This is shown to corrupt otherwise healthy rocksdb databases. Rename to
make it clear that it is generally not safe to run and shoud only be used
as a last resort.
Conflicts:
PendingReleaseNotes: drop this change as "repair" command did
not exist in luminous before this change.
qa/workunits/cephtool/test_kvstore_tool.sh: drop this change,
as this test was not added before this change.
src/tools/ceph_kvstore_tool.cc: trivial resolution.
Kefu Chai [Thu, 15 Nov 2018 05:56:19 +0000 (13:56 +0800)]
tools/ceph_kvstore_tool: do not open rocksdb when repairing it
before this change, the `need_open_db` parameter is passed to the
constructor of BlueStore as `min_alloc_size`. and rocksdb will fail to
repair because Repairer::Run() also tries to acquire the db lock, and it
will fail to do so if the lock file is already acquired by
BlueStore::_mount().
Conflicts:
src/librbd/api/Mirror.cc
- C_ImageGetInfo ctor takes only two arguments in nautilus
- nautilus does not have LambdaContext as a class; use FunctionContext
instead
Jan Fajerski [Wed, 13 Nov 2019 09:13:01 +0000 (10:13 +0100)]
ceph-volume: assume msgrV1 for all branches containing mimic
With nautilus and newer OSDs listen on v1 ports and v2 ports. Assume
that if mimic (or luminous) occur in the branch name, the OSDs are
running msgrv1 only.
Fixes: https://tracker.ceph.com/issues/42791 Signed-off-by: Jan Fajerski <jfajerski@suse.com>
(cherry picked from commit b8754919df61b118200e210e0bfc8d6df0261dfd)
This was broken by def50d04796, and implicitly fixed during
refactoring in the master (octopus) by adf1486e46c, hence it is a
direct commit to nautilus branch.
Conflicts:
src/tools/rbd_mirror/Mirror.cc (std::lock_guard vs Mutex::Locker, ceph_abort_msgf does not exist)
src/tools/rbd_mirror/PoolReplayer.cc (std::lock_guard vs Mutex::Locker, PoolReplayer is not a template)
Conflicts:
src/test/cli/osdmaptool/help.t (some options not present)
src/tools/osdmaptool.cc (ceph_assert is assert here)
src/test/cli/osdmaptool/missing-argument.t (usage included here)
Igor Fedotov [Thu, 16 Aug 2018 11:51:06 +0000 (14:51 +0300)]
os/bluestore: fix assertion in StupidAllocator::get_fragmentation
One might face an assertion (assert(intervals <= max_intervals))
in StupidAllocator::get_fragmentation method for clusters created
by early Luminous releases and before. The root cause is that block
volume size wasn't aligned with min_alloc_size and hence we missed
that last fraction interval during max_interval calculation.
Fixes: https://tracker.ceph.com/issues/43297
Note: This was a clean cherry-pick from master, but p2roundup was
introduced since mimic release, use P2ROUNDUP instead
Ali Maredia [Mon, 25 Nov 2019 02:30:03 +0000 (21:30 -0500)]
luminous: update s3-test download code for s3-test tasks
- Ensure the download code for all tasks running
s3-tests is consistent.
- Simplify download code to only use the config
variable 'force-branch' for the branch being
cloned.
- Make ceph-luminous the force-branch for all
suites using s3-tests.
- Add force-branch to suites running s3readwrite
& s3roundtrip tasks
osd/MissingLoc.cc: do not rely on missing_loc_sources only
In 624ade487ea4aeaf988cc1767e0b293f76addd5b, we relied on missing_loc_sources
to check for strays and remove an OSD from missing_loc. However, it is
possible that missing_loc_sources is empty while there are still OSDs
present in missing_loc. Since the aim is to just remove a stray OSD from
missing_loc, we do not need to rely on missing_loc_sources. We still
clean missing_loc_sources if any stray is present in it.
xie xingguo [Sat, 31 Aug 2019 02:17:57 +0000 (10:17 +0800)]
osd/PG: fix _finish_recovery vs repair race
On detecting a corrupted object, primary may automatically
repair that object by leveraging the existing recovery procedure,
which turned out to be racy with a previous unfinished _finish_recovery
callback - the problem would then be that _finish_recovery might
continue to purge some strays that we still want to pull data from.
Fix by re-checking if there are any newly added missing objects when
executing _finish_recovery.
Note that before https://github.com/ceph/ceph/pull/29756 we might
instead have to call needs_recovery to catch the race condition
since we did not evict pg from clean state when triggering an auto-repair..
Conflicts:
src/osd/PG.cc
- adjusted if conditional for luminous
- did not add the comment nor state_clear(PG_STATE_REPAIR);. Those lines were
moved but don't exist in luminous.
