qa: ignore warnings

Signed-off-by: Dhairya Parmar <dparmar@redhat.com>

qa: add test cases to check client eviction if an OSD is laggy

Signed-off-by: Dhairya Parmar <dparmar@redhat.com>

mds,messages: enable beacon to report clients lagginess
using new MDS health metric

Fixes: https://tracker.ceph.com/issues/58023
Signed-off-by: Dhairya Parmar <dparmar@redhat.com>

mds: do not evict client on laggy osds

A client might get unresponsive/laggy due to laggy OSD(s).
This change provides us a way to defer client eviction in
such scenarios

also adds helpers:
- get_laggy_clients()
- clear_laggy_clients()

and call clear_laggy_clients() before calling related
Server methods

Fixes: https://tracker.ceph.com/issues/58023
Signed-off-by: Dhairya Parmar <dparmar@redhat.com>

common: add new config option to defer client eviction

Fixes: https://tracker.ceph.com/issues/58023
Signed-off-by: Dhairya Parmar <dparmar@redhat.com>

osd: add method to check for laggy osds

Fixes: https://tracker.ceph.com/issues/58023
Signed-off-by: Dhairya Parmar <dparmar@redhat.com>

Merge pull request #51473 from zdover23/wip-doc-2023-05-15-rados-operations-devices

doc/rados: edit devices.rst

Reviewed-by: Anthony D'Atri <anthony.datri@gmail.com

doc/rados: edit devices.rst

Line-edit doc/rados/operations/devices.rst.

Co-authored-by: Anthony D'Atri <anthony.datri@gmail.com>
Co-authored-by: Cole Mitchell <cole.mitchell.ceph@gmail.com>
Signed-off-by: Zac Dover <zac.dover@proton.me>

Merge PR #51386 into main

* refs/pull/51386/head:
qa: ignore cluster warning when fs flag refuse_client_session is set

Reviewed-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Kotresh Hiremath Ravishankar <khiremat@redhat.com>
Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>

Merge pull request #50806 from rhcs-dashboard/dashboard-multisite-migrate

Dashboard multisite migrate

Reviewed-by: Nizamudeen A <nia@redhat.com>

mgr/dashboard: Migrate from single site to multi-site

Signed-off-by: Aashish Sharma <aasharma@redhat.com>

Merge pull request #51463 from zdover23/wip-doc-2023-05-13-fs-volumes-1-of-x

doc/cephfs: edit fs-volumes.rst (1 of x)

Reviewed-by: Anthony D'Atri <anthony.datri@gmail.com>

doc/cephfs: edit fs-volumes.rst (1 of x)

Edit the syntax of the English language in the file
doc/cephfs/fs-volumes.rst up to (but not including) the section called
"FS Subvolumes".

Signed-off-by: Zac Dover <zac.dover@proton.me>

Merge pull request #51458 from zdover23/wip-doc-2023-05-12-cephfs-fs-volumes-prompt-rectification

doc/cephfs: rectify prompts in fs-volumes.rst

Reviewed-by: Anthony D'Atri <anthony.datri@gmail.com>

doc/cephfs: rectify prompts in fs-volumes.rst

Make sure all prompts are unselectable. This PR is meant to be
backported to Reef, Quincy, and Pacific, to get all of the prompts into
a fit state so that a line-edit can be performed on the Englsh language
in this file.

Follows https://github.com/ceph/ceph/pull/51427.

Signed-off-by: Zac Dover <zac.dover@proton.me>

Merge pull request #51448 from Matan-B/wip-matanb-crimson-only-mclock-boot

crimson/osd/scheduler/mclock_scheduler: Fix OSD unable to start

Reviewed-by: Samuel Just <sjust@redhat.com>

Merge pull request #51388 from Matan-B/wip-matanb-c-enable-rbd-tests

qa/suites/crimson: Enhance rbd api testing

Reviewed-by: Samuel Just <sjust@redhat.com>
Reviewed-by: Radosław Zarzyński <rzarzyns@redhat.com>

crimson/osd/scheduler/mclock_scheduler: Fix OSD unable to start

https://github.com/ceph/ceph/pull/49975 Introduced changes to
mclock conf value types which caused the osd to stall while booting.

