010 16:25:43 +0000 (09:25 -0700)]
common: break init into set_defaults and init steps

This will let us tweak the default daemon/non-daemon settings as needed,
e.g. in the case of cfuse.

15 years agodebug: fixup
Sage Weil [Thu, 1 Jul 2010 16:10:41 +0000 (09:10 -0700)]
debug: fixup

15 years agocommon: disable logger by default; make daemon settings explicit in (3) callers
Sage Weil [Thu, 1 Jul 2010 15:55:52 +0000 (08:55 -0700)]
common: disable logger by default; make daemon settings explicit in (3) callers

15 years agodebug: add max count for name symlinks
Sage Weil [Thu, 1 Jul 2010 15:35:25 +0000 (08:35 -0700)]
debug: add max count for name symlinks

15 years agovstart: normalize pid file names
Sage Weil [Thu, 1 Jul 2010 15:29:01 +0000 (08:29 -0700)]
vstart: normalize pid file names

15 years agodebug: revamp debug/logging
Sage Weil [Thu, 1 Jul 2010 15:28:24 +0000 (08:28 -0700)]
debug: revamp debug/logging

- By default, append to $type.$name.log.
- Get old $hostname.$pid + $type.$name symlink behavior only with
  g_conf.log_per_instance
- Add new g_conf.log_file option to force a particular file.

15 years agoclient: set MetaRequest::inode to the inode or directory inode whenever possible.
Greg Farnum [Wed, 30 Jun 2010 18:34:45 +0000 (11:34 -0700)]
client: set MetaRequest::inode to the inode or directory inode whenever possible.
Provides link to caps, useful when requests come back ESTALE or similar.

15 years agoclient: set inode/dentry fields where available
Greg Farnum [Wed, 30 Jun 2010 18:03:50 +0000 (11:03 -0700)]
client: set inode/dentry fields where available

15 years agoclient: remove now-unused authority and pick_replica functions
Greg Farnum [Tue, 29 Jun 2010 22:52:04 +0000 (15:52 -0700)]
client: remove now-unused authority and pick_replica functions

15 years agoclient: rewrite choose_target_mds to more closely mirror kclient logic and be correct.
Greg Farnum [Tue, 29 Jun 2010 22:51:02 +0000 (15:51 -0700)]
client: rewrite choose_target_mds to more closely mirror kclient logic and be correct.

15 years agoclient: make fragmap updates a little more space efficient.
Greg Farnum [Tue, 29 Jun 2010 22:50:13 +0000 (15:50 -0700)]
client: make fragmap updates a little more space efficient.

If the auth is the same as the parent, don't add to the map (and remove if
it's already there). Absence is interpreted the same as a value of -1.

15 years agoclient: add mds_num to MDSSession struct
Greg Farnum [Tue, 29 Jun 2010 22:48:47 +0000 (15:48 -0700)]
client: add mds_num to MDSSession struct

15 years agoclient: clean up Inode::authority
Greg Farnum [Tue, 29 Jun 2010 17:32:18 +0000 (10:32 -0700)]
client: clean up Inode::authority

15 years agoclient: move all mds selection code into choose_target_mds
Greg Farnum [Tue, 29 Jun 2010 17:29:24 +0000 (10:29 -0700)]
client: move all mds selection code into choose_target_mds

15 years agoclient: remove dead code
Greg Farnum [Tue, 29 Jun 2010 17:28:24 +0000 (10:28 -0700)]
client: remove dead code

15 years agoclient: remove Inode::dir_auth data member
Greg Farnum [Tue, 29 Jun 2010 17:27:52 +0000 (10:27 -0700)]
client: remove Inode::dir_auth data member

15 years agoclient: initialize auth_caps in Inode! Prevents bad pointer derefs.
Greg Farnum [Wed, 30 Jun 2010 20:44:27 +0000 (13:44 -0700)]
client: initialize auth_caps in Inode! Prevents bad pointer derefs.

