]> git.apps.os.sepia.ceph.com Git - ceph.git/log
ceph.git
12 years agoMerge branch 'wip-5194' into next
Sage Weil [Mon, 17 Jun 2013 22:46:47 +0000 (15:46 -0700)]
Merge branch 'wip-5194' into next

Reviewed-by: Dan Mick <dan.mick@inktank.com>
Reviewed-by: Gary Lowell <gary.lowell@inktank.com>
12 years agoceph-disk: add some notes on wth we are up to
Sage Weil [Mon, 17 Jun 2013 22:43:40 +0000 (15:43 -0700)]
ceph-disk: add some notes on wth we are up to

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoOSD: we need to check pg ?.0 for resurrection
Samuel Just [Mon, 17 Jun 2013 20:09:21 +0000 (13:09 -0700)]
OSD: we need to check pg ?.0 for resurrection

Fixes: #5269
Signed-off-by: Samuel Just <sam.just@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>
12 years agoceph-disk: clear TERM to avoid libreadline hijinx
Sage Weil [Fri, 14 Jun 2013 23:29:10 +0000 (16:29 -0700)]
ceph-disk: clear TERM to avoid libreadline hijinx

The weird output from libreadline users is related to the TERM variable.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph-disk-udev: set up by-partuuid, -typeuuid symlinks on ancient udev
Sage Weil [Mon, 17 Jun 2013 16:49:46 +0000 (09:49 -0700)]
ceph-disk-udev: set up by-partuuid, -typeuuid symlinks on ancient udev

Make the ancient-udev/blkid workaround script for RHEL/CentOS create the
symlinks for us too.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph-disk: do not stop activate-all on first failure
Sage Weil [Sun, 16 Jun 2013 03:06:33 +0000 (20:06 -0700)]
ceph-disk: do not stop activate-all on first failure

Keep going even if we hit one activation error.  This avoids failing to
start some disks when only one of them won't start (e.g., because it
doesn't belong to the current cluster).

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph.spec: include partuuid rules in package
Sage Weil [Fri, 14 Jun 2013 23:30:24 +0000 (16:30 -0700)]
ceph.spec: include partuuid rules in package

Commit f3234c147e083f2904178994bc85de3d082e2836 missed this.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agorgw: escape prefix correctly when listing objects
Yehuda Sadeh [Fri, 14 Jun 2013 21:53:54 +0000 (14:53 -0700)]
rgw: escape prefix correctly when listing objects

Fixes: #5362
When listing objects prefix needs to be escaped correctly (the
same as with the marker). Otherwise listing objects with prefix
that starts with underscore doesn't work.
Backport: bobtail, cuttlefish

Signed-off-by: Yehuda Sadeh <yehuda@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>
12 years agoclient: fix ancient typo in caps revocation path
Sage Weil [Sat, 15 Jun 2013 15:48:37 +0000 (08:48 -0700)]
client: fix ancient typo in caps revocation path

If we have dropped all references to a revoked capability, send the ack
to the MDS.  This typo has been there since v0.7 (early 2009)!

Backport: cuttlefish
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>
12 years agoceph.spec: install/uninstall init script
Sage Weil [Fri, 14 Jun 2013 22:01:14 +0000 (15:01 -0700)]
ceph.spec: install/uninstall init script

This was commented out almost years ago in commit 9baf5ef4 but it is not
clear to me that it was correct to do so.  In any case, we are not
installing the rc.d links for ceph, which means it does not start up after
a reboot.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agosysvinit, upstart: ceph-disk activate-all on start
Sage Weil [Fri, 14 Jun 2013 20:39:03 +0000 (13:39 -0700)]
sysvinit, upstart: ceph-disk activate-all on start

On 'service ceph start' or 'service ceph start osd' or start ceph-osd-all
we should activate any osd GPT partitions.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph-disk: add 'activate-all'
Sage Weil [Fri, 14 Jun 2013 20:34:40 +0000 (13:34 -0700)]
ceph-disk: add 'activate-all'

Scan /dev/disk/by-parttypeuuid for ceph OSDs and activate them all.  This
is useful when the event didn't trigger on the initial udev event for
some reason.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoudev: /dev/disk/by-parttypeuuid/$type-$uuid
Sage Weil [Fri, 14 Jun 2013 20:23:52 +0000 (13:23 -0700)]
udev: /dev/disk/by-parttypeuuid/$type-$uuid