Signed-off-by: Matan Breizman <mbreizma@redhat.com>

Merge PR #51251 into main

* refs/pull/51251/head:
PendingReleaseNotes: add a note about deleting files from lost+found directory
qa: add checks that validate removal of entries from lost+found dir
mds: allow unlink operation under lost+found directory

Reviewed-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>

Merge PR #51201 into main

* refs/pull/51201/head:
qa: run scrub post file system recovery

Reviewed-by: Dhairya Parmar <dparmar@redhat.com>
Reviewed-by: Kotresh Hiremath Ravishankar <khiremat@redhat.com>

Merge PR #51188 into main

* refs/pull/51188/head:
client: use deep-copy when setting permission during make_request

Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Xiubo Li <xiubli@redhat.com>

Merge pull request #51423 from bigjust/replace-go-example-mods

examples: replace example go modules with instructions to run

Reviewed-by: Yuval Lifshitz <ylifshit@redhat.com>
Reviewed-by: Ken Dreyer <kdreyer@redhat.com>

Merge pull request #51427 from zdover23/wip-doc-2023-05-10-cephfs-fs-volumes-prompt-fix

doc/cephfs: fix prompts in fs-volumes.rst

Reviewed-by: Cole Mitchell <cole.mitchell.ceph@gmail.com>

doc/cephfs: fix prompts in fs-volumes.rst

Fixed a regression introduced in
e5355e3d66e1438d51de6b57eae79fab47cd0184 that broke the unselectable
prompts in the RST.

Signed-off-by: Zac Dover <zac.dover@proton.me>

Merge pull request #51345 from cbodley/wip-59639

rgw/dbstore: allow NULL RealmIDs in sqlite schema

Reviewed-by: Soumya Koduri <skoduri@redhat.com>
Reviewed-by: Daniel Gryniewicz <dang@redhat.com>

examples: replace example go modules with instructions to run

Signed-off-by: Justin Caratzas <jcaratza@ibm.com>

Merge pull request #50627 from AliMasarweh/wip-ali-masa-multipart-populate-etag

RGW: Solving the issue of not populating etag in Multipart upload result
Reviewed-by: Daniel Gryniewicz <dang1@ibm.com>

Merge pull request #49742 from ajarr/fix-56724

mgr/rbd_support: recover from rados client blocklisting

Reviewed-by: Ilya Dryomov <idryomov@gmail.com>

Merge pull request #51166 from chrisphoffman/wip-rbd-59393

librbd: localize snap_remove op for mirror snapshots

Reviewed-by: Mykola Golub <mgolub@suse.com>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>

Merge PR #43184 into main

* refs/pull/43184/head:
qa: fix journal flush failure issue due to the MDS daemon crashes
qa: add test support for the alloc ino failing
mds: do not take the ino which has been used

Reviewed-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>

qa: run scrub post file system recovery

Running file system scrub is recommended post running filesystem
data and metadata recovery. Running scrub isn't covered in tests.

Fixes: http://tracker.ceph.com/issues/59527
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>

Merge pull request #51167 from liu-chunmei/teuthology-multicore

crimson/qa: make crimson run multicore in teuthology test

Reviewed-by: Samuel Just <sjust@redhat.com>

Merge pull request #51301 from ceph/wip-yuriw-release-16.2.13-main

doc: 16.2.13 Release Notes

doc: 16.2.13 Release Notes

Signed-off-by: Yuri Weinstein <yweinste@redhat.com>
Signed-off-by: Laura Flores <lflores@redhat.com>

Merge pull request #50411 from xxhdx1985126/wip-58928

crimson/osd: start operations asynchrously

Reviewed-by: Samuel Just <sjust@redhat.com>
Reviewed-by: Yingxin Cheng <yingxin.cheng@intel.com>
Reviewed-by: Matan Breizman <mbreizma@redhat.com>

Merge pull request #51403 from zdover23/wip-doc-2023-05-09-start-get-involved-planet-ceph

doc/start: fix "Planet Ceph" link

Reviewed-by: Anthony D'Atri <anthony.datri@gmail.com>

crimson/qa: enable multicore for crimson in teuthology test

Signed-off-by: chunmei <chunmei.liu@intel.com>

Merge pull request #47749 from xxhdx1985126/wip-intra-fixedkvbtree-pointers-2

crimson/os/seastore/btree: link fixedkvbtree's nodes and logical extents with forward and backward pointers, and drop the pin_set

