Ilya Dryomov [Sat, 18 Jun 2022 11:00:34 +0000 (13:00 +0200)]
rbd-mirror: fix up PrepareReplayDisconnected test case
It was botched in commit 2bca9ee96c65 ("rbd-mirror: consolidate
prepare local/remote image steps to bootstrap") and went unnoticed
because currently no special handling is needed for disconnected
clients -- is_disconnected() check happens to be the last step
and it doesn't generate an error.
Ilya Dryomov [Mon, 20 Jun 2022 12:19:41 +0000 (14:19 +0200)]
rbd-mirror: generally skip replay/resync if remote image is not primary
Replay and resync should generally be skipped if the remote image is
not primary.
If this is not done for replay, snapshot-based mirroring can run into
a livelock if the primary image is demoted while a mirror snapshot is
being synced. On the demote site, rbd-mirror would pick up the just
demoted image, grab the exclusive lock on it and idle waiting for a new
mirror snapshot to be created. On the (still) non-primary site,
rbd-mirror would eventually finish syncing that mirror snapshot and
attempt to unlink from it on the demote site. These attempts would
fail with EROFS due to exclusive lock being held in the "refuse proxied
maintenance operations" mode, blocking forward progress (syncing of the
demotion snapshot so that the non-primary image can be orderly promoted
to primary, etc).
If this is not done for resync, data loss can ensue as the just demoted
image would be immediately trashed, underneath the non-primary site that
is still syncing.
Currently this is done in PrepareReplayRequest only for journal-based
mirroring. Note that it is conditional: if the local image is linked
to the remote image, proceeding is desirable.
Generalize this check, consolidate it with a related check in
PrepareRemoteImageRequest and move the result to BootstrapRequest to
cover both "local image does not exist" and "local image is unlinked"
cases for both modes.
Ilya Dryomov [Sat, 18 Jun 2022 10:35:51 +0000 (12:35 +0200)]
rbd-mirror: strengthen is_local_primary() and is_linked()
Initialize local_promotion_state and remote_promotion_state to UNKNOWN
instead of counterintuitive PRIMARY and NON_PRIMARY -- half the time the
final values are flipped. Then is_local_primary() and is_linked() can
be strengthened as a non-existent image should stay in UNKNOWN.
Zac Dover [Tue, 14 Jun 2022 22:15:33 +0000 (08:15 +1000)]
doc/dev: s/master/main/ in basic workflow
This PR changes "master" to "main" in the
basic_workflow.rst file. I have even changed
"master" to "main" in some terminal output from
several years ago. This isn't historically ac-
curate, of course, but my hope is that this change
will prevent someone in the future from being con-
fused about why an antiquated branch name is ref-
erred to.
Or Friedmann [Tue, 19 Apr 2022 12:00:28 +0000 (12:00 +0000)]
rgw: RGWCoroutine::set_sleeping() checks for null stack
users of the RGWOmapAppend coroutine don't manage the lifetime of its
underlying coroutine stack, so end up making calls on RGWOmapAppend
after its stack goes away. this null check is a band-aid, and there are
still several other calls in RGWCoroutine that don't check for null
stack
Fixes: https://tracker.ceph.com/issues/49302 Signed-off-by: Or Friedmann <ofriedma@redhat.com> Signed-off-by: Casey Bodley <cbodley@redhat.com>
(cherry picked from commit 3f0f831d66c7d43c9872f5de2aceb68aef4004d8)
mon/OSDMonitor: Ensure kvmon() is writeable before handling "osd new" cmd
Before proceeding to handle "osd new" mon command as part of
OSDMonitor::prepare_command_impl(), a check is made to verify if the
authmon is writeable. Later on, prepare_command_osd_new() invokes
KVMonitor::do_osd_new() to create pending dmcrypt keys and calls
propose_pending(). The propose could fail (with an assertion failure)
if there was a prior mon command that resulted in the kvmon invoking
propose_pending().