15 years agoqa: set runoncfuse.sh to run cfuse in background and kill at end of tests.
Greg Farnum [Wed, 30 Jun 2010 20:44:06 +0000 (13:44 -0700)]
qa: set runoncfuse.sh to run cfuse in background and kill at end of tests.

This is an imperfect solution at best, since it doesn't handle errors at all,
but it'll have to do for now.

15 years agorbd: some header cleanup
Yehuda Sadeh [Wed, 30 Jun 2010 22:36:58 +0000 (15:36 -0700)]
rbd: some header cleanup

15 years agorbdtool: implement snap rollback
Yehuda Sadeh [Wed, 30 Jun 2010 22:07:25 +0000 (15:07 -0700)]
rbdtool: implement snap rollback

15 years agobuffer: handle write_fd() on empty bufferlist
Sage Weil [Wed, 30 Jun 2010 22:44:38 +0000 (15:44 -0700)]
buffer: handle write_fd() on empty bufferlist

15 years agoosd: fix, cleanup ack/disk reply logic
Sage Weil [Wed, 30 Jun 2010 21:40:20 +0000 (14:40 -0700)]
osd: fix, cleanup ack/disk reply logic

There was a bug where we would get no reply if we could send ondisk but
the client didn't want it.  This simplifies and cleans up the checks
to make more sense, removing the can_* helpers that were hiding which
checks were being done.

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agobuffer: make write_fd() tolerate zero length ptrs in the bufferlist
Sage Weil [Wed, 30 Jun 2010 21:18:06 +0000 (14:18 -0700)]
buffer: make write_fd() tolerate zero length ptrs in the bufferlist

15 years agomsgr: clean up aborted message handling
Sage Weil [Wed, 30 Jun 2010 19:09:14 +0000 (12:09 -0700)]
msgr: clean up aborted message handling

No need for a dummy message.  Make the (single) caller handle returning
no message as a non-error.

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agomsgr: release bytes reserved from throttlers in failure paths
Sage Weil [Wed, 30 Jun 2010 19:08:18 +0000 (12:08 -0700)]
msgr: release bytes reserved from throttlers in failure paths

If we don't release those bytes, the throttler count eventually fills up
with bytes we were going to read but didn't (due to socket errors, etc)
until we can't read anything.

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agouclient: rename make_path; cleanup/move make_request_from_Meta
Sage Weil [Wed, 30 Jun 2010 18:14:50 +0000 (11:14 -0700)]
uclient: rename make_path; cleanup/move make_request_from_Meta

Document, make it clear what kind of "path" it's really building.

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agoman: add man pages for cclass and cclsinfo
Yehuda Sadeh [Wed, 30 Jun 2010 17:32:21 +0000 (10:32 -0700)]
man: add man pages for cclass and cclsinfo

15 years agofilejournal: fix buffer realignment condition; useful debug output on error
Sage Weil [Wed, 30 Jun 2010 15:34:26 +0000 (08:34 -0700)]
filejournal: fix buffer realignment condition; useful debug output on error

We don't care about contiguous, only about page alignment.

15 years agorbdtool: fix uninitialized variable
Yehuda Sadeh [Wed, 30 Jun 2010 05:58:52 +0000 (22:58 -0700)]
rbdtool: fix uninitialized variable

15 years agorbdtool: implement rename
Yehuda Sadeh [Wed, 30 Jun 2010 00:22:39 +0000 (17:22 -0700)]
rbdtool: implement rename

15 years agoconfig: update __isarg before reading config param
Yehuda Sadeh [Tue, 29 Jun 2010 23:23:27 +0000 (16:23 -0700)]
config: update __isarg before reading config param

15 years agodebian: include cclsinfo
Sage Weil [Tue, 29 Jun 2010 23:25:27 +0000 (16:25 -0700)]
debian: include cclsinfo

