Merge pull request #5283 from SUSE/wip-12397-hammer

ceph.spec.in: 95-ceph-osd.rules, mount.ceph, and mount.fuse.ceph not installed properly on SUSE

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #5206 from SUSE/wip-11998-hammer

/usr/bin/ceph from ceph-common is broken without installing ceph

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #5055 from SUSE/wip-12044-hammer

rgw/logrotate.conf calls service with wrong init script name

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #5040 from SUSE/wip-11964-hammer

systemd: Increase max files open limit for OSD daemon

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #5038 from SUSE/wip-11876-hammer

ceph-post-file fails on rhel7

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #5030 from SUSE/wip-12092-hammer

packaging: add SuSEfirewall2 service files

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #5028 from SUSE/wip-12090-hammer

rcceph script is buggy

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #5026 from SUSE/wip-12087-hammer

max files open limit for OSD daemon is too low

Reviewed-by: Loic Dachary <ldachary@redhat.com>

ceph.spec.in: install 95-ceph-osd.rules, mount.ceph, and mount.fuse.ceph properly on SUSE

http://tracker.ceph.com/issues/12261 Fixes: #12261

Signed-off-by: Nathan Cutler <ncutler@suse.com>
(cherry picked from commit 5ce38b9536efabf99a236c7a9d15c149fa4c16a6)

debian: move ceph_argparse into ceph-common

Prior to this commit, if a user installed the "ceph-common" Debian
package without installing "ceph", then /usr/bin/ceph would crash
because it was missing the ceph_argparse library.

Ship the ceph_argparse library in "ceph-common" instead of "ceph". (This
was the intention of the original commit that moved argparse to "ceph",
2a23eac54957e596d99985bb9e187a668251a9ec)

http://tracker.ceph.com/issues/11388 Refs: #11388

Reported-by: Jens Rosenboom <j.rosenboom@x-ion.de>
Signed-off-by: Ken Dreyer <kdreyer@redhat.com>
(cherry picked from commit 110608e5bdd9e2f03020ad41f0c2d756684d4417)

Conflicts:
debian/ceph.install
            There is no ceph_daemon.py in hammer
        debian/control
            Depends/Replaces/Breaks version adapted (from 9.0.0 to 0.94.2)
            also adapted ceph-dbg Replaces/Breaks

Merge pull request #5046 from ceph/wip-12109-hammer

librbd: new QA client upgrade tests

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #5159 from theanalyst/wip-11701-hammer

make the all osd/filestore thread pool suicide timeouts separately configurable

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

Merge pull request #5252 from ceph/wip-12021-hammer

OSDMonitor: allow addition of cache pool with non-empty snaps with co…

Reviewed-by: Sage Weil <sage@redhat.com>

Merge pull request #4891 from theanalyst/wip-11740-hammer

crush: take crashes due to invalid arg

Reviewed-by: Loic Dachary <ldachary@redhat.com>

OSD: add command_wq suicide timeout

Signed-off-by: Samuel Just <sjust@redhat.com>
(cherry picked from commit df4e5de819c30003cfbe50a071c49039cf534419)

Conflicts:
src/common/config_opts.h
Trivial merge conflict

OSD: add remove_wq suicide timeout

Signed-off-by: Samuel Just <sjust@redhat.com>
(cherry picked from commit f2fbfa32a16666be46359f0eab7b04ca80a753f5)

OSD: add scrub_wq suicide timeout

Signed-off-by: Samuel Just <sjust@redhat.com>
(cherry picked from commit 547a7041edc833f3cc8e04d388574809e30a8af6)

OSD: add snap_trim_wq suicide timeout

Signed-off-by: Samuel Just <sjust@redhat.com>
(cherry picked from commit e1073a4a577211672148a4112bd633831552d66f)

OSD: add recovery_wq suicide timeout

Signed-off-by: Samuel Just <sjust@redhat.com>
(cherry picked from commit 85311b656852af75bfbbc6699f92fc6aa233c316)

Conflicts: src/common/config_opts.h

There was a merge conflict due to introduction of `osd_recovery_sleep`
which was introduced in #3829