We need this to help trigger OSD activations.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agomon: make mark_me_down asserts match check
Sage Weil [Mon, 17 Jun 2013 03:13:51 +0000 (20:13 -0700)]
mon: make mark_me_down asserts match check

The OSD may have sent a request where the message source does not match
the target in the message.  Verify that the target matches so that it
matches the assert.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: remove space when prefix is blank
Sage Weil [Sun, 16 Jun 2013 23:49:05 +0000 (16:49 -0700)]
ceph: remove space when prefix is blank

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: fix return code for multi-target commands
Sage Weil [Sun, 16 Jun 2013 23:48:41 +0000 (16:48 -0700)]
ceph: fix return code for multi-target commands

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: error out properly when failing to get commands
Sage Weil [Sun, 16 Jun 2013 23:48:27 +0000 (16:48 -0700)]
ceph: error out properly when failing to get commands

If we make ret positive here we miss the failure check below.  Instead,
just set outs appropriately.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agotest/admin_socket/objecter_requests: fix test
Sage Weil [Sun, 16 Jun 2013 23:42:27 +0000 (16:42 -0700)]
test/admin_socket/objecter_requests: fix test

Commit 2bda9db1c24530cbaaa161b7ff0a80efa913aa78 added command_ops
to the result.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: do not print status to output file when talking to old mons
Sage Weil [Sun, 16 Jun 2013 20:36:19 +0000 (13:36 -0700)]
ceph: do not print status to output file when talking to old mons

The old cli would send the status message to stdout instead of stderr;
we try to emulate that behavior when talking to old monitors because
they send some useful data to outs instead of the data payload.
However, when outputting to a *file*, the outs would still go to
stdout.  Maintain that so that, e.g.,

 ceph mon getmap -o /tmp/foo

doesn't prefix the monmap with 'got latest monmap\n'.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agocommon/Preforker: fix broken recursion on exit(3)
Sage Weil [Sat, 15 Jun 2013 15:14:40 +0000 (08:14 -0700)]
common/Preforker: fix broken recursion on exit(3)

If we exit via preforker, call exit(3) and not recursively back into
Preforker::exit(r).  Otherwise you get a hang with the child blocked
at:

Thread 1 (Thread 0x7fa08962e7c0 (LWP 5419)):
#0  0x000000309860e0cd in write () from /lib64/libpthread.so.0
#1  0x00000000005cc906 in Preforker::exit(int) ()
#2  0x00000000005c8dfb in main ()

and the parent at

#0  0x000000309860eba7 in waitpid () from /lib64/libpthread.so.0
#1  0x00000000005cc87a in Preforker::parent_wait() ()
#2  0x00000000005c75ae in main ()

Backport: cuttlefish
Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoosd/OSDMap: fix is_blacklisted()
Sage Weil [Sat, 15 Jun 2013 16:10:46 +0000 (09:10 -0700)]
osd/OSDMap: fix is_blacklisted()

You can only call set_port() if is_ip() is true (there is an assert in
the accessor).

Fixes: #5366
Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: pass --format=foo to old monitors
Sage Weil [Sat, 15 Jun 2013 00:30:02 +0000 (17:30 -0700)]
ceph: pass --format=foo to old monitors

And --threshold too, although.. really.

Signed-off-by: Sage Weil <sage@inktank.com>
Reviwed-by: Dan Mick <dan.mick@inktank.com>
12 years agoceph: add newline when using old monitors
Sage Weil [Sat, 15 Jun 2013 00:30:44 +0000 (17:30 -0700)]
ceph: add newline when using old monitors

The old tool would print a newline after outs, e.g. from 'ceph osd create'.
Do the same when we are talking to old monitors.  Also, put outs at the
top, not the bottom!

Tweak the json code to not add the newline again if we already did so
above.

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Dan Mick <dan.mick@inktank.com>
12 years agoceph.in: zero-arg invocation was broken (check array length)
Dan Mick [Fri, 14 Jun 2013 23:51:40 +0000 (16:51 -0700)]
ceph.in: zero-arg invocation was broken (check array length)

Also remove stray comment char

Signed-off-by: Dan Mick <dan.mick@inktank.com>
12 years agorules: Don't disable tcmalloc on ARM (and other non-intel)
Gary Lowell [Thu, 13 Jun 2013 23:38:26 +0000 (16:38 -0700)]
rules:  Don't disable tcmalloc on ARM (and other non-intel)

Fixes #5342

Signed-off-by: Gary Lowell <gary.lowell@inktank.com>
12 years agoudev: drop useless --mount argument to ceph-disk
Sage Weil [Fri, 14 Jun 2013 05:02:03 +0000 (22:02 -0700)]
udev: drop useless --mount argument to ceph-disk