In order to avoid such a situation, introduce a check to verify that
kvmon is also writeable in OSDMonitor::prepare_command_impl(). If it
is not writeable, the op is pushed into the wait_for_active context
queue to be retried later.
Zac Dover [Mon, 13 Jun 2022 21:48:46 +0000 (07:48 +1000)]
doc/dev: s/master/main/ essentials.rst dev guide
This PR changes all reference to the "master" branch
to references to the "main" branch (because we renamed
"master" to main", and the docs now need to reflect that).
Zac Dover [Sun, 12 Jun 2022 23:41:28 +0000 (09:41 +1000)]
doc/start: rewrite CRUSH para
This PR supersedes https://github.com/ceph/ceph/pull/46584
and makes changes suggested by Anthony D'Atri that improve
the coherence and consistency of the paragraph that explains
the basics of the CRUSH algorithm.
Zac Dover [Wed, 8 Jun 2022 19:19:16 +0000 (05:19 +1000)]
doc/start: make OSD and MDS structures parallel
This PR makes the "Ceph OSDs" and "MDSs" bullet points
parallel by naming "object storage daemon" before referring
to the (admittedly more common and colloquial, but surely
unknown to people who genuinely require a document called
'Intro') acronym "OSD".
Zac Dover [Mon, 13 Jun 2022 04:34:36 +0000 (14:34 +1000)]
doc/start: rewrite hardware-recs networks section
This rewrites the first two-thirds of the "Networks"
section of the Hardware Recommendations page in the
Intro to Ceph document. I have tried to divide the
techincal content in this section into subsections
that foreground the various subjects covered.
7. Checking number of pg log's entries and dups
```
rzarz@ubulap:~/dev/ceph/build$ for pgid in `cat osd0_pgs.txt`; do echo $pgid; bin/ceph-objectstore-tool --data-path dev/osd0 --op log --pgid $pgid | jq '(.pg_log_t.log|length),(.pg_log_t.dups|length)'; done
2.7
10020
2500
2.6
10100
3000
2.3
10012
2800
2.1
10049
2900
2.2
10057
2700
2.0
10027
2900
2.5
10077
2700
2.4
10072
2900
1.0
97
0
```
Conflicts:
src/tools/ceph_objectstore_tool.cc -- undetected conflict
with d5445b8f113797718a0dbb05e884a6bffbfed76a. Fixed by
adopting the patch no not require the `unique_ptr<T>::get()`.
Although the chunking in off-line `dups` trimming (via COT) seems
fine, the `ceph-objectstore-tool` is a client of `trim()` of
`PGLog::IndexedLog` which means than a partial revert is not
possible without extensive changes. Moreover, trimming pg log
is not enough without modifying pg_info_t accordingly which
the reverted patch lacks.
The backport ticket is: https://tracker.ceph.com/issues/55989
Ilya Dryomov [Thu, 9 Jun 2022 11:42:01 +0000 (13:42 +0200)]
cmake: pass -Wno-error when building PMDK
It's hitting pacific with a nuisance -Werror=array-parameter= const
char * vs const char[37] mismatch. Follow commit 91a616b26e83 ("cmake:
pass RTE_DEVEL_BUILD=n when building dpdk") and just disable -Werror.
Jos Collin [Thu, 24 Mar 2022 04:57:58 +0000 (10:27 +0530)]
qa: make test_perf_stats_stale_metrics check only the clients created for the tests
Uses the client's global id to get the metrics, instead of using the index.
This ensures that test_perf_stats_stale_metrics checks only the clients mounted for
the tests.
Fixes: https://tracker.ceph.com/issues/54971 Signed-off-by: Jos Collin <jcollin@redhat.com>
(cherry picked from commit 1621308214cd18750c8be803fc014bdf73e2218a)
Jos Collin [Tue, 29 Jun 2021 09:59:01 +0000 (15:29 +0530)]
mgr, pybind/mgr, mgr/stats: be resilient to offline rank0 MDS
Reregister the user queries during the rank0 MDS failover event
by calling listener.handle_query_updated(). This enables
`ceph fs perf stats` to receive the updated metrics after the
failover.
