For the new read-based bench tests, flushing prior to the start of the test
will result in the exclusive lock being acquired and the object map being
utilized.
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
Our progress on a backfill target should only count toward degraded if
the target is needed in order for us to reach the target pool size. If
we have remapped other complete copies, we can backfill and not be degraded
at all. And we may be some combination of the two if there are multiple
backfill targets; use the target(s) with the most objects.
Signed-off-by: Sage Weil <sage@redhat.com>
buffer: avoid changing bufferlist ABI by removing new _mempool field
Reviewed-by: Kefu Chai <kchai@redhat.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
The OSD subops were replaced with replication ops (i.e., MSG_OSD_REPOP
and MSG_OSD_REPOPREPLY) in jewel. mClock incorrectly used them as an
op class rather than the newer replication ops. This fixes that and
places eight ops into the repop class -- MSG_OSD_REPOP,
MSG_OSD_REPOPREPLY, MSG_OSD_PG_UPDATE_LOG_MISSING,
MSG_OSD_PG_UPDATE_LOG_MISSING_REPLY, MSG_OSD_EC_WRITE,
MSG_OSD_EC_WRITE_REPLY, MSG_OSD_EC_READ, and MSG_OSD_EC_READ_REPLY.
This alo removes dupliated code related to op classes in
mClockOpClassQueue and mClockClientQueue and combines into
mClockOpClassSupport.
The three configuration options
osd_op_queue_mclock_osd_subop_{res,wgt,lim} have also been replaced
with osd_op_queue_mclock_osd_rep_op_{res,wgt,lim}.
This also removes OpQueueItem::OpQueueable::op_type_t::osd_subop since
it's not a category core ceph recognizes and is instead it's part of
an mclock view of ops.
Signed-off-by: J. Eric Ivancich <ivancich@redhat.com>
This is a common and recommended step that users should take once they
have clients that support straw2 (hammer!). Let's make it easy.
This may result in a small amount of data movement, but usually it is
very little.
Signed-off-by: Sage Weil <sage@redhat.com>
d3ac8d18 moves ceph-client-debug from ceph-test to ceph-base without
updating the package relationships between the two involved packages.
which results in:
dpkg: error processing archive /var/cache/apt/archives/ceph-test_12.2.1-241-g43e027b-1trusty_amd64.deb (--unpack):
trying to overwrite '/usr/bin/ceph-client-debug', which is also in package ceph-base 10.2.10-14-gcbaddae-1trusty
dpkg-deb: error: subprocess paste was killed by signal (Broken pipe)
Signed-off-by: Kefu Chai <kchai@redhat.com>
we have issues when running upgrade tests:
dpkg: error processing archive /var/cache/apt/archives/ceph-osd_13.0.0-2201-g6cc0b41-1trusty_amd64.deb (--unpack):
trying to overwrite '/usr/bin/ceph-osdomap-tool', which is also in package ceph-test 10.2.10-14-gcbaddae-1trusty
in 40caf6a6, we moves some tools from ceph-test out into ceph-osd,
ceph-mon and ceph-base respectively. but didn't update the relationships
between these packages accordingly. this causes the upgrade failure.
see https://www.debian.org/doc/debian-policy/#document-ch-relationships
for more details on "Breaks" and "Conflicts".
the reason why the package version to be replaced/conflicted is 12.2.2
is that: i assume that this change will be backported to luminous, and
the next release of it will be 12.2.2 .
Signed-off-by: Kefu Chai <kchai@redhat.com>
sizeof(data_off) in ceph_msg_header is uint16_t, so -1 will be 65535. When peer
side receive MOSDRepOp message, it will allocate extra 1 byte buffer to store
nonexisted buffer.
Signed-off-by: Haomai Wang <haomai@xsky.com>
Reset admin_socket raw value if it is defined in conffile.
Just in case it used metavarirables (e.g, $pid) which could be expanded
again to the correct value in child process.
Signed-off-by: Zhi Zhang <zhangz.david@outlook.com>
Testing BlueStore against both HDDs and OSDs with fully allocated
and sparse-allocated objects shows a performance improvement with
sparse-read between 32K and 64K.
Fixes: http://tracker.ceph.com/issues/21849
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
osdc/Objecter: skip sparse-read result decode if bufferlist is empty
Reviewed-by: Greg Farnum <gfarnum@redhat.com>
Reviewed-by: Sage Weil <sage@redhat.com>
If we are reencoding an incremental, and it embeds a crush map, we need
to reencode that in a compatible way too. This is especially true now
because we have the compat crush weight-sets. Otherwise, a client may
learn the crush map through an incremental but not understand choose_args,
and not see/understand the alternate weight set. It will send requests
to the wrong OSDs where they will just get dropped.
Fixes: http://tracker.ceph.com/issues/21882
Signed-off-by: Sage Weil <sage@redhat.com>
