test/rbd_mirror: grab timer lock before calling add_event_after()
add_event_after() expects an externally provided mutex to be held
for the call. This was missed in commit 8965a0f2a6f7 ("rbd-mirror:
synchronize with in-flight stop in ImageReplayer::stop()").
librbd/cache/pwl: remove RBD_FEATURE_DIRTY_CACHE check in DiscardRequest
"m_image_ctx.features &&RBD_FEATURE_DIRTY_CACHE" is obviously wrong
because it would pretty much always be true. However, even if bitwise
AND was used, this check would still be dead because DiscardRequest is
only invoked if RBD_FEATURE_DIRTY_CACHE is enabled:
int invalidate_cache(ImageCtx *ictx) {
{
...
// Delete writeback cache if it is not initialized
if ((!ictx->exclusive_lock ||
!ictx->exclusive_lock->is_lock_owner()) &&
ictx->test_features(RBD_FEATURE_DIRTY_CACHE)) {
C_SaferCond ctx3;
ictx->plugin_registry->discard(&ctx3);
r = ctx3.wait();
}
librbd/cache/pwl: don't crash if cache file removal fails
The non-ec overload will throw fs::filesystem_error on any error
(e.g. EPERM due to unprivileged "rbd persistent-cache invalidate"
being brought up against a privileged workload).
Yin Congmin [Wed, 22 Dec 2021 07:07:11 +0000 (15:07 +0800)]
librbd/cache/pwl: rename persistent cache key
librbd "internal" metadata keys was change to ".rbd" prefix. Change
peristent cache to ".rbd" too.
And the name of persistent cache key is IMAGE_CACHE_STATE. Since
this key is planned to be used outside the pwl directory, it seems
more appropriate to change it to a clear name as PERSISTENT_CACHE_STATE.
librbd/cache/pwl: avoid inconsistencies in ImageCacheState
When empty and/or clean bools are updated in I/O handling code paths,
ImageCacheState becomes inconistent for a short while: e.g. with clean
transitioned to true, dirty_bytes counter could still be positive
because the counters are updated only in periodic_stats(). Move to
updating the counters in update_image_cache_state(Context*) to avoid
this.
update_image_cache_state(Context*) now requires m_lock -- most call
sites already hold it anyway. The only problematic call site was
AbstractWriteLog::shut_down() callback chain: perf_stop() needed to
be moved to the very end since perf counters must be alive now for
update_image_cache_state() to work.
Don't override expect_op_work_queue() in unit tests: completing
context in the same thread now results in a deadlock on m_lock in
all test cases that call AbstractWriteLog::init().
get_json_format() and create_image_cache_state() attempt to get
particular keys which could result in an unhandled std::runtime_error
exception. Conversely, ImageCacheState constructor just swallows that
exception which could leave the newly constructed object incorrectly
initialized. Avoid doing parsing in the constructor and introduce
init_from_config() and init_from_metadata() methods instead.
While at it, move everything out from under "persistent_cache" key.
Also fix init_state_json_write test case which stopped working now
that types are enforced by json_spirit.
Yin Congmin [Tue, 29 Mar 2022 08:59:05 +0000 (16:59 +0800)]
librbd/cache/pwl: add basic metrics to ImageCacheState
Add basic metrics to ImageCacheState and persist them, including
allocated_bytes, cached_bytes, dirty_bytes, free_bytes and hit/miss
info.
Leverage periodic_stats() timer to call update_image_cache_state.
In order to avoid outputting too much debug information, the original
statistics output log level is changed to 5.
Switch to json_spirit for encoding because encode_json encodes bool as
"true"/"false" string.
Remove rbd_persistent_cache_log_periodic_stats option because we need
to always update cache state.
[ idryomov: add cached_bytes and hits_partial; report misses and
miss_bytes instead of respective totals; naming ]
Laura Flores [Thu, 7 Apr 2022 22:20:14 +0000 (22:20 +0000)]
mgr/telemetry: anonymize daemons in telemetry perf_counters
In the telemetry perf channel we collect 'perf_counters' of individual daemons.
The monitors appear with their full name, which includes the host name.
The host name part must be anonymized.
To err on the safe side, I have anonymized all daemons except for osds,
since they are not attached to host names.
Fixes: https://tracker.ceph.com/issues/55229 Signed-off-by: Laura Flores <lflores@redhat.com>
(cherry picked from commit 0fe47b974ccc591c6108eb7a1b26087e62932bce)
Venky Shankar [Tue, 29 Mar 2022 13:18:06 +0000 (09:18 -0400)]
mount.ceph: remove `ms_mode' mount option when switching to old-syntax
... and switch to using v1 addresses (if users haven't specified those
explicitly). kernel versions <5.11 do not understand `ms_mode' mount
option which would result in mount failure.