Reviewed-by: Yingxin Cheng <yingxin.cheng@intel.com>
Reviewed-by: Samuel Just <sjust@redhat.com>

Merge pull request #51308 from jzhu116-bloomberg/wip-59592

rgw/notification: remove non x-amz-meta-* attributes from bucket notifications

crimson/tools/store_nbd: read logical extents via
TransactionManager::read_pin()

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/cache: add comment about backref_extent_entry_t

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

test/crimson/seastore: complement lba test with logical extents

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

test/crimson/seastore: check intra-fixedkv-btree parent->child trackers during unittests

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/btree: drop btree_pin_set_t

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/transaction_manager: follow leaf<->logical extent pointers to read extent

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/lba_manager: link lba leaf nodes with logical extents by pointers

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/btree: "templatize" btree leaf node to distinguish leaf nodes with(out) children

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/btree: link fixed-kv-btree and root_block with pointers

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore: more debug logs

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/backref_manager: retrieve live backref extents throught the backref tree

After involving intra-fixed-kv-btree parent-child pointers, we need to keep the
invariant that it's only when extents are not in transactions' read_set that
we can directly query cache with inspecting the transaction

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/btree: avoid searching transactions' read_set when retrieving btree nodes

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/btree: search fixed-kv-btree by parent<->child pointers

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/cache: invalidate out-dated extent when initiating Cache

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/cached_extent: improve the representation of "has_been_invalidated"

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/btree: don't go to leaf nodes when updating internal mappings

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/btree: introduce parent<->child pointers for fixed-kv-btree nodes

maintain correct parent<->child pointers when modifying the btree

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

doc/start: fix "Planet Ceph" link

Fix a link to Planet Ceph on the doc/start/get-involved.rst page.

Reported 2023 Apr 21, here:
https://pad.ceph.com/p/Report_Documentation_Bugs

Signed-off-by: Zac Dover <zac.dover@proton.me>

Merge pull request #51355 from aravind-wdc/wip-crimson-zbd

crimson/os/seastore: enable SMR HDD

Reviewed-by: Yingxin Cheng <yingxin.cheng@intel.com>

Merge pull request #51392 from parth-gr/rgw-mutisite-ceph-doc

doc: update multisite doc

Reviewed-by: Jiffin Tony Thottan <jthottan@redhat.com>
Reviewed-by: Casey Bodley <cbodley@redhat.com>
Reviewed-by: Zac Dover <zac.dover@proton.me>

Merge pull request #50857 from kamoltat/wip-ksirivad-iswriteable

mon/Monitor.cc: exit function if !osdmon()->is_writeable()
Reviewd-by: Gregory Farnum <gfarnum@redhat.com>

Merge pull request #51394 from rzarzynski/wip-doc-encode-stdoptional

doc/dev/encoding.txt: update per std::optional

Reviewed-by: Zac Dover <zac.dover@proton.me>

qa/workunits/rbd: Add tests for rbd_support module recovery

... after the module's RADOS client is blocklisted.

Signed-off-by: Ramana Raja <rraja@redhat.com>

mgr/rbd_support: recover from rados client blocklisting

In certain scenarios the OSDs were slow to process RBD requests.
This lead to the rbd_support module's RBD client not being able to
gracefully handover a RBD exclusive lock to another RBD client.
After the condition persisted for some time, the other RBD client
forcefully acquired the lock by blocklisting the rbd_support module's
RBD client, and consequently blocklisted the module's RADOS client. The
rbd_support module stopped working. To recover the module, the entire
mgr service had to be restarted which reloaded other mgr modules.

Instead of recovering the rbd_support module from client blocklisting
by being disruptive to other mgr modules, recover the module
automatically without restarting the mgr serivce. On client getting
blocklisted, shutdown the module's handlers and blocklisted client,
create a new rados client for the module, and start the new handlers.

