Edwin Rodriguez [Thu, 7 Aug 2025 20:28:44 +0000 (16:28 -0400)]
test/osd: Suppress subobject-linkage warning in SelectMappingAndLayers class
Change SelectMapping and SelectLayers definitions to use non-static arrays of strings.
SelectMappingAndLayers::sma and SelectMappingAndLayers::sly have internal storage
duration, because it is a non-template, non-inline, non-extern const-qualified variable.
As a consequence, in each translation unit sma and sly is a different object.
And because ProgramOptionSelector takes a reference as template argument, then
ProgramOptionSelector<...> are different specializations of ProgramOptionSelector
in different translation units, because the template parameter references different objects in each.
Then, if you include the header in two different translation units, the program will
have undefined behavior, because the definitions of SelectMapping violates the one-definition
rule as they are, roughly said, not semantically identical. The compiler has no way to
decide whether SelectMapping is supposed to have ProgramOptionSelector<value1> or
ProgramOptionSelector<value2> as base class (where value1 and value2 are invented names
for the two instances of io_sequence::tester::lrc::mapping_layer_array_sizes in the
different translation units).
gal salomon [Wed, 11 Jun 2025 18:05:03 +0000 (21:05 +0300)]
the current connection setup is single and shared connection, the strand on that single connection may cause a serialization.
it should be noted that per s3-request there are several redis-operation that may run on co-routine.
the redis-connection pool implement the guarded acquire/release APIs.
adding configuration : rgw_redis_connection_pool_size.
re-factor of redis-exec* function.
shared pointer for Redis connection pool
adding branch predication optimization for redis-pool/single-shared-connection condition
adding a warning-report-method for the case there is a blocking state upon empty connection pool.
Dan Mick [Tue, 26 Aug 2025 00:45:21 +0000 (17:45 -0700)]
Remove git clean -fdx
either
1) a source tarball is supplied, in which case the local dir is
irrelevant, or
2) make-debs calls make-dist, which doesn't care about a dirty cwd
so it just punishes the unaware by removing things that they may
have wanted to keep.
Afreen Misbah [Wed, 13 Aug 2025 06:49:02 +0000 (12:19 +0530)]
mgr/dashboard: Add /health/snapshot api
Fixes https://tracker.ceph.com/issues/72609
- The current minimal API relies on fetching data from osdmap and pgmap.
- These commands produce large, detailed payloads that become a performance bottleneck and impact scalability, especially in large clusters.
- To address this, we propose switching to the ceph snapshot API using ceph status command, which retrieves essential information directly from the cluster map.
- ceph status is significantly more lightweight compared to osdmap/pgmap, reducing payload sizes and processing overhead.
- This change ensures faster response times, improves system efficiency in large deployments, and minimizes unnecessary data transfer.
- update tests
Dan Mick [Sat, 23 Aug 2025 00:43:24 +0000 (17:43 -0700)]
make-debs.sh: invoke tar with --no-same-owner
When running as a normal user, tar does not attempt to preserve
owners set on the tar content files. When running as root, it does.
Containerized builds are running as root. Stop make-debs.sh from
trying to set other owners for files, and leaving files in the
host system with mapped UIDs other than the user running the container
(which causes jenkins to be unable to clear the workspace).
rgw/d4n: modified update method to optionally take a bool
for dirty flag, such that when a flag is not set, then the method
re-uses the old value of dirty flag using in-memory data structure.
This is helpful in eliminating a call to directory to fetch the
'dirty' flag in the flush() method.
rgw/d4n: optimizing iterate method to align
last block also with max_chunk_size(object size
or rgw_max_chunk_size) and to perform checks
based on object size.
Zac Dover [Fri, 22 Aug 2025 08:39:29 +0000 (18:39 +1000)]
doc/cephfs: edit troubleshooting.rst (Slow MDS)
Move the "Slow requests (MDS)" section immediately after the first
section in this document ("Slow/Stuck Operations"), because the first
procedure on the page directs the reader to undertake the operation in
"Slow requests (MDS)" before trying anything else.
