John Mulligan [Thu, 9 Jul 2026 20:12:11 +0000 (16:12 -0400)]
pybind/mgr: disable type checking in diskprediction_local
Disable type checking in the __preprocess function of
diskprediction_local. This function is already collecting a lot of
`type: ignore` comments and adding more is getting difficult due to
the structure of this function. This time we were hitting errors
like:
diskprediction_local/predictor.py:160: error: Value of type "NDArray?[Any]" is not indexable [index]
diskprediction_local/predictor.py:161: error: "NDArray?[Any]" has no attribute "shape" [attr-defined]
on python 3.12.
Fixes: https://tracker.ceph.com/issues/78117 Signed-off-by: John Mulligan <jmulligan@redhat.com>
Super User [Wed, 8 Jul 2026 12:05:16 +0000 (17:35 +0530)]
doc/rbd: clarify mirror resync snapshot behavior
Add documentation explaining that rbd mirror resync only copies
image state and data up to the last mirror-snapshot for
snapshot-based mirroring. Document the requirement to create
a new mirror-snapshot on current primary so that new changes
(including resize operation) gets reflected on secondary after
resync.
crimson/osd: guard recovery clone overlap with try_lock_for_read
Crimson recovery could use a neighbor clone as a clone_subset source
during osd_recover_clone_overlap without checking whether that clone
was safe to read. Classic OSD takes a read lock on the neighbor clone
via try_lock_for_read() before using it.
Under snap trim + recovery stress, recovery could clone_range from a
clone that snap trim was modifying, producing wrong object content
(ceph_test_rados: incorrect buffer at pos 1462272).
calc_clone_subsets() and calc_head_subsets() now return clone overlap
candidates in preference order only. ReplicatedRecoveryBackend locks
the first usable neighbor clone from each list and holds those read
locks until push/pull recovery completes. RecoveryCloneLockManager
releases locks in its destructor so move-assign cannot leak a held
lock and block snap trim.
src/test/librados: enable FastEC tests with crimson
This commit enables fast EC tests such that they are run for crimson
OSDs. This is done towards an effort to establish reliable fast EC
support in crimson.
These tests where previously enabled as part of another PR but
ultimately were not merged due to insufficient confidence.
mgr/cephadm: don't treat unknown daemon state as error
cephadm ls may transiently report state='unknown' for daemons that are
still starting up. The state mapping in _refresh_host_daemons mapped
'unknown' to DaemonDescriptionStatus.error, which caused
update_failed_daemon_health_check to fire CEPHADM_FAILED_DAEMON for
daemons that were simply mid-startup. The warning clears on the next
refresh once the daemon reaches a real state, but it causes spurious
QA failures across many test suites.
Map 'unknown' to DaemonDescriptionStatus.unknown instead — the enum
value already exists and is handled by to_str() and the dashboard.
get_error_daemons() only checks for .error, so unknown daemons no
longer trigger the health warning.
Ville Ojamo [Wed, 8 Jul 2026 09:59:30 +0000 (16:59 +0700)]
doc: remove unused readthedocs-sphinx-search
Remove readthedocs-sphinx-search from the RTD build environment because
it seems to be unused and unsupported.
RTD has archived the readthedocs-sphinx-search repo since Apr 2025.
The docs website does not seem to show such function, the search box
in the top-left corner requires a search to be submitted and is not
live.
Fixes: https://tracker.ceph.com/issues/72918 Signed-off-by: Ville Ojamo <git2233+ceph@ojamo.eu>
Ville Ojamo [Wed, 8 Jul 2026 05:27:06 +0000 (12:27 +0700)]
doc: fix Sphinx complaints and some rendering errors
Fix 25 criticals, errors, and warnings from Sphinx.
Some of the changes did not change rendering but most fixed unintended
rendering, broken link etc.
Change Mon to Monitor in ceph-secrets.rst.
Signed-off-by: Ville Ojamo <git2233+ceph@ojamo.eu>
Nizamudeen A [Wed, 8 Jul 2026 05:21:50 +0000 (10:51 +0530)]
mgr/dashboard: teardown http requests for feature-toggle
angular 19.2.latest is aggressively tearing down the test env which
forces the active rxjs timer to emit one execution as it collapses which
produces a ghost http in the case of request wrapped in the timer
service. so for now flushing all the requests manually to prevent it
Fixes: https://tracker.ceph.com/issues/77729 Signed-off-by: Nizamudeen A <nia@redhat.com>
qa/cephadm: fix hardcoded 'sshd' unit name in setup_ca_signed_keys
setup_ca_signed_keys() hardcoded `systemctl restart sshd`, which fails
on Debian/Ubuntu where the SSH server unit is named `ssh.service`
rather than `sshd.service`. Try `ssh` first, fall back to `sshd` if
that fails, grouped so the fallback only fires on restart failure and
doesn't mask a failed config write.
qa/suites/rbd/valgrind: pin to centos_9.stream instead of rpm_latest
This used to be the case before commit d4b977afdc59 ("qa/distros:
rename centos_latest.yaml to rpm_latest.yaml") and subsequent changes
to enable Rocky 10. When paired with valgrind, python_api_tests* jobs
fail persistently on Ubuntu and Rocky, see [1]. Switch back until that
is resolved (or for as long as centos_9.stream remains in the mix).
