Rishabh Dave [Fri, 8 May 2020 17:58:27 +0000 (23:28 +0530)]
test_cephfs_shell: use StringIO instead of BytesIO
Code in qa/ uses both StringIO and BytesIO. Let's use StringIO
exclusively (unless necessary) for uniformity. The reason for using
StringIO over BytesIO is that tests mostly need stdout as string than
as bytes and StringIO is used more frequently used in qa/ code at this
point.
Tiago Melo [Fri, 22 May 2020 15:41:58 +0000 (15:41 +0000)]
mgr/dashboard: Migrate Tabs from ngx-bootstrap to ng-bootstrap
Using ng-bootstrap for Tabs will allow us to easily implement some new features
like only loading 1 tab at a time (already implemented here) and
saving/restoring last opened tab.
Modified the table component to use a clone of the columns list.
Making changes directly to columns var was causing problem when the table was
loaded a second time.
Jason Dillaman [Thu, 28 May 2020 21:59:39 +0000 (17:59 -0400)]
librbd: restore missing flush on write-block logic
When creating the new image dispatch layer, the original flush
upon write-block was dropped. This is resulting in some random race
conditions where object IO is still in-flight when the write-block
indicates it's complete.
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
Jason Dillaman [Wed, 27 May 2020 23:14:14 +0000 (19:14 -0400)]
librbd: exclusive lock image dispatch should not wait on IO when setting lock
IO from later dispatch layers might have caused the need to acquire the lock
(i.e. like an image refresh). In that case, the IO will be blocked waiting for
the exclusive lock to be acquired -- but will deadlock waiting for the IO
to flush.
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
Jason Dillaman [Wed, 3 Jun 2020 13:40:32 +0000 (09:40 -0400)]
rbd-mirror: track journal replay flush requests to prevent race
If a journal replay flush is in-progress when the ImageReplayer is stopped,
it can race and result in an assertion failure due to two attempted shutdowns
of the same journal replay state machine.
Fixes: https://tracker.ceph.com/issues/45409 Signed-off-by: Jason Dillaman <dillaman@redhat.com>
Kefu Chai [Wed, 3 Jun 2020 01:39:26 +0000 (09:39 +0800)]
qa/tasks/vstart_runner: do not teardown test_path if "create-cluster-only"
otherwise we could be removing a "None" directory when tearing down the cluster,
and have following failure:
Exception ignored in: <bound method LocalContext.__del__ of <__main__.LocalContext object at 0x7f99fd4a6cc0>>
Traceback (most recent call last):
File "../qa/tasks/vstart_runner.py", line 1189, in __del__
shutil.rmtree(self.teuthology_config['test_path'])
File "/tmp/tmp.mmM2ugspuR/venv/lib/python3.6/shutil.py", line 477, in rmtree
onerror(os.lstat, path, sys.exc_info())
File "/tmp/tmp.mmM2ugspuR/venv/lib/python3.6/shutil.py", line 475, in rmtree
orig_st = os.lstat(path)
TypeError: lstat: path should be string, bytes or os.PathLike, not NoneType
* refs/pull/34719/head:
ceph-fuse: compatible with libfuse3.5 or higher
cmake: to get the header and library from specified path
libfuse: check the libfuse version from the pkconfig/fuse{3}.pc file
Reviewed-by: Zheng Yan <zyan@redhat.com> Reviewed-by: Kefu Chai <kchai@redhat.com>
Samuel Just [Tue, 12 May 2020 04:02:07 +0000 (21:02 -0700)]
crimson: distinguish record and block relative paddrs
Blocks get read independently of the surrounding record,
so paddr's embedded directly in a block need to refer
to other blocks within the same record by a block_relative
addr relative to the block's own offset. By contrast,
deltas to existing blocks need to use record_relative
addrs relative to the first block of the record.
This patch distinguishes the two kinds of relative paddr
(mainly for debugging purposes) and adapts cache, journal,
etc to use the appropriate types.
* refs/pull/26004/head:
mds: forward mds metrics to ceph manager w/ quering interfaces
mds: track per session client metrics
mds: record metrics from all MDSs in MDS rank 0
mds: non-rank based interface for sending message to an mds
mds: inter-mds ping-pong message and type
mgr: introduce query/report types for ceph metadata server
mds: new intra-mds message type for forwarding aggregated metrics
client: new message type for providing client side metrics
Reviewed-by: Xiubo Li <xiubli@redhat.com> Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
* refs/pull/34838/head:
vstart_runner: don't use namespaces by default
qa/cephfs: run nsenter commands with superuser privileges
qa/cephfs: look for mountpoint in cmdline file
Reviewed-by: Xiubo Li <xiubli@redhat.com> Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
This follows b162541ac21e965a304ee6ffe604c43f22fa96c4.
The balancer was turned on by default in d4fbaf7, as a result of which we might see
PG_AVAILABILITY health warnings when pg-upmap-items are applied.
Kefu Chai [Sat, 30 May 2020 04:51:14 +0000 (12:51 +0800)]
rgw/reshard: use defined variable
use the defined reference for more concise code, this silences the
warning like:
```
../src/rgw/rgw_reshard.cc:530:15: warning: unused variable ‘bucket’ [-Wunused-variable]
530 | rgw_bucket& bucket = bucket_info.bucket;
| ^~~~~~
```
also move `ret` close to where it is used for the first time.
Jason Dillaman [Mon, 1 Jun 2020 18:09:34 +0000 (14:09 -0400)]
rbd-mirror: don't hold (stale) copy of local image journal pointer
The exclusive-lock manages its life cycle and can close the journal
at any point. This can result in rbd-mirror deferencing a freed pointer
or a journal state machine that is in an unexpected state.
Fixes: https://tracker.ceph.com/issues/45803 Signed-off-by: Jason Dillaman <dillaman@redhat.com>
* refs/pull/34782/head:
qa/tasks/cephfs/mount.py: remove netns name parsing in mountpoint setter
qa/tasks/vstart_runner.py: add kwargs parameter to ignore the ones it does not understand
Reviewed-by: Rishabh Dave <ridave@redhat.com> Reviewed-by: Jeff Layton <jlayton@redhat.com> Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
The goal is to never implicitly ignore errors that the function can
return, particularly the failure on pidfile locking due to the file
being hold by another instance. This problem happened recently in
crimson-osd.
Lenz Grimmer [Tue, 2 Jun 2020 13:11:20 +0000 (15:11 +0200)]
Merge pull request #35249 from rhcs-dashboard/wip-45705-master
mgr/dashboard: add API team to CODEOWNERS
Reviewed-by: Alfonso Martínez <almartin@redhat.com> Reviewed-by: Laura Paduano <lpaduano@suse.com> Reviewed-by: Stephan Müller <smueller@suse.com> Reviewed-by: Tatjana Dehler <tdehler@suse.com> Reviewed-by: Volker Theile <vtheile@suse.com>
Rishabh Dave [Wed, 29 Apr 2020 18:10:16 +0000 (23:40 +0530)]
qa/cephfs: run nsenter commands with superuser privileges
And add a method that sets self.fuse_daemon.subproc.pid to the PID of
the process that doesn't have sudo in its arguments. For example, when
"sudo ceph-fuse /mnt/cephfs" is run on the shell, it launches process
with arguments "ceph-fuse /mnt/cephfs". The added method gets PID of
latter/child process and sets that as the fuse daemon's PID. Not doing
so kills the former/parent process but the not the child process.
Also, since we are around cleanup this method a bit.
Fixes: https://tracker.ceph.com/issues/45339 Signed-off-by: Rishabh Dave <ridave@redhat.com>
projects / ceph.git / log