crimson/os/seastore/omap_manager: predict append (not overwrite) on initial_pending
An initial_pending LogNode always append_kv()s and never overwrites.
expect_overflow() was still called with can_ow=true whenever the tail entry is
_fastinfo. It then used only the overwrite gap while the append
itself consumed a full entryi - so the node rolled over far too late and
used_space() overran capacity().
This would result in `assert(capacity() >= used_space()`.
Gate can_ow on !is_initial_pending() so the append is predicted with the full
entry size and rolls over in time.
crimson/os/seastore/omap_manager: calculate the overwrite growth (gap)
The previous implementation computed ow_gap_from_last_entry() but never used
the result.
Use it to check only the additional space required when the tail
entry grows, which matches overwrite logic.
Note, this is a cleanup - behavior remains similar.
crimson/seastore: batch only txns with matching data_features
Transaction::append() asserts both transactions share data_features.
The per-collection batcher merged queued client txns unconditionally,
so a data_features mismatch would be asserted in build_next_batch().
Use diffrent data_features as a grouping boundary.
crimson/seastore: run pg-log trim txns solo, off the batched path
Per-collection batching merges many client txns into one seastore txn.
Batched OP_OMAP_RMKEYS is not stable yet, possibly due to an exposed
internal bug that is seen with the new behavior.
Avoid batching OMAP_RMKEYS for now.
crimson/seastore: fix LogNode overflow from unsigned underflow in expect_overflow
The can_ow branch of LogManager's expect_overflow computed:
```
remain = capacity() - get_last_pos() - reserved_len
```
all uint32 operands, once the log leaf node is near full
(get_last_pos() + reserved_len >= capacity()) this subtraction underflows.
The node never rolls to a new leaf, instead it overflows the
16 KB block at commit-time delta replay.
Captured at the replay overflow point (per-collection batching, 4K randwrite):
log_node.h : In function 'void LogKVNodeLayout::set_last_pos(uint32_t)',
ceph_assert(pos <= LOG_NODE_BLOCK_SIZE)
```
Here reserved_len (3098) is accurate: ~13543 committed + 3098 pending =
16641 = pos, i.e. the node was already ~13.5 KB full and the predictor
should have rolled over. Instead capacity - get_last_pos - reserved_len
went negative (16328 - ~13273 - 3098 < 0), wrapped in uint32, and the
check passed; pos reached 16641 > 16384 and set_last_pos asserted.
This was exposed by the per-coll batching where many overwrites append in
a single txn.
Previously, the per-collection ordering_lock provided both ordering and
single-holder exclusion, causing same-collection transactions to be serialized
one at a time.
Replace that with a per-collection queue and a single dispatch loop. When a
collection has a batch in flight, newly arriving transactions are queued.
The dispatch loop later drains the queue and merges transactions into one
SeaStore transaction (batch). Resulting in a one build, one ool_write and one journal record for the entire batch.
Only one batch is in flight per collection, so same-collection concurrency is
still prevented and ordering is preserved by the queue.
Callbacks are unchanged: FuturizedStore drains each caller's on_commit before
enqueueing, so only ops are merged. Each caller's promise is fulfilled when the
merged batch commits.
David Galloway [Fri, 17 Jul 2026 18:34:44 +0000 (14:34 -0400)]
container: default FROM_IMAGE to Rocky Linux 10
Since tentacle, the preferred base image for ceph containers has been
Rocky Linux 10, and the CI tag-naming logic in build.sh already assumes
rockylinux-10 is the default fromtag for every branch except reef and
squid. The actual build default was never flipped, though: anything
that ran build.sh without FROM_IMAGE set (e.g. the release container
job in ceph-build) still got a CentOS Stream 9 base.
Flip the Containerfile ARG and the build.sh fallback to
docker.io/rockylinux/rockylinux:10 so umbrella and later build from
Rocky 10 by default. Builds that want a different base can still pass
FROM_IMAGE explicitly, as the CI pipeline does.
Signed-off-by: David Galloway <david.galloway@ibm.com>
Aggressive Link-Time Optimization (LTO) and stricter linker defaults
in newer toolchains (like GCC 13 on Ubuntu 24.04) can strip or
localize external global variables if their visibility is not
explicitly defined. Tagging g_lockdep with default visibility ensures
it remains accessible in the dynamic symbol table for external
binaries like monmaptool.
Assisted-by: Gemini Fixes: https://tracker.ceph.com/issues/78321 Signed-off-by: Patrick Donnelly <pdonnell@ibm.com>
crimson/os/seastore/transaction: new conflict/no_conflict trans
synchronization
Use seastar::shared_mutex to synchronize transactions when committing
no_conflict ones.
Specifically, this commit achieves conflict/no_conflict trans
synchronization by:
1. The no_conflict trans acquires locks on all the stable extents which
it modifies.
2. The conflicting trans acquires shared locks on all the stable extents
which it modifies.
3. The no_conflict trans releases the locks after its committing.
4. The conflicting trans releases the shared locks once it enter the
prepare pipeline phase.
This commits makes sure that:
1. only transactions accessing the extents that the no_conflict trans
is committing will be blocked.