Fixes: https://tracker.ceph.com/issues/56724
Signed-off-by: Ramana Raja <rraja@redhat.com>

Merge pull request #51365 from nbalacha/fix-remove-unused-type

librbd: remove unused enum WriteOpType

Reviewed-by: Ilya Dryomov <idryomov@gmail.com>

Merge pull request #49975 from sseshasa/wip-fix-mclk-rec-backfill-cost

osd: mClock recovery/backfill cost fixes

Reviewed-by: Sam Just <sjust@redhat.com>
Reviewed-by: Radosław Zarzyński <rzarzyns@redhat.com>

pybind/rados: add ConnectionShutdown exception class

Signed-off-by: Ramana Raja <rraja@redhat.com>

mgr/rbd_support: notify the thread waiting on pending snapshot

... requests to be completed.

Signed-off-by: Ramana Raja <rraja@redhat.com>

Merge pull request #51381 from Matan-B/wip-matanb-c-blocklist-fix

crimson/osd/osd_operations/client_request: Fix client blocklisting

Reviewed-by: Yingxin Cheng <yingxin.cheng@intel.com>

Merge pull request #43245 from thiagoarrais/docs-java-examples

[rgw]: Update AWS SDK in Java examples

doc/dev/encoding.txt: update per std::optional

Signed-off-by: Radoslaw Zarzynski <rzarzyns@redhat.com>

librbd: localize snap_remove op for mirror snapshots

A client may attempt a lock request not quickly enough to
obtain exclusive lock for operations when another competing
client responds quicker. This can happen when a peer site has
different performance characteristics or latency. Instead of
relying on this unpredictable behavior, localize operation to
primary cluster.

Fixes: https://tracker.ceph.com/issues/59393
Signed-off-by: Christopher Hoffman <choffman@redhat.com>

doc: update multisite doc

cmd for getting zone group was spelled incorrectly
Updated to rdosgw-admin

Signed-off-by: parth-gr <paarora@redhat.com>

librbd : remove unused enum type WriteOpType

This removes the unused enum WriteOpType from
the librbd deep_copy code.

Signed-off-by: N Balachandran <nibalach@redhat.com>

qa: ignore cluster warning when fs flag refuse_client_session is set

Fixes: https://tracker.ceph.com/issues/59667
Introduced-by: https://github.com/ceph/ceph/pull/48720
Signed-off-by: Dhairya Parmar <dparmar@redhat.com>

Merge pull request #51387 from zdover23/wip-doc-2023-05-08-rados-operations-stretch-mode-other-commands

doc/rados: stretch-mode.rst (other commands)

Reviewed-by: Cole Mitchell <cole.mitchell.ceph@gmail.com>

doc/rados: stretch-mode.rst (other commands)

Edit the "Other Commands" section of
doc/rados/operations/stretch-mode.rst.

Signed-off-by: Zac Dover <zac.dover@proton.me>

qa/suites/crimson: Introduce rbd_python_api_tests.yaml

Test python api with new image format.

Signed-off-by: Matan Breizman <mbreizma@redhat.com>

qa/suites/crimson: Skip unsupported tests (Crimson)

Align with `rbd_api_tests` and skip deep_copy and breaklock tests
in Crimson.

Signed-off-by: Matan Breizman <mbreizma@redhat.com>

qa/: Override mClock profile to 'high_recovery_ops' for qa tests

The qa tests are not client I/O centric and mostly focus on triggering
recovery/backfills and monitor them for completion within a finite amount
of time. The same holds true for scrub operations.

Therefore, an mClock profile that optimizes background operations is a
better fit for qa related tests. The osd_mclock_profile is therefore
globally overriden to 'high_recovery_ops' profile for the Rados suite as
it fits the requirement.

Also, many standalone tests expect recovery and scrub operations to
complete within a finite time. To ensure this, the osd_mclock_profile
options is set to 'high_recovery_ops' as part of the run_osd() function
in ceph-helpers.sh.

A subset of standalone tests explicitly used 'high_recovery_ops' profile.
Since the profile is now set as part of run_osd(), the earlier overrides
are redundant and therefore removed from the tests.

Signed-off-by: Sridhar Seshasayee <sseshasa@redhat.com>

doc/: Modify mClock configuration documentation to reflect profile changes

Modify the relevant documentation to reflect:

- change in the default mClock profile to 'balanced'
- new allocations for ops across mClock profiles
- change in the osd_max_backfills limit
- miscellaneous changes related to warnings.