15 years agoMakefile: include cclsinfo; lose the old .cc
Sage Weil [Tue, 29 Jun 2010 23:23:11 +0000 (16:23 -0700)]
Makefile: include cclsinfo; lose the old .cc

15 years agoMakefile: include cclass.in in dist
Sage Weil [Tue, 29 Jun 2010 23:08:33 +0000 (16:08 -0700)]
Makefile: include cclass.in in dist

15 years agobuffer: write_fd should skip empty buffers
Yehuda Sadeh [Tue, 29 Jun 2010 22:59:20 +0000 (15:59 -0700)]
buffer: write_fd should skip empty buffers

15 years agorbdtool: init rbd block id, later will be used for rename
Yehuda Sadeh [Tue, 29 Jun 2010 22:15:00 +0000 (15:15 -0700)]
rbdtool: init rbd block id, later will be used for rename

15 years agocclass.in: LIBDIR=.libs in certain cases
Yehuda Sadeh [Tue, 29 Jun 2010 22:13:56 +0000 (15:13 -0700)]
cclass.in: LIBDIR=.libs in certain cases

15 years agocls: cls_read, cls_cxx_read return the number of bytes read
Yehuda Sadeh [Tue, 29 Jun 2010 22:11:04 +0000 (15:11 -0700)]
cls: cls_read, cls_cxx_read return the number of bytes read

15 years agoconfig: use <<20 for MB
Sage Weil [Mon, 28 Jun 2010 22:32:04 +0000 (15:32 -0700)]
config: use <<20 for MB

15 years agoscript/plot.pl: don't pause
Sage Weil [Mon, 28 Jun 2010 21:18:31 +0000 (14:18 -0700)]
script/plot.pl: don't pause

Run like so:

 $ script/plot.pl path/to/log osd c_wrb [smooth bezier] | gnuplot -persist

15 years agomsgr: use dedicated reaper thread
Sage Weil [Mon, 28 Jun 2010 21:15:59 +0000 (14:15 -0700)]
msgr: use dedicated reaper thread

We were calling the reaper from the wait() loop.  The problem is that
the OSD has two messengers, and only the first was in wait().. the second
wait() was only called after the first terminated (i.e, when the OSD was
shutting down).

Instead, launch a separate reaper thread when we bind, and close it out
on shutdown right after the accepter.

15 years agoosd: removed unused RepGather::indata
Sage Weil [Tue, 29 Jun 2010 21:38:39 +0000 (14:38 -0700)]
osd: removed unused RepGather::indata

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agoosd: change write osd logging timing
Sage Weil [Tue, 29 Jun 2010 21:38:28 +0000 (14:38 -0700)]
osd: change write osd logging timing

15 years agoosd: always use original Connection when replying
Sage Weil [Tue, 29 Jun 2010 21:32:28 +0000 (14:32 -0700)]
osd: always use original Connection when replying

...even when the op came from another OSD.  Not that that should happen
anyway, since we don't forward messages currently.  (And can't, since the
OSD doesn't initiate connections to the client!)

15 years agoosd: always include osd op result, result code in the first reply
Sage Weil [Tue, 29 Jun 2010 21:31:12 +0000 (14:31 -0700)]
osd: always include osd op result, result code in the first reply

15 years agoosd: track open repops in logger
Sage Weil [Mon, 28 Jun 2010 20:19:35 +0000 (13:19 -0700)]
osd: track open repops in logger

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agoosd: add 'logger reset' command
Sage Weil [Mon, 28 Jun 2010 20:19:26 +0000 (13:19 -0700)]
osd: add 'logger reset' command

15 years agojournal: set max journal write to 10MB
Sage Weil [Mon, 28 Jun 2010 18:44:26 +0000 (11:44 -0700)]
journal: set max journal write to 10MB

If we take too big a bite of data to write in a single writev(2), we can
end up making performance worse, because everyone waits for the full write
to complete.  Bigger writes mean better throughput but higher latency.
So, balance the two by placing some upper limit.