It doesn't mean anything anymore; drop it.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph-disk-udev: activate-journal
Sage Weil [Fri, 14 Jun 2013 05:01:34 +0000 (22:01 -0700)]
ceph-disk-udev: activate-journal

Trigger 'ceph-disk activate-journal' from the alt udev rules.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph-disk: do not use mount --move (or --bind)
Sage Weil [Fri, 14 Jun 2013 04:56:23 +0000 (21:56 -0700)]
ceph-disk: do not use mount --move (or --bind)

The kernel does not let you mount --move when the parent mount is
shared (see, e.g., https://bugzilla.redhat.com/show_bug.cgi?id=917008
for another person this also confused).  We can't use --bind either
since that (on RHEL at least) screws up /etc/mtab so that the final
result looks like

 /var/lib/ceph/tmp/mnt.HNHoXU /var/lib/ceph/osd/ceph-0 none rw,bind 0 0

Instead, mount the original dev in the final location and then umount
from the old location.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph.spec: include by-partuuid udev workaround rules
Sage Weil [Fri, 14 Jun 2013 04:22:53 +0000 (21:22 -0700)]
ceph.spec: include by-partuuid udev workaround rules

These are need for old or buggy udev.  Having them for new and unbroken
udev is harmless.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph.spec: add missing ceph_test_rados_api_cmd to package
Sage Weil [Fri, 14 Jun 2013 04:21:28 +0000 (21:21 -0700)]
ceph.spec: add missing ceph_test_rados_api_cmd to package

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: flush stderr, stdout for sane output; add prefix
Sage Weil [Fri, 14 Jun 2013 19:35:46 +0000 (12:35 -0700)]
ceph: flush stderr, stdout for sane output; add prefix

Aie.

e.g., ceph tell mon.* injectargs '--debug-ms 1'

 mon.a: injectargs:debug_ms=1/1
 mon.b: injectargs:debug_ms=1/1
 mon.c: injectargs:debug_ms=1/1

or

 osd.0: debug_ms=1/1
 osd.1: debug_ms=1/1
 osd.2: Problem getting command descriptions from ('osd', '2'), ENXIO
 osd.3: Problem getting command descriptions from ('osd', '3'), ENXIO
 osd.4: Problem getting command descriptions from ('osd', '4'), ENXIO
 osd.5: Problem getting command descriptions from ('osd', '5'), ENXIO

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Dan Mick <dan.mick@inktank.com>
12 years agoceph-disk: work around buggy rhel/centos parted
Sage Weil [Fri, 14 Jun 2013 19:10:49 +0000 (12:10 -0700)]
ceph-disk: work around buggy rhel/centos parted

parted on RHEL/Centos prefixes the *machine readable output* with

 1b 5b 3f 31 30 33 34 68

Note that the same thing happens when you 'import readline' in python.

Work around it!

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agomon: OSDMonitor: don't ignore apply_incremental()'s return on UfP [1]
Joao Eduardo Luis [Fri, 14 Jun 2013 16:11:43 +0000 (17:11 +0100)]
mon: OSDMonitor: don't ignore apply_incremental()'s return on UfP [1]

apply_incremental() may return -EINVAL.  Don't ignore it.

[1] UfP = Update from Paxos

Fixes: #5343
Signed-off-by: Joao Eduardo Luis <joao.luis@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
12 years agoupstart: start ceph-all on runlevel [2345]
Sage Weil [Fri, 14 Jun 2013 18:21:25 +0000 (11:21 -0700)]
upstart: start ceph-all on runlevel [2345]

Starting when only one network interface has started breaks machines with
multiple nics in very problematic ways.

There may be an earlier trigger that we can use for cases where other
services on the local machine depend on ceph, but for now this is better
than the existing behavior.

See #5248

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: fix mon.*
Sage Weil [Fri, 14 Jun 2013 18:00:46 +0000 (11:00 -0700)]
ceph: fix mon.*

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoMerge pull request #362 from ceph/wip-4984
Dan Mick [Fri, 14 Jun 2013 02:37:37 +0000 (19:37 -0700)]
Merge pull request #362 from ceph/wip-4984

ceph-disk: udev/partprobe redo, zap command, activate-journal command

12 years agoceph-fuse: fix uninitialized variable
Sage Weil [Fri, 14 Jun 2013 01:13:34 +0000 (18:13 -0700)]
ceph-fuse: fix uninitialized variable

There is a delete call in the out_mc_start_failed path.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph-disk: implement 'activate-journal' 362/head
Sage Weil [Thu, 13 Jun 2013 22:54:58 +0000 (15:54 -0700)]
ceph-disk: implement 'activate-journal'

Activate an osd via its journal device.  udev populates its symlinks and
triggers events in an order that is not related to whether the device is
an osd data partition or a journal.  That means that triggering
'ceph-disk activate' can happen before the journal (or journal symlink)
is present and then fail.