Fixes: https://tracker.ceph.com/issues/50033 Signed-off-by: Jos Collin <jcollin@redhat.com>
(cherry picked from commit c2470f271cce4d512f2cf00552c9b753e4c69f71)
Revert "osd/PGLog.cc: Trim duplicates by number of entries"
This reverts commit d49ff13c80bbed5fb6c064f9689bb2b6a0a251b8.
which is the in-OSD part of the fix for accumulation of `dup`
entries in a PG Log. Brainstorming it has brought questions
on the OSD's behaviour during an upgrade if there are tons of
dups in the log. What must be double-checked before bringing
it back is ensuring we chunk the deletions properly to not
impose OOMs / stalls in, to exemplify, RocksDB.
The backport ticket is: https://tracker.ceph.com/issues/55989
We really want to have the ability to know how many
entries `PGLog::IndexedLog::dups` has inside.
The current ways are either invasive (stopping an OSD)
or indirect (examination of `dump_mempools`).
client/fuse: Fix directory DACs overriding for root
DACs are overridable for directories. For files,
Read/write DACs are always overridable but executable
DACs are overridable when there is at least one exec bit
set.
The files and directory DACS overriding were handled the
same way for root which is incorrect. This patch fixes
DACs overriding as described above for the root.
This was attempted in commit 69a7ed4eab36 ("run-make-check: enable
WITH_RBD_RWL when WITH_PMEM is true") but never completed. We soon
bumped the requirement on libpmem, so WITH_SYSTEM_PMDK=ON wouldn't
have worked anyway.
Enable the RWL mode conditionally based on WITH_RBD_RWL variable.
Enable the SSD mode unconditionally as it has no special dependencies
and can be built on any architecture.
test/encoding/check-generated.sh: show diff if binary reencode check fails
Take bf0b161115aa ("test/encoding/check-generated.sh: show diff if cmp
fails") a bit further. Suggesting "cmp $tmp1 $tmp2" isn't very helpful
since cmp would report just the mismatch offset.
librbd/cache/pwl: WriteLogCacheEntry constructor must initialize flags
Initializing the individual bit field members leaves the remaining two
bits uninitialized and that garbage state gets persisted.
In general, using bit fields in a structure where the layout actually
matters is not desirable. Even with a few single bits, such as here,
their order, strictly speaking, is not guaranteed:
An implementation may allocate any addressable storage unit large
enough to hold a bit-field. If enough space remains, a bit-field
that immediately follows another bit-field in a structure shall be
packed into adjacent bits of the same unit. If insufficient space
remains, whether a bit-field that does not fit is put into the next
unit or overlaps adjacent units is implementation-defined. The
order of allocation of bit-fields within a unit (high-order to
low-order or low-order to high-order) is implementation-defined.
The alignment of the addressable storage unit is unspecified.
this mutex was only held by one function, OpsLogFile::flush(). this
private member function is only ever called from the background thread,
so doesn't need to be protected by a mutex
as a further cleanup, i renamed 'cond' and 'mutex' now that we don't
need to differentiate between different locks
this shuts up ceph::debug_condition_variable's assertion that the
associated mutex is held during notify_one(). this is not strictly
required for correct use, but is a common source of bugs
Xiubo Li [Mon, 20 Dec 2021 02:21:50 +0000 (10:21 +0800)]
client: force a request to update the perms
If the caller perms is different with the current dirtier's, we
can still cache the attributes locally if possible, but will
mark the ctime needs to be updated forcely and the request will
implicitly flush the dirty caps for us.
Xiubo Li [Mon, 11 Oct 2021 08:45:16 +0000 (16:45 +0800)]
client: improve the code of setattr for atime/mtime
When the Fw caps is issued and increasing the atime/mtime, we can
buffer the atime/mtime. And when the Fs caps is issued and they
are not changing we can just ignore them.
In case when recovering the atime/mtime maybe decreased, no matter
the Fsw caps are issued, always send the request immediatelly.
projects / ceph.git / log