John Mulligan [Mon, 28 Feb 2022 20:46:44 +0000 (15:46 -0500)]
mgr/nfs: remove redundant check
Remove the extra check of the cluster id from _apply method. As _apply
is a "private" method that should be only called from other private
methods that have already validated the cluster_id. It also removes
a dependency on the orch-requiring func available_clusters.
John Mulligan [Mon, 28 Feb 2022 20:43:47 +0000 (15:43 -0500)]
mgr/nfs: fix unintentional recursion
The `exports` property of the ExportMgr exists to cache the exports
configuration found in the .nfs namespace. Using that property
within the property method is probably not intentional and is probably
only working due to the lucky construction of the _exports dict
immediately after the None check so that the _exports dict is returned
(and is a mutable type).
John Mulligan [Mon, 28 Feb 2022 20:39:24 +0000 (15:39 -0500)]
mgr/nfs: add known_cluster_ids to generalize nfs cluster id fetching
The changes to the nfs module in 8c711afc are working but when I began
writing more test automation I found a few more places in the
export-configuration code path relying on the orchestration module
only. This change generalizes the logic to source nfs clusters from
orchestration when it's enabled but from the .nfs pool when
orchestration is disabled. It then uses that call when loading
the exports cache on the ExportMgr object.
John Mulligan [Wed, 2 Feb 2022 20:58:08 +0000 (15:58 -0500)]
doc/mgr/nfs: document that nfs exports related mgr call requirements
A recent change in the mgr/nfs module should enable the functioning
of export management commands/API calls as long as the rados namespaces
and objects have been already established. Document this fact, noting
that now only the `ceph nfs cluster ...` calls *require* an
orchestration module.
John Mulligan [Sat, 29 Jan 2022 16:23:00 +0000 (11:23 -0500)]
mgr/nfs: support managing exports without orchestration enabled
This change allows the `ceph nfs export ...` commands to function
without the entire mgr/nfs subsystem requiring orchestration to be
enabled. When there's no orchestration available, the code falls back
to examining the namespaces in the ".nfs" rados pool to determine what
cluster_id values are valid.
This change does not add support for creating the rados objects and
namespace needed to manage a nfs cluster. As discussed with the
orchestration group on 2022-01-22, rook does not need the mgr module to
establish the namespace. So, for now, we'll defer the work needed to
create the namespace/objects when orchestration is disabled.
Fixes: https://tracker.ceph.com/issues/54043 Signed-off-by: John Mulligan <jmulligan@redhat.com>
(cherry picked from commit 8c711afc4ab898942a2569b619eb8379ee02ffba)
John Mulligan [Thu, 27 Jan 2022 22:06:02 +0000 (17:06 -0500)]
mgr/nfs: clean up rados object naming code
The naming of rados objects used to store the nfs config was spread
all over the code, including inline f-strings, not-static methods,
etc.
This change unifies the naming by putting constant string prefixes
and name generating functions into the utils.py file.
John Mulligan [Thu, 27 Jan 2022 21:06:50 +0000 (16:06 -0500)]
mgr/nfs: make _check_rados_notify a function
This was previously a staticmethod. This static method was only used by
NFSRados object. Staticmethods are nearly always better implemented as
functions, which is done so here.
John Mulligan [Thu, 27 Jan 2022 20:50:13 +0000 (15:50 -0500)]
mgr/nfs: limit dependency of NFSRados object
Previously, the NFSRados object accepted the "Module" as the
first argument but only used the rados attribute (type rados.Rados).
It's better to limit the scope of types when reasonably possible
so we can see what the true dependencies are. So we restrict
NFSRados to accepting a rados.Rados as the argument.
Ernesto Puerta [Fri, 25 Mar 2022 15:26:48 +0000 (16:26 +0100)]
mgr/dashboard: fix api test issue with pip
Fix
```
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
apache-libcloud 3.5.0 requires requests>=2.26.0, but you have requests 2.25.1 which is incompatible.