Dan Mick [Thu, 21 Aug 2025 20:00:43 +0000 (13:00 -0700)]
make-debs.sh: make "skip debug packages" conditional
Now that we're using make-debs.sh as a builder inside containers,
the default should be to build all the packages, including debug.
(Also, fix a typo.)
Ilya Dryomov [Thu, 21 Aug 2025 19:39:29 +0000 (21:39 +0200)]
mon/MonClient: post version request completions outside of monc_lock
dispatch() is allowed to invoke the completion object in the current
thread, before control returns from dispatch(). This isn't desirable
when it comes to discarding version requests in MonClient::shutdown()
and MonClient::_reopen_session() because completion objects could then
be invoked under monc_lock. In case of MonClient::_reopen_session() in
particular, this leads to an attempt to acquire monc_lock once again in
MonClient::get_version() on a retry due to monc_errc::session_reset
that is converted to errc::resource_unavailable_try_again:
MonClient::ms_handle_reset
< takes monc_lock >
MonClient::_reopen_session
< invokes the completion object via dispatch() with ec == monc_errc::session_reset >
Objecter::CB_Objecter_GetVersion::operator() [ ec == errc::resource_unavailable_try_again ]
Objecter::_wait_for_latest_osdmap
MonClient::get_version
< attempts to take monc_lock in the body of the lambda >
The end result is either a lockup or some form of undefined behavior.
The best possible outcome here is an exception (std::system_error with
"Resource deadlock avoided" error) and a successive call to
std::terminate().
This is a regression introduced in commit e81d4eae4e76 ("common/async:
Update `use_blocked` for newer asio"). Revert to posting version
request completions for the error cases in a way that is uniform with
the success case in MonClient::handle_get_version_reply().
Samuel Just [Tue, 12 Aug 2025 00:36:16 +0000 (17:36 -0700)]
crimson/.../store-bench: refactor arguments and workloads
- Adds a workload abstraction grouping arguments with associated
workload.
- Reworks argument parsing to occur prior to seastar app, allows
passing unparsed arguments to ceph.
- Refactors time usages as necessary to use std::chrono types.
- Removes LOG_PREFIX usages that don't currently have log lines.
- Other minor cleanups not worth separating out.
make_snapmapper_oid was missing for the new PGs created post-splitting.
This was causing a scrub error due to the missing snap mapper object in the children PGs.
crimson/osd/scrub: check if stats have been marked invalid
In case of splits/merges the PeeringState::split_into function will mark
the child and parent pgs stats invalid.
We need to check for this when scrub finishes, update the stats and mark them valid.
Kefu Chai [Mon, 9 Jun 2025 12:35:44 +0000 (20:35 +0800)]
src: Fix memory leaks in generate_test_instance() by returning values instead of pointers
Problem:
The current `generate_test_instance()` function returns `std::list<T*>`,
which creates memory management issues:
- Inconsistent lifecycle management of test instances
- Callers don't always properly clean up allocated memory
- Results in ASan memory leak reports in unit tests and ceph-dencoder
Solution:
Change `generate_test_instance()` to return `std::list<T>` instead of `std::list<T*>`:
Core Changes:
- Modified all classes with `generate_test_instance()` to return `std::list<T>`
- Use `emplace_back()` without parameters** to avoid copy/move
constructors for classes that don't define them
- Updated ceph-dencoder to handle the new return type
ceph-dencoder Adaptations:
Since `m_list` now holds `T` objects instead of `T*`, and we can't
assume `T` is copyable/moveable:
- Keep `m_object` as a pointer for flexibility
- Handle two scenarios:
1. `m_object` points to an element in `m_list`
2. `m_object` points to a decoded instance (requires manual cleanup)
- Introduce `make_ptr()` as a factory function to create a smart pointer
to conditionally free the managed pointer.
Additional Cleanup:
- Simplify DencoderBase constructor from template to plain
function (extra parameters were never used in derived classes)
With this change, object lifecycles are now managed by value semantics
instead of raw pointers, eliminating memory leaks.
projects / ceph.git / log