Similarly, it may be that they are on different disks that are hotplugged
with the journal second.

This can be wired up to the journal partition type to ensure that osds are
started when the journal appears second.

Include the udev rules to trigger this.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph-disk: call partprobe outside of the prepare lock; drop udevadm settle
Sage Weil [Wed, 12 Jun 2013 01:35:01 +0000 (18:35 -0700)]
ceph-disk: call partprobe outside of the prepare lock; drop udevadm settle

After we change the final partition type, sgdisk may or may not trigger a
udev event, depending on how well udev is behaving (it varies between
distros, it seems).  The old code would often settle and wait for udev to
activate the device, and then partprobe would uselessly fail because it
was already mounted.

Call partprobe only at the very end, after prepare is done.  This ensures
that if partprobe calls udevadm settle (which is sometimes does) we do not
get stuck.

Drop the udevadm settle.  I'm not sure what this accomplishes; take it out,
at least until we determine we need it.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph-disk: add 'zap' command
Sage Weil [Thu, 13 Jun 2013 18:03:37 +0000 (11:03 -0700)]
ceph-disk: add 'zap' command

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agolibrados: add missing #include
Sage Weil [Fri, 14 Jun 2013 00:38:02 +0000 (17:38 -0700)]
librados: add missing #include

librados/librados.cc: In function 'int rados_mon_command_target(void*, const char*, const char**, size_t, const char*, size_t, char**, size_t*, char**, size_t*)':
error: librados/librados.cc:1877: 'LONG_MAX' was not declared in this scope
error: librados/librados.cc:1877: 'LONG_MIN' was not declared in this scope

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agolibrados: wait for osdmap for commands that need it
Sage Weil [Thu, 13 Jun 2013 23:39:30 +0000 (16:39 -0700)]
librados: wait for osdmap for commands that need it

In commit 7e1cf87b5158c870e2a118ed6d316be8cb9818ce we stopped waiting for
the osdmap on start because the Objecter will normally wait, but for some
commands we assume the osdmap is recent(ish).

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agoMerge branch 'wip-objecter' into next
Sage Weil [Thu, 13 Jun 2013 23:15:44 +0000 (16:15 -0700)]
Merge branch 'wip-objecter' into next

Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agoosdc/Objecter: dump command ops
Sage Weil [Thu, 13 Jun 2013 23:01:31 +0000 (16:01 -0700)]
osdc/Objecter: dump command ops

Dump command_ops along with everything else.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoosdc/Objecter: ping osds for which we have pending commands
Sage Weil [Thu, 13 Jun 2013 22:57:57 +0000 (15:57 -0700)]
osdc/Objecter: ping osds for which we have pending commands

As with ops and linger_ops, this ensures we detect connection resets.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph.in: refuse 'ceph <type> tell' commands; suggest 'ceph tell <type>'
Dan Mick [Thu, 13 Jun 2013 22:48:32 +0000 (15:48 -0700)]
ceph.in: refuse 'ceph <type> tell' commands; suggest 'ceph tell <type>'

Signed-off-by: Dan Mick <dan.mick@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
12 years agoceph.in: argparsing cleanup: suppress --completion, add help
Dan Mick [Thu, 13 Jun 2013 22:30:38 +0000 (15:30 -0700)]
ceph.in: argparsing cleanup: suppress --completion, add help

Options -v, --verbose, --concise didn't have helpstrings
Option --completion doesn't quite work yet, and should be hidden anyway

Signed-off-by: Dan Mick <dan.mick@inktank.com>
12 years agoosdc/Objecter: kick command ops on osd con resets
Sage Weil [Thu, 13 Jun 2013 22:13:47 +0000 (15:13 -0700)]
osdc/Objecter: kick command ops on osd con resets

Resend osd/pg commands on the OSDSession, just as we do with other request
types.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoosdc/Objecter: add perfcounters for commands
Sage Weil [Thu, 13 Jun 2013 22:13:18 +0000 (15:13 -0700)]
osdc/Objecter: add perfcounters for commands