Matan Breizman [Mon, 8 Jun 2026 12:45:20 +0000 (15:45 +0300)]
crimson/osd: fix CyanStore boot with crimson OSD
- Remove dead cyan_store.h includes from pg.cc and shard_services.cc.
- Fix CyanStore::mkfs to write type="cyanstore" instead of "memstore"
- Handle missing device class in OSD::start: stores that return an empty
device class (e.g CyanStore) now fall back to is_rotational=false with
a warning instead of throwing
Before optimizing the onode lookup paths we should measure the existing
cost if each possible optimization.
The per-node search is already binary search(see search_result_bs_t
binary_search). Possible costs are:
- per-level node traversal
- STAGE_STRING ns/oid memcmp (rbd has long shared oid prefixes)
Adding the following stats to be exposed as seastar-metrics (not logs).
Keep the new sampling per-comparison overhead out unless needed for
measurment runs.
Candidate optimization leads from the numbers:
* cross-node string/prefix dedup (marked as TODOs in stage.h) this could
result in cheaper compares for shared (rbd) prefixes.
* shrink onode_layout_t - marked as onode.h TODO.
stop inlining oi and ss to get more objects per leaf.
mgr/cephadm: pass RGW zonegroup hostnames as custom SANs
Fix RGW certificate generation so `zonegroup_hostnames` and wildcard
SANs are passed to `get_certificates()` using the correct keyword
argument. Previously, RGW built the expected SAN list from
`zonegroup_hostnames` and `wildcard_enabled`, but passed it
positionally so it was interpreted as ips list.
Also add a unit test to cover the RGW `zonegroup_hostnames` +
`wildcard_enabled` path and ensure SANs are passed as `custom_sans`.
Fixed other calls to enforce the usage of keywords for arguments just
to avoid similar issues in the future.
Sun Yuechi [Mon, 6 Jul 2026 08:06:09 +0000 (01:06 -0700)]
test/subprocess: add missing space in SubshellKilled subcommand
9abf2388f24 rewrote "sh -c cat" as SHELL "-c cat", which concatenates
to "/bin/sh-c cat", so the subshell tries to execute a non-existent
"/bin/sh-c". The test still passed because the SIGTERM sent by kill()
usually wins the race against the shell startup; under CI load the
shell occasionally prints
/bin/sh: line 1: /bin/sh-c: No such file or directory
into the stderr pipe first, ASSERT_TRUE(buf.empty()) fails, and the
skipped join() turns this into a SIGABRT in ~SubProcess().
crimson/osd: fix backfill deadlock when only drop operations remain
BackfillState::Waiting requires an ObjectPushed event to transition
back to Enqueuing, but drop (remove) operations are fire-and-forget
and never generate ObjectPushed. When a backfill round enqueues only
drops and no pushes, the FSM permanently stalls in the Waiting state,
holding the backfill reservation and blocking all other PGs from
backfilling to that OSD.
Fix by adding a pending-pushes counter to ProgressTracker to
distinguish "only drops remain" from "pushes still in-flight". When
all pushes are complete and the namespace is exhausted, drain remaining
drop entries and advance last_backfill to MAX so that
OP_BACKFILL_FINISH triggers the normal completion path via
RequestDone.
The classic OSD avoids this by design: drops are never tracked as
in-flight operations in recover_backfill() and are drained inline.
Fixes: https://tracker.ceph.com/issues/77936 Assisted-by: Claude <claude.ai> Signed-off-by: Ronen Friedman <rfriedma@redhat.com>
monitoring/ceph-mixin: scope test queries per dashboard
get_dashboards_data() keeps every query in a single map keyed by
"<panel title>-<legendFormat>", but that id is not unique: several
dashboards have a panel titled "IOPS", "OSDs" or "Throughput". The
dashboard read last overwrites the earlier entry, so one query shadows
another and never gets tested. glob() walks the files in filesystem
order, so which query survives, and whether the test passes, depends on
readdir.
run-tox-promql-query-test hit this in "Test IOPS Read"
(ceph-cluster.feature): the scenario feeds ceph_osd_op_r but the id
resolved to a CephFS pool panel.
pool_id is "UNSET VARIABLE" because the scenario does not set $mdatapool,
and no ceph_pool_rd series were fed, so the query returns nil. It only
fails when readdir returns cephfsdashboard after ceph-cluster-advanced,
so the earlier change that walked nested rows and made the winner
deterministic did not fix it; it fixed which query wins, not that the
wrong one can win.
Key the queries per dashboard (data['queries'][<dashboard>][<id>]) and
have each .feature name its dashboard in a Background step. The lookup is
confined to that dashboard, so a title/legend used elsewhere no longer
shadows it. A duplicate within one dashboard is still an error, unless it
is in a collapsed row.
ceph-cluster.feature covers ceph-cluster-advanced; the rest map to one
dashboard each.
projects / ceph.git / log