OSD: add op_wq suicide timeout

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

Merge pull request #4899 from theanalyst/wip-11911-hammer

start_flush: filter out removed snaps before determining snapc's

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

Merge pull request #4868 from SUSE/wip-11879-hammer

Clock skew causes missing summary and confuses Calamari

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

Merge pull request #4883 from SUSE/wip-11638-hammer

ceph.spec.in: ceph-common subpackage def needs tweaking for SUSE/openSUSE

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

Merge pull request #5208 from tchaikov/wip-11975-hammer

tests: TEST_crush_reject_empty must not run a mon

Reviewed-by: Loic Dachary <ldachary@redhat.com>

crush/CrushTester: return EINVAL if crushtool returns non-zero

this backports a tiny part of ec02441, otherwise
CrushTester will return 1, and "ceph" cli will take it
as EPERM, which is miss leading, and fails
osd-crush.sh:TEST_crush_reject_empty.

Signed-off-by: Kefu Chai <kchai@redhat.com>

tests: TEST_crush_reject_empty must not run a mon

* Back in Hammer, the osd-crush.sh individual tests did not run the
  monitor, it was taken care of by the run() function. An attempt to run
  another mon fails with:

  error: IO lock testdir/osd-crush/a/store.db/LOCK: Resource temporarily
  unavailable

  This problem was introduced by cc1cc033930e8690a57674e842a003f6bbc7a242
  from https://github.com/ceph/ceph/pull/4936
* replace test/mon/mon-test-helpers.sh with test/ceph-helpers.sh as
  we need run_osd() in this newly added test
* update the run-dir of commands: ceph-helpers.sh use the different
  convention for the run-dir of daemons.

http://tracker.ceph.com/issues/11975 Refs: #11975

Signed-off-by: Loic Dachary <ldachary@redhat.com>

ceph-helpers: implement test_expect_failure

To display the output in case the command did not fail with the expected
output.

Signed-off-by: Loic Dachary <ldachary@redhat.com>
(cherry picked from commit 5871781b10ff0b26c731b70d1898c474006cbee3)

tests: display the output of failed make check runs

After a make check fails, it shows a summary but not the output of the
failed tests although they contain information to diagnose the problem.

Set the VERBOSE=true automake variable which is documented to collect
and display the failed script output at the end of a run (the content of
the test-suite.log file (valid from automake-1.11 up).

http://www.gnu.org/software/automake/manual/automake.html#index-VERBOSE

Also remove the run-make-check.sh that did the same in a way that is not
compatible with automake-1.11.

Signed-off-by: Loic Dachary <ldachary@redhat.com>
(cherry picked from commit 3a55cb029bb7db9542d2b14f2deda90feb0ae0f6)

Merge pull request #4889 from theanalyst/wip-11484-hammer

OPT_INT option interprets 3221225472 as -1073741824, and crashes in Throttle::Throttle()

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #4776 from tchaikov/wip-11279-hammer

ceph: cli interactive mode does not understand quotes

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #4657 from ceph/wip-hammer-11535-admin-socket

common/admin_socket: close socket descriptor in destructor

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #4687 from SUSE/wip-7387-hammer

utf8 and old gcc breakage on RHEL6.5

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #5122 from theanalyst/wip-11982-hammer

ceph fails to compile with boost 1.58

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #4936 from ceph/wip-11975-hammer

mon crashes when "ceph osd tree 85 --format json"

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #4892 from theanalyst/wip-11760-hammer

ceph-disk: get_partition_type fails on /dev/cciss...

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #4877 from SUSE/wip-11902-hammer

admin/build-doc: script fails silently under certain circumstances

Reviewed-by: Loic Dachary <ldachary@redhat.com>

mon: add "--check" to CrushTester::test_with_crushtool()

so we don't need to call CrushTester::check_name_maps() in OSDMonitor.cc
anymore.

Fixes: #11680
Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit c6e634875316cf17368d497e6dc4f6f4b5dd65d2)

crushtool: rename "--check-names" to "--check"

* because "--check" also checks for the max_id

Note: edited since we do not have the fix introduced in 46103b2 in
      hammer.

Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit 9381d53acdce85fcbff828926b911e050ba36e51)

mon: check the new crush map against osdmap.max_osd

Fixes: #11680
Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit 22e6bd6e01d5df3f3e897562597e22ca1737f8c8)

crushtool: enable check against max_id

add an argument "max_id" for "--check-names" to check if any item
has an id greater or equal to given "max_id" in crush map.

Note: edited since we do not have the fix introduced in 46103b2 in
      hammer.

Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit d0658dd3cdf072b2a7c2a1986f8785a697c591ee)

crush/CrushTester: check if any item id is too large

Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit e640d89240017956b8c7411babb86be0f1e2b172)

mon: validate new crush for unknown names

* the "osd tree dump" command enumerates all buckets/osds found in either the
  crush map or the osd map. but the newly set crushmap is not validated for
  the dangling references, so we need to check to see if any item in new crush
  map is referencing unknown type/name when a new crush map is sent to
  monitor, reject it if any.

Fixes: #11680
Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit a955f36a509e5412b1f72632a1a956d99e768e35)

crushtool: add the "--check-names" option

* so one is able to verify that the "ceph osd tree" won't chock on the
  new crush map because of dangling name/type references

Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit d6b46d4c7b722945ce24ac2930381a109b1e3dda)

crush/CrushTester: add check_name_maps() method

* check for dangling bucket name or type names referenced by the
  buckets/items in the crush map.
* also check for the references from Item(0, 0, 0) which does not
  necessarily exist in the crush map under testing. the rationale
  behind this is: the "ceph osd tree" will also print stray OSDs
  whose id is greater or equal to 0. so it would be useful to
  check if the crush map offers the type name indexed by "0"
  (the name of OSDs is always "OSD.{id}", so we don't need to
  look up the name of an OSD item in the crushmap).

Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit b75384d73958faf81d45847a7dfa56f4fa347e6f)

Merge pull request #4667 from SUSE/wip-11611-hammer

ceph.spec: update OpenSUSE BuildRequires

Reviewed-by: Loic Dachary <ldachary@redhat.com>

OSDMonitor: allow addition of cache pool with non-empty snaps with config

We need to be able to allow the version of ceph_test_* from earlier
versions of ceph to continue to work.  This patch also adjusts the
work unit to use a single rados snap to test the condition without
--force-nonempty to ensure that we don't need to be careful about
the config value when running that script.

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

Merge pull request #5123 from theanalyst/wip-11979-hammer

MDSMonitor: handle MDSBeacon messages properly

Reviewed-by: Kefu Chai <kchai@redhat.com>

Always provide summary for non-healthy cluster

This fixes a problem, wherein calamari does not provide
popup drill-downs for warnings or errors, should the summary
be missing.

Calamari gets health info from /api/v1/cluster/$FSID/health.
If the data here has a summary field, this summary is provided
in a popup window:

  /api/v1/cluster/$FSID/health is populated (ultimately) with
  status obtained via librados python bindings from the ceph
  cluster. In the case where there's clock skew, the summary
  field supplied by the ceph cluster is empty.

No summary field, no popup window with more health details.

Signed-off-by: Thorsten Behrens <tbehrens@suse.com>
(cherry picked from commit eaf6e0cf48488fe604d0ef0db164d44948d4e8d4)

Merge pull request #4862 from SUSE/wip-11874-hammer

Bucket header is enclosed by quotes

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>

Merge pull request #4885 from theanalyst/wip-11755-hammer

Object copy bug

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>

Merge pull request #4884 from theanalyst/wip-11722-hammer

Keystone PKI token expiration is not enforced

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>

Merge pull request #4875 from ceph/wip-11770-hammer

librbd: aio calls may block

Reviewed-by: Jason Dillaman <dillaman@redhat.com>
Reviewed-by: Loic Dachary <ldachary@redhat.com>

mon: always reply mdsbeacon

the MDS (Beacon) is always expecting the reply for the mdsbeacon messages from
the lead mon, and it uses the delay as a metric for the laggy-ness of the
Beacon. when it comes to the MDSMonitor on a peon, it will remove the route
session at seeing a reply (route message) from leader, so a reply to
mdsbeacon will stop the peon from resending the mdsbeacon request to the
leader.

if the MDSMonitor re-forwards the unreplied requests after they are
outdated, there are chances that the requests reflecting old and even wrong
state of the MDSs mislead the lead monitor. for example, the MDSs which sent
the outdated messages could be dead.