This matches the other counters we maintain for other kinds of ops.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agomon: fix idempotency of 'osd crush add'
Sage Weil [Thu, 13 Jun 2013 21:01:01 +0000 (14:01 -0700)]
mon: fix idempotency of 'osd crush add'

If we add an item that already exists in particular position, we should
update instead of inserting it; the CrushWrapper methods are not
idempotent.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agolibrados: do not wait for osdmap on start
Sage Weil [Thu, 13 Jun 2013 21:42:03 +0000 (14:42 -0700)]
librados: do not wait for osdmap on start

If we abort while waiting, we incorrect clean up (we switch the state value
incorrectly, and also fail to clean up the initialized objecter).

Intead, skip this wait.. it's useless!

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agomon/MonmapMonitor: remove unused label
Sage Weil [Thu, 13 Jun 2013 18:27:49 +0000 (11:27 -0700)]
mon/MonmapMonitor: remove unused label

mon/MonmapMonitor.cc: In member function 'bool MonmapMonitor::preprocess_command(MMonCommand*)':
mon/MonmapMonitor.cc:273:2: warning: label 'out' defined but not used [-Wunused-label]

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agomon/MonCap: bootstrap-* need to subscribe to osdmap, monmap
Sage Weil [Thu, 13 Jun 2013 18:27:23 +0000 (11:27 -0700)]
mon/MonCap: bootstrap-* need to subscribe to osdmap, monmap

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoMerge branch 'wip-tell' into next
Sage Weil [Thu, 13 Jun 2013 16:27:15 +0000 (09:27 -0700)]
Merge branch 'wip-tell' into next

Reviewed-by: Dan Mick <dan.mick@inktank.com>
12 years agomon: remove support for 'mon tell ...' and 'osd tell ...'
Sage Weil [Wed, 12 Jun 2013 23:56:45 +0000 (16:56 -0700)]
mon: remove support for 'mon tell ...' and 'osd tell ...'

It doesn't work.  The commands the ceph cli sends are vector<string>, and
the mon expects json.

Leave the MDS on in place since ceph-mds still takes strings.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: add support for 'tell mon.X ...'
Sage Weil [Wed, 12 Jun 2013 23:55:03 +0000 (16:55 -0700)]
ceph: add support for 'tell mon.X ...'

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agolibrados: new rados_mon_command_target to talk to a specific monitor
Sage Weil [Wed, 12 Jun 2013 23:36:39 +0000 (16:36 -0700)]
librados: new rados_mon_command_target to talk to a specific monitor

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoUpdate README dependency lists
Dan Mick [Thu, 13 Jun 2013 05:25:04 +0000 (22:25 -0700)]
Update README dependency lists

Signed-off-by: Dan Mick <dan.mick@inktank.com>
12 years agoceph-disk: extra dash in error message
Dan Mick [Thu, 13 Jun 2013 05:22:42 +0000 (22:22 -0700)]
ceph-disk: extra dash in error message

Signed-off-by: Dan Mick <dan.mick@inktank.com>
12 years agoClean up CrushWrapper methods that take string: no c_str() necessary
Dan Mick [Thu, 13 Jun 2013 03:59:49 +0000 (20:59 -0700)]
Clean up CrushWrapper methods that take string: no c_str() necessary

Signed-off-by: Dan Mick <dan.mick@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
12 years agoOSDMonitor: osd id when id already exists needs to come to stdout too
Dan Mick [Thu, 13 Jun 2013 03:59:08 +0000 (20:59 -0700)]
OSDMonitor: osd id when id already exists needs to come to stdout too

Found by qa/workunits/mon/osd.sh

Signed-off-by: Dan Mick <dan.mick@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
12 years agoceph, mon/OSDMonitor: fix up osd crush commands for <osd.N> or <N>
Dan Mick [Thu, 13 Jun 2013 01:08:17 +0000 (18:08 -0700)]
ceph, mon/OSDMonitor: fix up osd crush commands for <osd.N> or <N>

The new parsing code had been trying to allow flexibility for the
'old form' commands (where id could be different from N in osd.N),
but also accept 'new form' commands.  The new rule is that where
there's an OSD specified in the osd crush command, it is of type
CephOsdName, which can be an id *or* 'osd.<id>', but not both.