15 years agoconf: fix parsing when there's no eol at the end of file
Yehuda Sadeh [Tue, 29 Jun 2010 16:57:01 +0000 (09:57 -0700)]
conf:  fix parsing when there's no eol at the end of file

15 years agomsg: fix entity_addr_t::parse() to return false on failure
Sage Weil [Tue, 29 Jun 2010 15:34:53 +0000 (08:34 -0700)]
msg: fix entity_addr_t::parse() to return false on failure

15 years agomkcephfs: Fix wrong maxosd when OSD ids are random ordered in ceph.conf
CC Lien [Tue, 29 Jun 2010 14:29:11 +0000 (22:29 +0800)]
mkcephfs: Fix wrong maxosd when OSD ids are random ordered in ceph.conf

Hi

I got a trouble that mkcephfs will have wrong "maxosd" when you have
ceph.conf with OSD ids in random order like:

[osd2]
...
[osd0]
...
[osd1]
...

In this case, you will got "2" for the "maxosd", instead of 3.
After adding a sort, the problem seems solved.

Cheers,
CC Lien

Signed-off-by: CC Lien <cc_lien@tcloudcomputing.com>
Signed-off-by: Sage Weil <sage@newdream.net>
15 years agofilejournal: fix journal write_pos advance
Sage Weil [Mon, 28 Jun 2010 18:34:29 +0000 (11:34 -0700)]
filejournal: fix journal write_pos advance

This was broken by bd4188a02abff9efffb87a0a2031efe51c1b4d9a.  @pos needs to
be advanced (it is pass by reference) or else we just overwrite the same
bytes at the journal start over and over again.

15 years agomount.ceph: update mount options
Sage Weil [Mon, 28 Jun 2010 16:55:12 +0000 (09:55 -0700)]
mount.ceph: update mount options

Signed-off-by: Thomas Mueller <thomas@chaschperli.ch>
Signed-off-by: Sage Weil <sage@newdream.net>
15 years agomsgr: fix throttle deadlock
Sage Weil [Sat, 26 Jun 2010 17:28:38 +0000 (10:28 -0700)]
msgr: fix throttle deadlock

Do msgr throttle after peer policy throttle.  The msgr (dispatch) throttle
is shortlived and won't deadlock (unless dispatch blocks), so it's safe to
take last.  In contrast, the policy throttle carries over the lifetime of
the message, and may block until replication completes or whatever else.

15 years agocrushwrapper: gracefully handle crush error
Sage Weil [Sat, 26 Jun 2010 04:46:23 +0000 (21:46 -0700)]
crushwrapper: gracefully handle crush error

crush_do_rule can return <0 in certain error cases (e.g., forcefed device
does not exist in crush map).  We should take that to mean an empty []
result instead of crashing.

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agocrushtool: add simple test function
Sage Weil [Sat, 26 Jun 2010 02:54:18 +0000 (19:54 -0700)]
crushtool: add simple test function

15 years agocrush: fix "in" threshold to be 1.0, not 0.1
Sage Weil [Sat, 26 Jun 2010 02:20:18 +0000 (19:20 -0700)]
crush: fix "in" threshold to be 1.0, not 0.1

We were effectively counting any item with weight > .1 to be weight 1.0.

15 years agocfuse: remove some NPEs with ESTALE from MDSes.
Greg Farnum [Fri, 25 Jun 2010 23:10:35 +0000 (16:10 -0700)]
cfuse: remove some NPEs with ESTALE from MDSes.

Under certain circumstances this continues to let you get ESTALE
and not handle it, but those are still being analyzed for a fix.

15 years agomds: keep cap follows above in->first in FLUSHSNAP
Sage Weil [Thu, 24 Jun 2010 23:49:12 +0000 (16:49 -0700)]
mds: keep cap follows above in->first in FLUSHSNAP

The client has a follows of 0 initially, which is correct (it does follow
0, and there are no prior snaps).  But the inode has ->first of 2, which
is also fine.  The follows here needs to be at least higher than the
inode first, though, or the caps cloning gets off...