Fixes: #11590
Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit b3555e9c328633c9e1fbc27d652c004b30535e5b)

mon/MDSMonitor: rename labels to a better name

* s/ignore/reply/
* s/out/ignore/

Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit f00ecb8b3df73ce6337985bc6d43bce5143ee537)

mon: send no_reply() to peon to drop ignored mdsbeacon

so the peon can remove the ignored mdsbeacon request from the
routed_requets at seeing this reply, and hence no longer resend the
request.

Fixes: #11590
Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit 72a37b3a8e145d8522ea67fc14ce2c5510b6852b)

mon: remove unnecessary error handling

msg.get_session() should always return a non-zero pointer in
Monitor.dispatch()

Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit 16e8e2cc82a90c49cd8aa3d0e3acc4694ba659a0)

mon: remove unused variable

* as a side effect, this change silences
  http://tracker.ceph.com/issues/11576

Fixes: #11576
Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit e7b196a4a091c0ea258866559ba06e7ed0cc4247)

Merge pull request #4886 from theanalyst/wip-11737-hammer

Reviewed-by: Greg Farnum <gfarnum@redhat.com>

Merge pull request #5095 from ceph/wip-fix-doc-hammer

doc: fix doc build

Reviewed-by: Loic Dachary <ldachary@redhat.com>

doc: add the corresponding @endcond command for @cond

* they are used to applease asphyxiate, as it
  is not able to handle "enum" sections

Signed-off-by: Kefu Chai <kchai@redhat.com>

doc: remove orphan: directive in ceph-create-keys.rst

* it is used to silence the sphinx warning, but conf.py
  does not like it.

Signed-off-by: Kefu Chai <kchai@redhat.com>

doc: let doxygen ignore src/tracing

Signed-off-by: Kefu Chai <kchai@redhat.com>

Merge pull request #4902 from theanalyst/wip-11908-hammer

Fixes for rados ops with snaps

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

Merge pull request #5069 from dachary/wip-11806-hammer

ceph / ceph-dbg steal ceph-objecstore-tool from ceph-test / ceph-test-dbg

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

debian: ceph-dbg steals ceph-objectstore-tool from ceph-test-dbg (take 2)

968573b8930a7c8485bf53e3a989ce2f7d0a2fff incorrectly backported
6f11fbf41fab10924b1e0e41fcf27864779d4073. It should instead reflect that
ceph-dbg in 0.94.2 and above will break ceph-test-dbg in all versions
prior to 0.94.2.

In other words, 0.94-XXXX being lower than 0.94.1, upgrading from
0.94.1 to 0.94.2 will not notice that ceph-dbg breaks ceph-test-dbg.

$ dpkg --compare-versions 0.94-XXXX lt 0.94.1 && echo yes || echo no
yes
$ dpkg --compare-versions 0.94.2 lt 0.94.1-xxx && echo yes || echo no
no

http://tracker.ceph.com/issues/11806 Fixes: #11806

Signed-off-by: Loic Dachary <loic@dachary.org>

rgw/logrotate.conf: Rename service name

The service name for ceph rados gateway was changed to "ceph-radosgw",
the previous version of service name "radosgw" would cause a failed reload,
and finally make it impossible to write any log data to the log file.

Signed-off-by: wuxingyi <wuxingyi2015@outlook.com>
(cherry picked from commit 9df3f798179481fe8ae6ae873dcb793de7d8f367)

tests: add librbd watch/notify version compatibility test

Fixes: #11405
Backport: hammer
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 41e4cbe032e32762e3a9e8bc3eff8ece19f91a54)

qa/workunits/rbd: add notify_master/slave bootstrap scripts

Backport: hammer
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 124b1d3d2d0dbd685bbd717856d29b316e62e660)

qa/workunits/rbd: add new test_librbd_api workunit

This only tests the public librbd API for use during upgrade tests.