Pass CephOsdName as int64_t 'id' for convenience in mon code

Signed-off-by: Dan Mick <dan.mick@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
12 years agomon/MonClient: send commands to a specific monitor
Sage Weil [Wed, 12 Jun 2013 23:36:21 +0000 (16:36 -0700)]
mon/MonClient: send commands to a specific monitor

This implementation is limited: we direct our command by reopening
a session with the specific monitor.  If there is more than one of these
queued we will fail to reach either.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: implement 'ceph tell osd.* ...'
Sage Weil [Wed, 12 Jun 2013 21:55:15 +0000 (14:55 -0700)]
ceph: implement 'ceph tell osd.* ...'

Send the command to each target.  Do this in series, for now.  Error out if
any one fails.

Later, we should do them in parallel.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoosdc/Objecter: fix handling for osd_command dne/down cases
Sage Weil [Thu, 13 Jun 2013 01:13:12 +0000 (18:13 -0700)]
osdc/Objecter: fix handling for osd_command dne/down cases

Generalize the map check machinery that the pool dne check uses to also
get the latest map for OSD down/dne checks.  This is better semantics, but
more important fixes the more immediate bug of returning the error code
to the caller from the osd_command -> _submit_command (that is ignored by
pretty much any caller) and then never triggering the callback.

Fixes: #5331
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agoceph: only use readline when in interactive mode
Sage Weil [Wed, 12 Jun 2013 23:22:45 +0000 (16:22 -0700)]
ceph: only use readline when in interactive mode

A mere

  import readline

line is dumping this to stdout on CentOS 6.3:

  00000000  1b 5b 3f 31 30 33 34 68  .[?1034h

That confuses non-terminals that read from stdout, so only import when we
are in the interactive mode.

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Dan Mick <dan.mick@inktank.com>
12 years agomon: fix read of format_version out of leveldb
Sage Weil [Wed, 12 Jun 2013 18:23:23 +0000 (11:23 -0700)]
mon: fix read of format_version out of leveldb

The get_version(string, string) is the wrong method; it combines the two
args into a key that is nested inside prefix (so it's prefix/a/b), but we
want perfix/format_version.  Add a method to grab an int for this
particular combo and use that.

This fixes an infinite loop when we actually trigger this code.

Bug introduced by f43c974571beac0c8e54fa699bfa96a1befaf56c.

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Joao Eduardo Luis <joao.luis@inktank.com>
12 years agoMerge branch 'next'
Gary Lowell [Wed, 12 Jun 2013 22:00:05 +0000 (15:00 -0700)]
Merge branch 'next'

12 years agoceph: filter out empty lines from osdids()
Sage Weil [Wed, 12 Jun 2013 21:54:10 +0000 (14:54 -0700)]
ceph: filter out empty lines from osdids()

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: accept osd.* as a valid name
Sage Weil [Wed, 12 Jun 2013 21:53:56 +0000 (14:53 -0700)]
ceph: accept osd.* as a valid name

This will be used for 'ceph tell osd.* ...'

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: make life easier on developers by handling in-tree runs
Dan Mick [Wed, 12 Jun 2013 02:46:53 +0000 (19:46 -0700)]
ceph: make life easier on developers by handling in-tree runs

If <path-to-ceph> contains pybind and .libs:
- prepend <path-to-ceph>/pybind to PYTHONPATH
- append <path-to-ceph>/.libs to LD_LIBRARY_PATH if not already there
  and exec self so it takes effect

Signed-off-by: Dan Mick <dan.mick@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
12 years agoqa/workunits/cephtool/test.sh: look for 'ceph log' via -w, not in log file
Sage Weil [Wed, 12 Jun 2013 21:00:24 +0000 (14:00 -0700)]
qa/workunits/cephtool/test.sh: look for 'ceph log' via -w, not in log file

'ceph-conf ...' doesn't give you final/default values, only what is in the
conf file.  Use -w output to test this instead.

Fixes: #5327
Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoceph: flush stdout on watch print
Sage Weil [Wed, 12 Jun 2013 20:59:37 +0000 (13:59 -0700)]
ceph: flush stdout on watch print

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agoMerge pull request #357 from atwardowski/patch-1
Sage Weil [Wed, 12 Jun 2013 20:50:15 +0000 (13:50 -0700)]
Merge pull request #357 from atwardowski/patch-1

Usage log and ops log are disabled by defaults since 0.56

12 years agoUsage log and ops log are disabled by defaults since 0.56 357/head
atwardowski [Wed, 12 Jun 2013 20:48:44 +0000 (17:48 -0300)]
Usage log and ops log are disabled by defaults since 0.56

http://ceph.com/docs/next/release-notes/#v0-56-bobtail

12 years agomon: fix 'pg dump_stuck' stuckops type
Sage Weil [Wed, 12 Jun 2013 20:39:30 +0000 (13:39 -0700)]
mon: fix 'pg dump_stuck' stuckops type

It's a list.