2. avoid persisting extents that a committing no_conflict trans is
modified
Compared to the old approach, this commit's strategy is much more
straightforward and understandable.
This commit is more performant as the no_conflict trans only blocks
the trans the modified extent set of which intersects with it, while,
in the old approach, all other transactions will be blocked once a
conflicting trans is blocked because it's blocked after it enters the
prepare pipeline phase
Anoop C S [Wed, 15 Jul 2026 18:41:13 +0000 (00:11 +0530)]
cephadm/smb: Update CTDB bind mount for new socket path
Samba recently changed CTDB to derive its runtime paths from Samba's
configure options instead of using a monolithic CTDB_RUNDIR[1].
With current packaging[2], the CTDB socket path moved from
/var/run/ctdb/ctdbd.socket to /run/samba/ctdb/ctdbd.socket.
Update the CTDB runtime bind mount target from /var/run/ctdb to
/run/samba/ctdb to match the new CTDB_SOCKETDIR. Additionally,
create the CTDB pid directory, /run/ctdb, and lock directory,
/var/lib/samba/lock/ctdb, on the host side as they are hidden
when their parent directories are bind mounted over.
Ideally these directories should be created by sambacc but sambacc is
included in samba-container images that may be built with different
Samba packages where these path configurations might not hold true.
Kefu Chai [Fri, 23 Jan 2026 07:04:47 +0000 (15:04 +0800)]
debian/rules: exclude ceph-osd-crimson from dwz compression
When building with DWZ enabled, the debian packaging fails with:
```
dh_dwz: error: Aborting due to earlier error
```
Running the dwz command manually reveals the root cause:
```
$ dwz -mdebian/ceph-osd-crimson/usr/lib/debug/.dwz/x86_64-linux-gnu/ceph-osd-crimson.debug \
-M/usr/lib/debug/.dwz/x86_64-linux-gnu/ceph-osd-crimson.debug -- \
debian/ceph-osd-crimson/usr/bin/ceph-osd-crimson \
debian/ceph-osd-crimson/usr/bin/crimson-store-nbd
dwz: debian/ceph-osd-crimson/usr/bin/ceph-osd-crimson: Too many DIEs, not optimizing
dwz: Too few files for multifile optimization
```
The dwz tool has a limit on the number of DWARF DIEs (Debug Information
Entries) it can process. The ceph-osd-crimson binary, being a large C++
executable with extensive template usage, exceeds this limit, causing
dwz to exit with status 1 and fail the build.
This change excludes only ceph-osd-crimson from dwz processing using
the -X flag, allowing other binaries in the package to still benefit
from DWARF compression while avoiding the build failure.
Please note, we always disable DWZ in ceph-build's ceph-dev-pipeline.
* refs/pull/68813/head:
tools/cephfs: make journal pointer/header failure explicit
doc/cephfs: add note explainin the new batching technique
tools/cephfs: document --max-rss usage in first-damage.py
qa: test cephfs-journal-recovery event recover_dentries summary
tools/cephfs: flush every journal event as soon as it is scanned
Reviewed-by: Venky Shankar <vshankar@redhat.com> Reviewed-by: Patrick Donnelly <pdonnell@ibm.com>
Naman Munet [Wed, 15 Jul 2026 14:11:14 +0000 (19:41 +0530)]
mgr/dashboard: services page flickering with empty table
fixes: https://tracker.ceph.com/issues/78247
why the changes were made:
The Services table had autoSave enabled by default, which persists table state (including sort configuration) to browser localStorage. This created a conflict because:
The table uses server-side pagination ([serverSide]="true")
The table auto-reloads every 5 seconds ([autoReload]="5000")
On initialization, the table would load stale sort configuration from localStorage
This cached sort state would override the intended default sort (by service_name ascending)
User sort interactions would be saved to localStorage but could conflict with server-side sort handling
* refs/pull/69320/head:
qa:/cephfs: separate out ceph-fuse upgrade tests under fs:upgrade
qa/cephfs: reject ceph-fuse upgrade jobs to reduce failure noise
rimson/net: cap outgoing message batch size to prevent seastar packet fragment overflow
During heavy backfill, sweep_out_pending_msgs_to_sent() could accumulate
more than 65535 buffer fragments in a single bufferlist, overflowing
seastar's uint16_t packet fragment counter and triggering an assertion.
Limit the sweep to crimson_osd_max_send_buffers fragments
per write; remaining messages are sent in subsequent dispatch iterations.
Afreen Misbah [Thu, 25 Jun 2026 18:58:06 +0000 (00:28 +0530)]
mgr/dashboard: fix hardware status parsing for health
Atollon return status as a plain string ("OK") instead of
the dict format ({"health": "OK", "state": "Enabled"}) used by
other BMCs. This caused KeyError/AttributeError.
Added _get_health() helper that handles both formats:
- dict with nested health key
- plain string status
- missing or non-dict components
Reviewed-by: Redouane Kachach <rkachach@redhat.com> Reviewed-by: Laura Flores <lflores@redhat.com> Reviewed-by: Adam C. Emerson <aemerson@redhat.com> Reviewed-by: Patrick Donnelly <pdonnell@ibm.com>
projects / ceph.git / log