Backport: hammer
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 9039955f42d01044cfcf20c56ca2181e51c317ee)

tests: create librbd API-only integration test suite

The QA client-upgrade test suite requires a librbd test
that is dynamically linked to librbd.  Since the current
ceph_test_librbd includes tests against the non-public API,
it is statically linked against librbd and thus cannot be
used to test a client upgrade scenario.

Backport: hammer
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 6fe94c8cbb924c31c1007e2d0c76f28ce9dbca57)

Increase max files open limit for OSD daemon.

Under heavy load the number of file descriptors opened
by the OSD can go beyond the 64K file limit. This patch
increases the default to 128K.

Signed-off-by: Owen Synge <osynge@suse.com>
(cherry picked from commit ebda4ba1c67172852587e47a8e6fb538809a1b1c)

Makefile: install ceph-post-file keys with mode 600

Otherwise ssh (may) prompt for a password.

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 106a1c3081d02446aa3d8e13865da0c3393bae90)

ceph-post-file: improve check for a source install

Signed-off-by: Joseph McDonald <joseph.mcdonald@alcatel-lucent.com>
(cherry picked from commit ee170eadcdcb4b54d36a7d474558484de9d917eb)

ceph-post-file: behave when sftp doesn't take -i

Fixes: #11836
Signed-off-by: Joseph McDonald <joseph.mcdonald@alcatel-lucent.com>
(cherry picked from commit b84031ed5eaace1222e14d3c4076a3ab1155da96)

packaging: move SuSEfirewall2 templates out of src

Better to have static config like this that needs to be installed in
/etc in a separate subdirectory of the project.

Signed-off-by: Tim Serong <tserong@suse.com>
(cherry picked from commit 70292658d5febb6c5f94af4df9c3e93551772d12)

packaging: add SuSEfirewall2 service files

This adds SuSEfirewall2 service files for Ceph MON, OSD and MDS, for use
on SLES and openSUSE.  The MON template opens port 6789 and the OSD/MDS
template opens the range 6800-7300 as per
http://ceph.com/docs/master/rados/configuration/network-config-ref/

Signed-off-by: Tim Serong <tserong@suse.com>
(cherry picked from commit 77685f5b787c56bcb1c4d9f1e058e25312fa62fe)

Bug fix to ceph systemV compatability script.

Was failing with more than one OSD / MON deamon on a single node.
Fixes suse bugzilla #927862

Signed-off-by: Owen Synge <osynge@suse.com>
(cherry picked from commit dfda3ff8741fcdbac3150456ca7614cf75ef1776)

Fixes to rcceph script

 - only start OSDs if mon daemons are also present
 - adds support for mask and unmask
 - removes support for cluster with non default cluster name,
   as this was very limited and inconsistent
 - Reapplied from a patch as could not cherry-pick
   66cb46c411d874be009c225450eea5021cf1219b from Mon Jan 12
   as this produced issues with src/gmock

Signed-off-by: Owen Synge <osynge@suse.com>
(cherry picked from commit bfa0c4a626fdbb2bf978ccfab783ac428156144b)

Increase max files open limit for OSD daemon.

Under heavy load the number of file descriptors opened
by the OSD can go beyond the 64K file limit. This patch
increases the default to 128K.

Signed-off-by: Owen Synge <osynge@suse.com>
(cherry picked from commit ebda4ba1c67172852587e47a8e6fb538809a1b1c)

qa/workunits/rados/test-upgarde-v9.0.1: fix exclude syntax

It's -, then a list of all exclusions separated by :.  There are just 2.

Signed-off-by: Sage Weil <sage@redhat.com>

qa/workunits/rados/test-upgrade-v9.0.1: skip one more evict test

Signed-off-by: Sage Weil <sage@redhat.com>

qa: add compatibility filtered rados api tests for upgrades

Post-9.0.1, the evict op returns success when an object doesn't exist
in the cache tier. Skip the tests that are incompatible across
versions.