Fixes: #5332
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Dan Mick <dan.mick@inktank.com>
12 years agoMerge remote-tracking branch 'gh/wip_5238'
Sage Weil [Wed, 12 Jun 2013 20:31:22 +0000 (13:31 -0700)]
Merge remote-tracking branch 'gh/wip_5238'

Reviewed-by: Sage Weil <sage@inktank.com>
12 years agoqa: multiple_rsync.sh: more output
Sage Weil [Wed, 12 Jun 2013 20:26:03 +0000 (13:26 -0700)]
qa: multiple_rsync.sh: more output

Trying to track down this failure:

2013-06-12T06:11:13.430 INFO:teuthology.task.workunit.client.0.err:+ rsync -auv --exclude local/ /usr/ usr.2
2013-06-12T06:11:13.430 INFO:teuthology.task.workunit.client.0.err:+ tee a
2013-06-12T06:11:13.527 INFO:teuthology.task.workunit.client.0.out:sending incremental file list
2013-06-12T06:11:46.206 INFO:teuthology.task.workunit.client.0.out:
2013-06-12T06:11:46.208 INFO:teuthology.task.workunit.client.0.out:sent 1689627 bytes  received 8302 bytes  50684.45 bytes/sec
2013-06-12T06:11:46.208 INFO:teuthology.task.workunit.client.0.out:total size is 3274130495  speedup is 1928.31
2013-06-12T06:11:46.209 INFO:teuthology.task.workunit.client.0.err:+ wc -l a
2013-06-12T06:11:46.209 INFO:teuthology.task.workunit.client.0.err:+ grep 4
2013-06-12T06:11:46.211 INFO:teuthology.task.workunit:Stopping misc on client.0...

...and am perplexed!

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agov0.64 v0.64
Gary Lowell [Wed, 12 Jun 2013 16:54:06 +0000 (09:54 -0700)]
v0.64

12 years agoceph-fuse: older libfuses don't support FUSE_IOCTL_COMPAT
Dan Mick [Wed, 12 Jun 2013 01:33:08 +0000 (18:33 -0700)]
ceph-fuse: older libfuses don't support FUSE_IOCTL_COMPAT

Signed-off-by: Dan Mick <dan.mick@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
12 years agoceph-create-keys: Make sure directories for admin and bootstrap keys exist
Peter Wienemann [Tue, 11 Jun 2013 19:38:51 +0000 (21:38 +0200)]
ceph-create-keys: Make sure directories for admin and bootstrap keys exist

Signed-off-by: Peter Wienemann <wienemann@physik.uni-bonn.de>
12 years agostore_test: create_collection prior to split
Samuel Just [Tue, 11 Jun 2013 18:24:54 +0000 (11:24 -0700)]
store_test: create_collection prior to split

Fixes: #5310
Signed-off-by: Samuel Just <sam.just@inktank.com>
Reviewed-by: David Zafman <david.zafman@inktank.com>
12 years agomon: adjust trim defaults
Sage Weil [Tue, 11 Jun 2013 23:30:41 +0000 (16:30 -0700)]
mon: adjust trim defaults

User testing has shown that smaller values yield better results; see #4917.
Jim's testing has had good results with even more aggressive trimming, but I
would like to do more validation yet before changing defaults.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agodoc: Reworked the landing page.
John Wilkins [Tue, 11 Jun 2013 22:32:23 +0000 (15:32 -0700)]
doc: Reworked the landing page.

Signed-off-by: John Wilkins <john.wilkins@inktank.com>
12 years agodoc: Added a hostname resolution section for local host execution.
John Wilkins [Tue, 11 Jun 2013 21:46:35 +0000 (14:46 -0700)]
doc: Added a hostname resolution section for local host execution.

Signed-off-by: John Wilkins <john.wilkins@inktank.com>
12 years agodoc: Added some tips and re-organized to simplify the process.
John Wilkins [Tue, 11 Jun 2013 21:46:12 +0000 (14:46 -0700)]
doc: Added some tips and re-organized to simplify the process.