15 years agoqa: read recently written/deleted data back from snap
Sage Weil [Thu, 24 Jun 2010 23:48:01 +0000 (16:48 -0700)]
qa: read recently written/deleted data back from snap

15 years agomds: fix client cap condition
Sage Weil [Thu, 24 Jun 2010 22:50:47 +0000 (15:50 -0700)]
mds: fix client cap condition

In 551a12f52e36 we fixed a bug with cow_inode() where the
cap->client_follows didn't match last precisely.  Instead, we compare
to first.  But the == is too strict.. cap follows that is equal _or_older_
than the clone's first should be copied to the clone inode.

This fixes the simple test case
 $ echo asdf > bar ; mkdir .snap/bar ; rm bar ; cat .snap/bar/bar
 asdf
(Previously we would get nothing unless we waited for the cap to flush on
its own.)

15 years agoceph_fs: add CEPH_LOCK_IFLOCK so its inclusion elsewhere continues to build
Greg Farnum [Thu, 24 Jun 2010 18:51:59 +0000 (11:51 -0700)]
ceph_fs: add CEPH_LOCK_IFLOCK so its inclusion elsewhere continues to build

15 years agomds: add IFLOCK to wait bits to prevent collisions with lock branch
Greg Farnum [Thu, 24 Jun 2010 18:37:29 +0000 (11:37 -0700)]
mds: add IFLOCK to wait bits to prevent collisions with lock branch

15 years agocrush: fix recursion through intervening types
Sage Weil [Thu, 24 Jun 2010 17:42:20 +0000 (10:42 -0700)]
crush: fix recursion through intervening types

This fixes pretty core behavior when doing recursion down the tree.  I
suspect it was broken when changing the retry behavior.

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agocrush: make CHOOSE_LEAF to behave when leaf type is encountered
Sage Weil [Thu, 24 Jun 2010 17:40:14 +0000 (10:40 -0700)]
crush: make CHOOSE_LEAF to behave when leaf type is encountered

We may not want to recursively call crush_choose() if we start out with a
leaf.  If that happens, we need to fill out the out2[] vector with
our result immediately.

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agoclient: resync with kernel's ioctl.h
Sage Weil [Thu, 24 Jun 2010 00:02:04 +0000 (17:02 -0700)]
client: resync with kernel's ioctl.h

This got munged up by the #ifndef/#define guard cleanup.

15 years agomds: fix SimpleLock wait_shift()
Sage Weil [Thu, 24 Jun 2010 17:38:36 +0000 (10:38 -0700)]
mds: fix SimpleLock wait_shift()

DVERSION was missing, others were overlapping...

15 years agoceph_fs: add CEPH_FEATURE_FLOCK to ceph_fs so its bit doesn't get covered again
Greg Farnum [Tue, 22 Jun 2010 18:52:34 +0000 (11:52 -0700)]
ceph_fs: add CEPH_FEATURE_FLOCK to ceph_fs so its bit doesn't get covered again

15 years agoosdmap: negative osd ids do not exist
Sage Weil [Wed, 23 Jun 2010 21:12:05 +0000 (14:12 -0700)]
osdmap: negative osd ids do not exist

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agocrush: behave when chooseleaf is given leaf type
Sage Weil [Wed, 23 Jun 2010 21:08:39 +0000 (14:08 -0700)]
crush: behave when chooseleaf is given leaf type

Fill in the out2 choose_leaf vector if it's defined.  This is necessary
because we may not recursively call choose on out2 if the item we're on is
not a bucket (e.g., when chooseleaf is given the leaf type 0).

