Leonid Usov [Sat, 13 Apr 2024 12:37:03 +0000 (15:37 +0300)]
mds/quiesce: don't take mirrored cap-related locks on the replica
For every mirrored lock, the auth will message the replica to ensure
the replicated lock state. When we take x/rdlock on the auth, it will
ensure the LOCK_LOCK state on the replica, which has the file caps we
want for quiesce: CACHE and BUFFER.
It should be sufficient to only hold the quiesce local lock
on the replica side.
Leonid Usov [Mon, 8 Apr 2024 11:35:02 +0000 (14:35 +0300)]
mds/quiesce: xlock the file to let clients keep their buffered writes
With the quiesce protocol taking a `rdlock` on the file,
it also revokes the `Fb` capability, which the clients can't release
until they are done flushing, and that may take up arbitrarily long,
evidently, more than 10 minutes.
We went for the rdlock to avoid affecting readonly clients,
but given the evidence above we should not optimize for those.
Ideally, we’d like to have a QUIESCE file lock mode where both rd
and buffer are allowed, but as of now it seems like our best
available option is to `xlock` the file which will let the writing
clients keep their buffers for the duration of the quiesce.
We can only afford this change for a `splitauth` config,
i.e. where we drop the lock immediately after all `Fw`s are revoked
Remove information about the installation of the Zabbix module and link
to a discussion of the reasoning behind Ceph's refusal to support
Zabbix.
John Jasen developed a procedure explaining how to install "Zabbix 2".
This commit removes outdated procedures and explains why those
procedures were removed. Immediately following this explanation, the
text includes an explanation of how to install "Zabbix 2".
Adam King [Tue, 9 Apr 2024 16:19:06 +0000 (12:19 -0400)]
mgr/cephadm: make enable_monitor_client configurable for nvmeof
Currently, the mon client work is not merged on main, but our
default nvmeof container will attempt to make use of it by default,
causing it to crash. This makes it configurable and defaults the
behavior to false. That can be changed once the work is actually
present in main.
this may happen when a test fails, and does not cleanup topics
it created. other tests that verify the number of topics may fail
because of that.
all tests that verify number of topics, should delete all topics at the
start of the test.
doc: reorder "releases" entries for reef to fix the diagram
The Gantt diagram currently shows "reef (latest 18.2.0)" instead of
18.2.2. This is because ReleasesGantt expects releases array to be
sorted in reverse order:
for code_name, info in releases.items():
last_release = info['releases'][0]
first_release = info['releases'][-1]
Nizamudeen A [Wed, 27 Mar 2024 16:29:55 +0000 (21:59 +0530)]
install-deps: enable copr ceph/grpc
In dashboard, to generate nvmeof apis in el8 this is needed so that it
can download the python3-grpcio packages.
https://copr.fedorainfracloud.org/coprs/ceph/grpc/
Fixes: https://tracker.ceph.com/issues/65184 Signed-off-by: Nizamudeen A <nia@redhat.com>
Adam King [Wed, 3 Apr 2024 17:11:08 +0000 (13:11 -0400)]
mgr/cephadm: pass daemon's current image when reconfiguring
Important to note here is that a reconfig will rewrite
the config files of the daemon and restart it, but not
rewrite the unit files. This lead to a bug where the
grafana image we used between the quincy and squid release
used different UIDs internally, which caused us to rewrite
the config files as owned by a UID that worked with the
new image but did not work with the image still specified
in the unit.run file. This meant the grafana daemon was down
from a bit after the mgr upgrade until the end
of the upgrade when we redeploy all the monitoring images.
Fixes: https://tracker.ceph.com/issues/65234 Signed-off-by: Adam King <adking@redhat.com>
Adam King [Thu, 4 Apr 2024 19:49:27 +0000 (15:49 -0400)]
qa/cephadm: update images for test_cephadm workunit
This patch doesn't actually fix any tests, but looking
at https://tracker.ceph.com/issues/64208 has made me
realize we haven't been updating the images on here
at all. This patch just removes the octopus and pacific
image variables and puts in ones for reef and squid. I
think this is a reasonable way to try and maintain these,
as we're just keeping around the image for the last two
releases, which matches with what versions could one day
upgrade to what main will become (the "T" release as
of writing this)
rgw/notification: Load bucket attrs before calling publish_reserve.
As part of PR# 55657, publish_reserve would reload bucket to ensure bucket_attrs are loaded. However for lc events, where the bucket attrs were already loaded, the reloading was causing crash but there was no obvious root cause, so to avoid the crashes, remove reloading of bucket in publish_reserve and put the onus on callers to load the bucket before calling publish_reserve.
Adam King [Thu, 4 Apr 2024 18:11:11 +0000 (14:11 -0400)]
mgr/cephadm: make client-keyring deploying ceph.conf optional
There are cases where users would like to manage their own
ceph.conf but still have cephadm deploy the client keyrings,
so this is being added to facilitate that.
Fixes: https://tracker.ceph.com/issues/65335 Signed-off-by: Adam King <adking@redhat.com>
Adam King [Wed, 3 Apr 2024 18:34:08 +0000 (14:34 -0400)]
mgr/cephadm: handle setting required osd release with no OSDs during upgrade
A change to the `ceph osd require-osd-release` command made it so
it fails if no OSDs are up unless --yes-i-really-mean-it is passed.
For real clusters this is likely not an issue, but it can be an
annoyance for trying upgrades on test clusters that may not have
OSDs deployed. This patch is to try and just pass the flag in cases
where we have no OSDs rather than failing the upgrade
rgw: don't map to EIO in rgw_http_error_to_errno()
the http client uses EIO to detect connection errors specifically. if we
map normal http errors to EIO, we incorrectly mark their endpoint as
failed and route requests to other endpoints (if any exist)
default to ERR_INTERNAL_ERROR (500 InternalError) instead
Rishabh Dave [Thu, 22 Feb 2024 12:31:59 +0000 (18:01 +0530)]
cephfs-shell: don't use pkg_resources since its deprecated
Currently, cephfs-shell prints warnings, hangs and aborts when launched.
This occurs because Python module "pkg_resources" has been deprecated.
We use that module only checking version of other Python modules used in
cephfs-shell. Use "Version" from "packaging.version" instead.
Fixes: https://tracker.ceph.com/issues/64538 Signed-off-by: Rishabh Dave <ridave@redhat.com>
rgw: udpate options yaml file so LDAP uri isn't an invalid example
LDAP tries to bind the URI configuration option when RGW starts. The
default value is an example used to show the form of the URI and is
not itself valid. The default value is used, unless overrideen, and
can cause delays in start-up in some situations. The example is now
provided in the description and the default is the empty string.
Signed-off-by: J. Eric Ivancich <ivancich@redhat.com>
Log at lower level at log level 5 while the
mds is stopping to track down issues like mds
becoming laggy while stopping. It was being
logged at level 7 before.
With the recent code added to handle connection errors
(commit#e200499bb3c5703862b92a4d7fb534d98601f1bf), RGWRESTStreamS3PutObj
initialization could fail at times if there were any failed requests to the
cloud endpoint within CONN_STATUS_EXPIRE_SECS period.
This fix is to handle such errors and abort the transition/sync
requests which can be retried later by LC/Sync worker threads.
projects / ceph.git / log