Neha Ojha [Sat, 31 Aug 2019 01:15:58 +0000 (18:15 -0700)]
osd/MissingLoc, PeeringState: remove osd from missing loc in purge_strays()
We should always try to keep osds in missing_loc consistent with peer_missing
and peer_info. When we remove an osd from peer_missing and peer_info, we
should also remove it from missing_loc during purging strays.
Conflicts:
src/osd/MissingLoc.cc
src/osd/MissingLoc.h
src/osd/PeeringState.cc
- these files do not exist in luminous; made the changes manually to
src/osd/PG.cc and src/osd/PG.h
- ldout(cct, ...) -> ldout(pg->cct, ...)
We should have done this while cherry-picking from master, but we
didn't. And here we are now. It's simpler to apply this one-off patch
than going back to the cherry-picking maze to adjust this one thing.
Conflicts:
src/pybind/mgr/telemetry/module.py
Due to missing context resulting from missing patches.
PendingReleaseNotes
Dropped to prevent conflicts in the future
Note:
This commit was heavily modified. We wanted to provide the number of
ipv4 and ipv6 monitors in the report, so we rewrote that part so we
can report on it; but we had to drop everything else (msgr1 and
msgr2), as well as 'min_mon_release'. Those do not exist in
luminous. In the end, the commit message itself is misleading, but
we are somehow (*shrug*) opting for leaving the commit as the original.
Additionally, we removed PendingReleaseNotes changes to prevent
conflicts in the future.
Matt Benjamin [Wed, 11 Dec 2019 22:52:57 +0000 (17:52 -0500)]
rgw: lc: continue past get_obj_state() failure
The get_obj_state() failure in particular could indicate a race with
an object being deleted, so likely is non-fatal. By returning, lifecycle
processing for the current bi-shard would not resume until re-scheduled,
likely in 24 hours.
Fixes: https://tracker.ceph.com/issues/43269 Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>
Conflicts:
src/osd/OSD.cc
src/osd/OSD.h
src/osd/PrimaryLogPG.h
- no OSD::get_osd_delete_sleep() in luminous, no OSD::get_recovery_max_active()
in luminous
- use cct->_conf->get_val instead of cct->_conf.get_val
Sage Weil [Mon, 28 Jan 2019 20:58:26 +0000 (14:58 -0600)]
osd: refuse to start if release > recorded min_osd_release + 2
If we try to start up the objectstore, we may make writeable changes to
(say) rocksdb that are not backwards compatible. This happens, for
example, if you start a mimic osd. Even if the compatset checks fail,
rocksdb may have written something that is not backwards compatible.
Sage Weil [Mon, 28 Jan 2019 21:05:53 +0000 (15:05 -0600)]
osd: record require_osd_release in objectstore meta
Record the require_osd_release value from the OSDMap in the 'meta' portion
of the osd's metadata that can be accessed without actually mounting the
OSD. This will be useful as a safety gate to prevent you from mounting
an osd thet is too new that may make incompatible changes to the store.
Conflicts:
doc/rados/operations/health-checks.rst
We don't have the crash module, hence neither its docs.
src/pybind/mgr/telemetry/module.py
Issues due to context
Conflicts:
src/pybind/mgr/telemetry/module.py
Slight conflicts due to past cherry-picks (or lack thereof)
Using set_config() instead of set_module_option()
Conflicts:
src/pybind/mgr/telemetry/module.py
Due to lack of 'crash' and 'devicehealth' modules, and a bit
on how we keep options (self.config[] vs class attributes)
Conflicts:
src/pybind/mgr/telemetry/module.py
Don't backport code related to the 'crash' module, and adjust
how we read option variables (luminous goes through a config
map, instead of master's that goes through class attributes)
Sage Weil [Mon, 29 Apr 2019 19:32:44 +0000 (14:32 -0500)]
mgr/telemetry: use cluster-provided timestamp unmolested
The cluster stamp is now ISO 8601; just use that.
(The isoformat() puts a : in +hh:mm the timezone offset, which is slightly
different than what Ceph does; just pass Ceph's value through for
consistency.)
Conflicts:
src/pybind/mgr/telemetry/module.py
Past commit in master had introduced field types and a
'minimum' value for the interval. We concluded that the field
types commit does not affect the telemetry module in a
significant way to force us to backport it, and the minimum
value commit is introduced for the benefit of the dashboard
(which, in luminous, does not have control over telemetry)
Conflicts:
src/pybind/mgr/telemetry/module.py
mostly due to store_get/store_set not existing in luminous,
and we relying instead on config_get/config_set.
Conflicts:
src/pybind/mgr/telemetry/module.py
master no longer has 'telemetry selftest' due to some other
major changes that we did not backport, as they would require
too many changes that were not, in an obvious manner, relevant
for us.
Conflicts:
src/pybind/mgr/telemetry/module.py
We don't have some other scaffolding that exists on master,
and we are not cherry-picking it because it changes
significantly the module's code in a way that is not a clear
advantage for the telemetry module (in 'luminous' context)
projects / ceph.git / log