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agomds: handle_export_caps, copy cap_import map (instead of swap content)
Yehuda Sadeh [Tue, 22 Jun 2010 21:11:53 +0000 (14:11 -0700)]
mds: handle_export_caps, copy cap_import map (instead of swap content)

This fixes #200. The client_map is later on swapped again in the new ESession(),
which left finish->client_map empty.

15 years agoclient: properly handle ESTALE by redirecting to auth MDS
Greg Farnum [Mon, 21 Jun 2010 23:27:32 +0000 (16:27 -0700)]
client: properly handle ESTALE by redirecting to auth MDS

15 years agoclient: don't send cap snap message to MDS if not dirty or writing.
Greg Farnum [Mon, 21 Jun 2010 17:54:32 +0000 (10:54 -0700)]
client: don't send cap snap message to MDS if not dirty or writing.

From Sage.

15 years agomds: only acquire_locks once in handle_client_open
Sage Weil [Mon, 21 Jun 2010 18:32:43 +0000 (11:32 -0700)]
mds: only acquire_locks once in handle_client_open

Subsequent calls will just return with 'already locked'

15 years agomds: rename handle_client_opent
Sage Weil [Mon, 21 Jun 2010 18:32:16 +0000 (11:32 -0700)]
mds: rename handle_client_opent

15 years agorados: more descriptive ls usage failure, stdout set implicitly
Yehuda Sadeh [Mon, 21 Jun 2010 18:43:30 +0000 (11:43 -0700)]
rados: more descriptive ls usage failure, stdout set implicitly

15 years agorbdtool: fix --list
Yehuda Sadeh [Mon, 21 Jun 2010 18:27:41 +0000 (11:27 -0700)]
rbdtool: fix --list

15 years agoosd: fix incorrect logic check on fsid comparison
Greg Farnum [Mon, 21 Jun 2010 17:35:54 +0000 (10:35 -0700)]
osd: fix incorrect logic check on fsid comparison

15 years agoosd: Warn and shutdown on a mismatched fsid, instead of failing an assert
Greg Farnum [Mon, 21 Jun 2010 16:45:43 +0000 (09:45 -0700)]
osd: Warn and shutdown on a mismatched fsid, instead of failing an assert

15 years agoadd helptext for option "snapdirname" to manpage of mount.ceph
Thomas Mueller [Mon, 21 Jun 2010 10:32:26 +0000 (10:32 +0000)]
add helptext for option "snapdirname" to manpage of mount.ceph

[ The following text is in the "UTF-8" character set. ]
    [ Your display is set for the "iso-8859-1" character set.  ]
    [ Some characters may be displayed incorrectly. ]

inspired by the addition to
http://ceph.newdream.net/wiki/Snapshots about the snapdirname
 option i've created a patch for the mount.ceph manpage

- Thomas

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agojournal: initialize applied_seq during journal replay
Sage Weil [Sun, 20 Jun 2010 21:41:19 +0000 (14:41 -0700)]
journal: initialize applied_seq during journal replay

This should avoid

#0  0x00007f41b1a18a75 in raise () from /lib/libc.so.6
#1  0x00007f41b1a1c5c0 in abort () from /lib/libc.so.6
#2  0x00007f41b22cd8e5 in __gnu_cxx::__verbose_terminate_handler() () from /usr/lib/libstdc++.so.6
#3  0x00007f41b22cbd16 in ?? () from /usr/lib/libstdc++.so.6
#4  0x00007f41b22cbd43 in std::terminate() () from /usr/lib/libstdc++.so.6
#5  0x00007f41b22cbe3e in __cxa_throw () from /usr/lib/libstdc++.so.6
#6  0x00000000005b39f8 in ceph::__ceph_assert_fail (assertion=0x5ec3b2 "seq >= last_committed_seq", file=<value optimized out>, line=711, func=<value optimized out>) at common/assert.cc:30
#7  0x00000000005649e1 in FileJournal::committed_thru (this=0x1116310, seq=0) at os/FileJournal.cc:711
#8  0x000000000055d265 in JournalingObjectStore::commit_finish (this=0x1125740) at os/JournalingObjectStore.cc:186
#9  0x00000000005543f3 in FileStore::sync_entry (this=0x1125740) at os/FileStore.cc:1714
#10 0x00000000004ef93d in FileStore::SyncThread::entry() ()
#11 0x0000000000469a4a in Thread::_entry_func (arg=0x6315) at ./common/Thread.h:39
#12 0x00007f41b28ab9ca in start_thread () from /lib/libpthread.so.0
#13 0x00007f41b1acb6cd in clone () from /lib/libc.so.6
#14 0x0000000000000000 in ?? ()