Signed-off-by: Sridhar Seshasayee <sseshasa@redhat.com>

common/options/osd.yaml.in: Change mclock max sequential bandwidth for SSDs

The osd_mclock_max_sequential_bandwidth_ssd is changed to 1200 MiB/s as
a reasonable middle ground considering the broad range of SSD capabilities.
This allows the mClock's cost model to extract the SSDs capability
depending on the cost of the IO being performed.

Signed-off-by: Sridhar Seshasayee <sseshasa@redhat.com>

osd/: Retain the default osd_max_backfills limit to 1 for mClock

The earlier limit of 3 was still aggressive enough to have an impact on
the client and other competing operations. Retain the current default
for mClock. This can be modified if necessary after setting the
osd_mclock_override_recovery_settings option.

Signed-off-by: Sridhar Seshasayee <sseshasa@redhat.com>

common/options/osd.yaml.in: change mclock profile default to balanced

Let's use the middle profile as the default.
Modify the standalone tests accordingly.

Signed-off-by: Samuel Just <sjust@redhat.com>
Signed-off-by: Sridhar Seshasayee <sseshasa@redhat.com>

osd/scheduler/mClockScheduler: avoid limits for recovery

Now that recovery operations are split between background_recovery and
background_best_effort, rebalance qos params to avoid penalizing
background_recovery while idle.

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/: add counters for ops delayed due to degraded|unreadable target

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/: add counters for queue latency for PGRecovery[Context]

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/: add per-op latency averages for each recovery related message

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/: differentiate priority for PGRecovery[Context]

PGs with degraded objects should be higher priority.

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/: add MSG_OSD_PG_(BACKFILL|BACKFILL_REMOVE|SCAN) as recovery messages

Otherwise, these end up as PGOpItem and therefore as immediate:

class PGOpItem : public PGOpQueueable {
...
  op_scheduler_class get_scheduler_class() const final {
    auto type = op->get_req()->get_type();
    if (type == CEPH_MSG_OSD_OP ||
  type == CEPH_MSG_OSD_BACKOFF) {
      return op_scheduler_class::client;
    } else {
      return op_scheduler_class::immediate;
    }
  }
...
};

This was probably causing a bunch of extra interference with client
ops.

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/: differentiate scheduler class for undersized/degraded vs data movement

Recovery operations on pgs/objects that have fewer than the configured
number of copies should be treated more urgently than operations on
pgs/objects that simply need to be moved to a new location.

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/.../OpSchedulerItem: add MSG_OSD_PG_PULL to is_recovery_msg

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/: move PGRecoveryMsg check from osd into PGRecoveryMsg::is_recovery_msg

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/: move get_recovery_op_priority into PeeringState next to get_*_priority

Consolidate methods governing recovery scheduling in PeeringState.

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/scheduler: simplify qos specific params in OpSchedulerItem

is_qos_item() was only used in operator<< for OpSchedulerItem.  However,
it's actually useful to see priority for mclock items since it affects
whether it goes into the immediate queues and, for some types, the
class.  Unconditionally display both class_id and priority.

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/scheduler: remove unused PGOpItem::maybe_get_mosd_op

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/scheduler: remove OpQueueable::get_order_locker() and supporting machinery

Apparently unused.

Signed-off-by: Samuel Just <sjust@redhat.com>

osd/scheduler: remove OpQueueable::get_op_type() and supporting machinery

Apparently unused.

Signed-off-by: Samuel Just <sjust@redhat.com>

PeeringState::clamp_recovery_priority: use std::clamp

Signed-off-by: Samuel Just <sjust@redhat.com>

doc: Modify mClock configuration documentation to reflect new cost model

Signed-off-by: Sridhar Seshasayee <sseshasa@redhat.com>

osd: Retain overridden mClock recovery settings across osd restarts

Fix an issue where an overridden mClock recovery setting (set prior to
an osd restart) could be lost after an osd restart.

For e.g., consider that prior to an osd restart, the option
'osd_max_backfill' was successfully set to a value different from the
mClock default. If the osd was restarted for some reason, the
boot-up sequence was incorrectly resetting the backfill value to the
mclock default within the async local/remote reservers. This fix
ensures that no change is made if the current overriden value is
different from the mClock default.

Modify an existing standalone test to verify that the local and remote
async reservers are updated to the desired number of backfills under
normal conditions and also across osd restarts.

Signed-off-by: Sridhar Seshasayee <sseshasa@redhat.com>