Signed-off-by: John Wilkins <john.wilkins@inktank.com>
12 years agoclient: set issue_seq (not seq) in cap release
Sage Weil [Sun, 9 Jun 2013 00:38:07 +0000 (17:38 -0700)]
client: set issue_seq (not seq) in cap release

We regularly have been observing a stall where the MDS is blocked waiting
for a cap revocation (Ls, in our case) and never gets a reply.  We finally
tracked down the sequence:

 - mds issues cap seq 1 to client
 - mds does revocation (seq 2)
 - client replies
 - much time goes by
 - client trims inode from cache, sends release with seq == 2
 - mds ignores release because its issue_seq is 1
 - mds later tries to revoke other caps
 - client discards message because it doesn't have the inode in cache

The problem is simply that we are using seq instead of issue_seq in the
cap release message.  Note that the other release call site in
encode_inode_release() is correct.  That one is much more commonly
triggered by short tests, as compared to this case where the inode needs to
get pushed out of the client cache.

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>
12 years agodoc: Added some Java S3 API troubleshooting entries.
John Wilkins [Tue, 11 Jun 2013 19:12:46 +0000 (12:12 -0700)]
doc: Added some Java S3 API troubleshooting entries.

Signed-off-by: John Wilkins <john.wilkins@inktank.com>
12 years agodoc: Added install ceph-common instruction.
John Wilkins [Tue, 11 Jun 2013 19:11:51 +0000 (12:11 -0700)]
doc: Added install ceph-common instruction.

Signed-off-by: John Wilkins <john.wilkins@inktank.com>
12 years agodoc: Added install ceph-common instruction.
John Wilkins [Tue, 11 Jun 2013 19:11:26 +0000 (12:11 -0700)]
doc: Added install ceph-common instruction.

Signed-off-by: John Wilkins <john.wilkins@inktank.com>
12 years agodoc: Fixed :term" syntax.
John Wilkins [Tue, 11 Jun 2013 19:10:52 +0000 (12:10 -0700)]
doc: Fixed :term" syntax.

Signed-off-by: John Wilkins <john.wilkins@inktank.com>
12 years agoceph-create-keys: Remove unused caps parameter on bootstrap_key()
Gary Lowell [Tue, 11 Jun 2013 15:25:36 +0000 (08:25 -0700)]
ceph-create-keys:  Remove unused caps parameter on bootstrap_key()

The caps parameter was removed except for one place.

Signed-off-by: Gary Lowell <gary.lowell@inktank.com>
12 years agoMerge branch 'next'
Sage Weil [Tue, 11 Jun 2013 06:22:49 +0000 (23:22 -0700)]
Merge branch 'next'

12 years agoosd: fix con -> session ref change after hb reset
Sage Weil [Tue, 11 Jun 2013 06:21:50 +0000 (23:21 -0700)]
osd: fix con -> session ref change after hb reset

set_priv() expects to be given a reference to own; take one.  This fixes
various crashes after we see a hb connection reset.

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agocommon/admin_socket: fix leak of new m_getdescs_hook
Sage Weil [Sat, 8 Jun 2013 20:56:00 +0000 (13:56 -0700)]
common/admin_socket: fix leak of new m_getdescs_hook

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agocommon/cmdparse: no need to use (and leak to) the heap
Sage Weil [Tue, 11 Jun 2013 00:11:44 +0000 (17:11 -0700)]
common/cmdparse: no need to use (and leak to) the heap

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Dan Mick <dan.mick@inktank.com>
12 years agoCrushWrapper: dump tunables along with crush map
Dan Mick [Tue, 11 Jun 2013 00:25:55 +0000 (17:25 -0700)]
CrushWrapper: dump tunables along with crush map

Signed-off-by: Dan Mick <dan.mick@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
12 years agoceph: --keyring must be passed to parse_argv, which means not argparse
Dan Mick [Tue, 11 Jun 2013 01:08:31 +0000 (18:08 -0700)]
ceph: --keyring must be passed to parse_argv, which means not argparse

If argparse gets its hands on it, it's not available for parse_argv()
and is therefore ignored.

Signed-off-by: Dan Mick <dan.mick@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
12 years agoOSD: create collection in handle_pg_create before _create_lock_pg
Samuel Just [Tue, 11 Jun 2013 00:01:02 +0000 (17:01 -0700)]
OSD: create collection in handle_pg_create before _create_lock_pg

Fixes: #5270
Signed-off-by: Samuel Just <sam.just@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
12 years agoObjecter: fail osd_command if OSD is down
Dan Mick [Mon, 10 Jun 2013 22:03:37 +0000 (15:03 -0700)]
Objecter: fail osd_command if OSD is down

Signed-off-by: Dan Mick <dan.mick@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>