Signed-off-by: Sage Weil <sage@newdream.net>
15 years agomkcephfs: push conf to remote machines
Sage Weil [Sun, 20 Jun 2010 16:45:05 +0000 (09:45 -0700)]
mkcephfs: push conf to remote machines

Signed-off-by: Fred Ar <ar.fred@yahoo.com>
Signed-off-by: Sage Weil <sage@newdream.net>
15 years agomon: allow 10ms clock drift before complaining
Sage Weil [Sun, 20 Jun 2010 16:44:37 +0000 (09:44 -0700)]
mon: allow 10ms clock drift before complaining

15 years agodebian: include cclass
Fred Ar [Sat, 19 Jun 2010 15:40:55 +0000 (08:40 -0700)]
debian: include cclass

15 years agoinitscript: fix for dash
Fred Ar [Sat, 19 Jun 2010 15:40:00 +0000 (08:40 -0700)]
initscript: fix for dash

Signed-off-by: Fred Ar <ar.fred@yahoo.com>
15 years agocclass: fix for dash
Fred Ar [Sat, 19 Jun 2010 15:39:37 +0000 (08:39 -0700)]
cclass: fix for dash

Signed-off-by: Fred Ar <ar.fred@yahoo.com>
15 years agoinitscript: remove class loading for now
Sage Weil [Sat, 19 Jun 2010 04:26:41 +0000 (21:26 -0700)]
initscript: remove class loading for now

- only need to do it once, by connecting to a random monitor, not for
  each monitor
- not sure we should try it every time we start the monitor for all time,
  as opposed to once after mkfs, or whenever the admin chooses to load
  new classes

15 years agovstart: only need to load classes once, not for each monitor
Sage Weil [Sat, 19 Jun 2010 02:12:01 +0000 (19:12 -0700)]
vstart: only need to load classes once, not for each monitor

15 years agoMerge branch 'rbd-class' into unstable
Sage Weil [Fri, 18 Jun 2010 23:32:10 +0000 (16:32 -0700)]
Merge branch 'rbd-class' into unstable

15 years agoRevert "class: able to add classes when doing mkfs"
Sage Weil [Fri, 18 Jun 2010 23:31:38 +0000 (16:31 -0700)]
Revert "class: able to add classes when doing mkfs"

This reverts commit 65371482dd661c6dc3d251fb9fd4e01812bdaa0e.

15 years agofilestore: op_start when op is _queued_, so that q is drained on commit
Sage Weil [Fri, 18 Jun 2010 22:59:36 +0000 (15:59 -0700)]
filestore: op_start when op is _queued_, so that q is drained on commit

We need the store in a consistent state on commit, which means flushing
transactions such that we have all ops <= a given seq applied.  That is
handled by the commit_start()/commit_started() pair, but will only include
ops in the FileStore queue if we op_start when it is initially queued.
Which is exactly what we want, because the queue can reorder things, so
stopping just currently-being-applied updates will only keep transactions
atomic but not ordered.

Yup!

15 years agojournal: maintain applied_seq, use that value during the commit.
Sage Weil [Fri, 18 Jun 2010 22:09:05 +0000 (15:09 -0700)]
journal: maintain applied_seq, use that value during the commit.

Track which seq's have successfully applied.

This fixes writeahead mode.  Parallel mode is still broken...