Consider the following workload:
writeA(0, 4096)
writeB(0, 512).
pwl can makre sure writeA persist to cache before writeB.
But when flush to osd, it use async-read to read data from cache and in
the callback function they issue write to osd.
So although we by order issue aio-read(4096), aio-read(512). But we
can't make sure the return order.
If aio-read(512) firstly return, the write order to next layer is
writeB(0, 512)
writeA(0, 4096).
This is wrong from the user point.
To avoid this occur, we should firstly read all data from cache. And
then send write by order.
Fiexs: https://tracker.ceph.com/issues/52511
Tested-by: Feng Hualong <hualong.feng@intel.com>
Signed-off-by: Jianpeng Ma <jianpeng.ma@intel.com>
mgr/dashboard: all pyfakefs must be pinned on same version
Reviewed-by: Ernesto Puerta <epuertat@redhat.com>
Reviewed-by: ljflores <NOT@FOUND>
Reviewed-by: Nizamudeen A <nia@redhat.com>
Reviewed-by: rishabh-d-dave <NOT@FOUND>
Without this patch execution of install-deps.sh fails with an error.
Fixes: https://tracker.ceph.com/issues/53088
Signed-off-by: Rishabh Dave <ridave@redhat.com>
This reverts commit 78d6ac0d35, reversing
changes made to 9a8de97f7a.
This broke cephadm (infers config when it should not)
Signed-off-by: Sage Weil <sage@newdream.net>
doc/releases/pacific.rst: add note about data corruption bug
Reviewed-by: Igor Fedotov <igor.fedotov@croit.io>
Reviewed-by: Laura Flores <lflores@redhat.com>
msg/async: support disabling data crc for protocol v2
Reviewed-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
Reviewed-by: Chunmei Liu <chunmei.liu@intel.com>
Reviewed-by: Yingxin Cheng <yingxin.cheng@intel.com>
To fix failure like
Failure Reason:
Command failed on smithi085 with status 1: 'sudo yum -y install ceph-volume'
Signed-off-by: Neha Ojha <nojha@redhat.com>
* refs/pull/43583/head:
changes to logging code requested by Adam
remove allocation file after rollback to real-fm
bug-fix close db when umount failed to store allocation file
Bug-Fix: When restoring allocation from file use a temp allocator and only copy the allocation to the shared-allocator after the file was verified and all extents were cleared
Reviewed-by: Adam Kupczyk <akucpzyk@redhat.com>
* refs/pull/43645/head:
os/bluestore: Disable compaction then no-column-b is storing allocations to bluefs file
Reviewed-by: Igor Fedotov <ifedotov@suse.com>
Reviewed-by: Gabriel Benhanokh <gbenhano@redhat.com>
Put the vault token file in a location that ceph can read.
Make it readable only by ceph.
On rhel8 (and indeed, any vanilla rhel machine), $HOME is liable to be
mode 700. This means the ceph user can't read things in that user's
directory. This causes radosgw to emit the confusing message "ERROR:
Vault token file ... not found" even though the teuthology log will
plainly show it was created and made readable by ceph.
Fixes: http://tracker.ceph.com/issues/51539
Signed-off-by: Marcus Watts <mwatts@redhat.com>
Warn users upgrading their clusters from nautilus/octopus to pacific about
https://tracker.ceph.com/issues/53062
Signed-off-by: Neha Ojha <nojha@redhat.com>
doc/cephadm: improve "Deploying OSDs on multiple similar hosts"
Reviewed-by: Joao Eduardo Luis <joao@suse.com>
Reviewed-by: Michael Fritch <mfritch@suse.com>
doc/cephadm: note that replacment osd needs to be on same host as removed osd
Reviewed-by: Adam King <adking@redhat.com>
Reviewed-by: Michael Fritch <mfritch@suse.com>
The upgrade process can get stuck if an MDS crashes. This should be rare
when straddling v16.2.5 where the compatset of the file system inherits
the FSMap "default". The MDS from pre-v16.2.5 do not yet share a
compatset with the mons so the mons will do no promotions, causing
upgrade task to get stuck.
Fixes: https://tracker.ceph.com/issues/53074
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
