firewalld in RHEL 7.3 contains service definitions for ceph's ports, so
we can simply use "--add-service=ceph-mon" or "--add-service=ceph"
rather than having to remember specific port numbers.
Break out two examples for both MONs and OSDs/MDSs.
Add the "firewall-cmd --reload" step to ensure the --permanent configs
take effect immediately.
Signed-off-by: Ken Dreyer <kdreyer@redhat.com>
we have
2017-02-04T16:15:46.090 INFO:tasks.workunit.client.0.mira032.stdout:error in 22088
2017-02-04T16:15:46.092 INFO:tasks.workunit.client.0.mira032.stderr:bash: line 1: 22092 Alarm clock ceph_test_rados_api_aio 2>&1
2017-02-04T16:15:46.096 INFO:tasks.workunit.client.0.mira032.stderr: 22093 Done | tee ceph_test_rados_api_aio.log
2017-02-04T16:15:46.099 INFO:tasks.workunit.client.0.mira032.stderr: 22094 Done | sed "s/^/ api_aio: /"
2017-02-04T16:15:46.102 INFO:tasks.workunit.client.0.mira032.stderr:+
if a unittest in rados/test.sh fails in teuthology.log, but it would
be desirable to have the failed test name in the line of "error in
22088".
Signed-off-by: Kefu Chai <kchai@redhat.com>
In the case of recovery pull using sub op, and if the prep push on
remote peers fail for some reasons, recovery_info.soid is not set
in prep_push_op_blank. Use it on the primary would lead to an
assertion failure.
Signed-off-by: Zhiqiang Wang <zhiqiang@xsky.com>
the version command is handled by HelpHook, which does not print newline
at the end of the returned JSON. so the output of "status" subcommand
looks like:
=== osd.1 ===
osd.1: running {"version":"Development"}=== osd.2 ===
osd.2: not running.
with this change, a newline is added to the end of version.
also use the "daemon" subcommand to talk to the asok, let the "ceph" cli
to figure out the asok path.
Signed-off-by: Kefu Chai <kchai@redhat.com>
tests: osd-scrub-repair.sh disable scrub backoff in test
Reviewed-by: David Zafman <dzafman@redhat.com>
Reviewed-by: Loic Dachary <ldachary@redhat.com>
there is good chance that the newly scheduled scrub fail to pass the
flip-coin check: the probability of scheduling a new scrub 1/3 every
time we flip the coin. so, instead hardwiring the probability to 1/3,
a new option named "osd_scrub_backoff_ratio" is introduced. so we can
override it when performing test.
Signed-off-by: Kefu Chai <kchai@redhat.com>
Update all calls to Messenger::create()
take 'ms_public_type' or 'ms_cluster_type' instead of 'ms_type'
issue: 971661
Change-Id: Ib234daf393d1838ef596dde35a6d83c1aeecaf10
Signed-off-by: Sarit Zubakov <saritz@mellanox.com>
FileStore holds a number of recources like op thread pool and work
queue, key/value DB threads, etc. These should be properly stopped
(released) before exiting to avoid segfaults on exit.
Note: more code paths (mkfs, dump_journal, etc) need similar fixes,
these will be submitted as separate patches.
Fixes: http://tracker.ceph.com/issues/18820
Signed-off-by: Alexey Sheplyakov <asheplyakov@mirantis.com>
We use m_work_queue of type ContextWQ in handle_update function but we
do not include common/WorkQueue.h that defines ContextWQ. This results
in dereference of an incomplete type and causes build error in latest
Fedora rawhide (future 26).
Fixes: http://tracker.ceph.com/issues/18862
Signed-off-by: Boris Ranto <branto@redhat.com>
* support xenial
* support "rados" cli debugging
* do not include the ending "'" in the prog name sending to gdb
Signed-off-by: Kefu Chai <kchai@redhat.com>