Fixes: #11548
Signed-off-by: Josh Durgin <jdurgin@redhat.com>

Merge pull request #4961 from ceph/wip-11493-hammer

backport 11493 fixes, and test, prevetning ec cache pools

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

OSDMonitor: disallow ec pools as tiers

Fixes: 11650
Signed-off-by: Samuel Just <sjust@redhat.com>
(cherry picked from commit 11b7801bb57cb25cd2d26d58722d49691747725b)

mon: prevent pool with snapshot state from being used as a tier

If we add a pool with snap state as a tier the snap state gets clobbered
by OSDMap::Incremental::propogate_snaps_to_tiers(), and may prevent OSDs
from starting.  Disallow this.

Include a test.

Fixes: #11493
Backport: hammer, giant, firefly
Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit bbec53edf9e585af4e20bbc9ba9057d6fdfda342)

test/librados/tier.cc: destroy and recreate cache pool on every test

Namespaces are not sufficient with the checks for 11493 in the mon.

Signed-off-by: Samuel Just <sjust@redhat.com>
(cherry picked from commit bef09e0cdb274cb1c87335a2af9ee532d14a4596)

Merge pull request #4846 from SUSE/wip-11862-hammer

missing man pages for ceph-create-keys, ceph-disk-*

Reviewed-by: Kefu Chai <kchai@redhat.com>
Reviewed-by: Ken Dreyer <kdreyer@redhat.com>

Merge pull request #4934 from dachary/wip-releases-hammer

doc/release-notes: v0.94.2

doc/release-notes: v0.94.2

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 306345b29c259ab04a58ed5d40f801645485b29d)

Conflicts:
doc/release-notes.rst

tests: verify librbd blocking aio code path

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 4cf41486e9c9e1efcb863960a8f3e0326ffca7e5)

Conflicts:
src/test/librbd/test_librbd.cc: trival resolution

librbd: new rbd_non_blocking_aio config option

Setting this option to false reverts librbd to legacy behavior
where AIO operations could potentially block.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 769cad12716b85d87eacc1069dd9f5c21cad3915)

PendingReleaseNotes: document changes to librbd's aio_read methods

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: AioRequest::send no longer returns a result

The librados calls used by AioRequest::send should always return
zero unless there is a bug.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit c77bce3311ab62892eb8c1d883263ba7ed663b20)

Conflicts:
src/librbd/AioRequest.cc: trivial resolution
src/librbd/AsyncFlattenRequest.cc: trivial resolution

tests: update librbd AIO tests to remove result code

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 948b15eb52fd5d9ce842fa12ee0cecda17353b01)

Conflicts:
src/test/librbd/test_internal.cc: trivial resolution
src/test/librbd/test_librbd.cc: trivial resolution

librbd: internal AIO methods no longer return result

All failures should be returned via the AioCompletion.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 9ab42d613128ab08c688ddbea93df4c95068b9cd)

Conflicts:
src/librbd/AioRequest.cc: trivial resolution
src/librbd/internal.cc: trivial resolution

Throttle: added pending_error method to SimpleThrottle

Allow the client of SimpleThrottle to detect an async error
so that it can exit early.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit b88b88c5df91325fb713c2031a56bffe421268e0)

librbd: add new fail method to AioCompletion

Helper method to handle passing fatal errors generated within
librbd (not from the OSDs) back to the client.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 6d1d0c867855a96bee4c13a0c0a39a0e002ccd12)

librbd: avoid blocking AIO API methods

Enqueue all AIO API methods within the new librbd thread pool to
reduce the possibility of any blocking operations. To maintain
backwards compatibility with the legacy return codes of the API's
AIO methods, it's still possible to block attempting to acquire
the snap_lock.

Fixes: #11056
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 3a7b5e30efdb21aa1a0aeb68a5d02a1ac2a5faf3)

librbd: add task pool / work queue for requests

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit afb896d91f886b647baf38f7ec94cc3739f6d2a9)

Conflicts:
src/librbd/ImageCtx.cc: trivial resolution
src/librbd/ImageCtx.h: trivial resolution

0.94.2