Successfully installed CherryPy-13.1.0 PyJWT-2.0.1 Routes-2.4.1 bcrypt-3.1.4 ceph-1.0.0 chardet-4.0.0 cheroot-8.6.0 idna-2.10 jaraco.functools-3.5.0 more-itertools-4.1.0 natsort-8.1.0 portend-3.1.0 pyopenssl-22.0.0 pytz-2022.1 repoze.lru-0.7 requests-2.25.1 tempora-5.0.1
```
Conflicts:
src/cephadm/cephadm
src/pybind/mgr/cephadm/module.py:
- Accept quincy changes and bring only updates in the Grafana,
Prometheus, Alertmanager and Node Exporter versions
Casey Bodley [Tue, 29 Mar 2022 21:46:04 +0000 (17:46 -0400)]
cmake: WITH_SYSTEM_UTF8PROC defaults to OFF
change the default value of WITH_SYSTEM_UTF8PROC from ON to OFF, so that
centos/rhel users can build with the default cmake configuration. no other
WITH_SYSTEM_* variable in ceph defaults to ON, so this is consistent
with other bundled libraries like boost and rocksdb
unfortunately, this also means that users that do have system packages
must opt-in to using them with -DWITH_SYSTEM_UTF8PROC=ON
both deb and rpm builds dependended on the previous default value, so
their logic was negated to match the new default
os/BlueStore: NCB fix for SimpleBitmap boundary check
The boundary check in SimpleBitmap is off by one causing an assert to trigger
Also fixed a bug when asking for the next clear_extent on a unaligned map when the last bits in the map were set.
Adding unit-tests Fixes: https://tracker.ceph.com/issues/55145 Signed-off-by: Gabriel BenHanokh <gbenhano@redhat.com>
(cherry picked from commit 7dfa20863090d5eb58c798b6903386dcce6a52f8)
The trailing '3' was missed in one instance, ceph-mgr-cephadm, leading to:
Some packages could not be installed. This may mean that you have
requested an impossible situation or if you are using the unstable
distribution that some required packages have not yet been created
or been moved out of Incoming.
The following information may help to resolve the situation:
The following packages have unmet dependencies:
ceph-mgr-cephadm : Depends: python3-cherrypy but it is not installable
osd/osd_types: Increasing decode version of scrub_duration in pg stats
All new fields added to pg stats after quincy RC need to have the decode field bumped up to avoid decoding errors during an upgrade from quincy RC to the quincy stable version
Dimitri Savineau [Mon, 28 Mar 2022 14:50:51 +0000 (10:50 -0400)]
cephadm: set quincy as stable release
Quincy isn't master anymore so we don't need the DEFAULT_IMAGE_IS_MASTER
variable set to true (which produces a warning message).
This also sets the LATEST_STABLE_RELEASE variable to quincy to match the
DEFAULT_IMAGE_RELEASE variable.
Tatjana Dehler [Wed, 9 Mar 2022 15:23:49 +0000 (16:23 +0100)]
mgr/cephadm: try to get FQDN for inventory address
Fixes: https://tracker.ceph.com/issues/54502 Signed-off-by: Tatjana Dehler <tdehler@suse.com>
(cherry picked from commit 4f14993b1667fff309cd9cd6f9dad638a5a7e502)
Conflicts:
src/pybind/mgr/cephadm/services/monitoring.py
Fixed conflict because https://github.com/ceph/ceph/pull/44751 has not been
backported to quincy (yet).
Tatjana Dehler [Wed, 9 Mar 2022 15:10:49 +0000 (16:10 +0100)]
mgr/cephadm: unify way to get the host address
There are two different ways to get the host address. From the
inventory of the mgr object directly or via the `_inventory_get_addr`
method of `CephadmService`. Update the code in order to use the
`_inventory_get_addr` method only.
qa/standalone: Fix test_activate_osd() test in ceph-helpers.sh
Modify test_activate_osd() to get the type of scheduler in use and then
verify the value of osd_max_backfills. This is because mclock scheduler
overrides this option to 1000 upon OSD initialization.
The test earlier used to pass because the OSD daemon was killed but not
marked down and upon being brought up, the wait for OSD up check was
passing quickly. But the OSD still didn't have the latest config values.
But now upon killing the OSD, the osd_fast_shutdown sequence notifies the
mon (see PR: https://github.com/ceph/ceph/pull/44807) and is marked down
and dead. Upon bringing it up, the wait for OSD up check takes a longer
time and this is sufficient for the config values to be updated. This
results in the correct values being read from the config 'Values' map.
Satoru Takeuchi [Thu, 18 Nov 2021 20:48:18 +0000 (20:48 +0000)]
osd: make osd_fast_shutdown_notify_mon option true by default
osd_fast_shutdown_notify_mon option is false by default. So users suffer
from error log flood, slow ops, and the long I/O timeouts on voluntary OS
shutdown before they are aware of the existence of this option. Let's
make this option true by default.
projects / ceph.git / log