Commit Graph

48 Commits

Author SHA1 Message Date
Sage Weil
01924a22d4 disable rbd thrash workload, #2174 2012-03-16 13:28:44 -07:00
Sage Weil
b4572351a9 Revert "disable rbd thrash workload, #2174"
This reverts commit 1bec416c7c.

Fixed with #2174
2012-03-15 10:32:39 -07:00
Sage Weil
1bec416c7c disable rbd thrash workload, #2174 2012-03-14 15:51:51 -07:00
Sage Weil
b90354dbab thrash: put client on separate machine from osds
This allows us to run kenrel clients (kclient, rbd) against the thrashing
cluster.
2012-03-13 10:49:33 -07:00
Sage Weil
096427d589 remove dup ceph tasks from new thrash workloads 2012-03-12 15:22:17 -07:00
Sage Weil
51d817fe57 Revert "disable s3tests on valgrind/lockdep until #2103 is fixed"
This reverts commit 9f757ca951.
2012-03-11 21:32:45 -07:00
Sage Weil
af445189c2 add rbd, kclient workloads to regression thrash collection
This will get us some kernel osd_client osd restart coverage.
2012-03-11 21:28:45 -07:00
Sage Weil
b84897e56f use dbench workunit, not the autotest one
The autotest one uses an old tarball that doesn't build.  Workunit assumes
the dbench package is installed.
2012-03-10 20:01:57 -08:00
Sage Weil
9f757ca951 disable s3tests on valgrind/lockdep until #2103 is fixed 2012-03-01 22:04:19 -08:00
Josh Durgin
b2bbede826 dump-stuck: set pg stuck threshold to match test 2012-02-29 15:45:25 -08:00
Sage Weil
722af1a4dd no peer as part of lost_unfound 2012-02-27 14:52:35 -08:00
Sage Weil
9afafdf164 move peer to separate test for now 2012-02-26 17:09:41 -08:00
Sage Weil
6295578f16 lost_unfound: do peer after, until wait_for_clean propagates last_epoch_started
The peer task does wait_for_clean, and then lost_unfound immediately marks
something down.  But the PGs become clean before the replica last_epoch_started
is moved forward in time, which means they block waiting for the now down
OSD.  Needlessly.

Until we fix this, just do the peer test after.
2012-02-25 21:35:31 -08:00
Sage Weil
d944e7ee1e fix lockdep.yaml conf syntax 2012-02-24 21:39:55 -08:00
Sage Weil
5d5a022cc1 run radosgw through valgrind for s3tests 2012-02-24 15:20:00 -08:00
Sage Weil
c9c1a4ab42 do peer test along with lost_unfound 2012-02-24 15:04:27 -08:00
Sage Weil
fc531a9177 rename valgrind -> verify, add in runs under lockdep 2012-02-24 12:49:33 -08:00
Josh Durgin
52a52cf4f2 Add test for 'ceph pg dump_stuck' 2012-02-21 16:21:05 -08:00
Sage Weil
df5f5738e6 add valgrind collection to regression suite
Run a smaller set of tests with valgrind on the mon, osd, and mds.

Valgrind is currently ignoring leaks, but this will pick up use-after-free
and similar badness.
2012-02-21 10:02:44 -08:00
Sage Weil
84bd876cd4 cfuse -> ceph-fuse 2012-02-20 12:49:35 -08:00
Sage Weil
b5668cf64b thrashing: whitelist 'objects unfound and apparently lost' message
This can happen when we mark OSDs down... if the objects are found when
the osds come back up then we're fine.  if not, it won't go clean, and the
test will fail for that reason.
2012-02-18 13:56:47 -08:00
Sage Weil
bc0e4068be add regression/multifs collection; run rgw tests under both xfs and btrfs 2012-02-14 21:49:26 -08:00
Sage Weil
3fbb5714cc rename fs files 2012-02-14 08:58:30 -08:00
Sage Weil
10a94d2b40 regression/thrash on xfs and btrfs both 2012-02-13 16:45:04 -08:00
Sage Weil
04f3e44514 btrfs: 1 -> fs: btrfs 2012-02-13 15:29:52 -08:00
Sage Weil
4fad1317da add snap thrashing covering a small number of objects
The snaps-many-objects has a relatively low density of ops-per-object. This
hammers on a small number of them and does a better job of validating the
correctness wrt snaps.
2012-02-11 13:40:44 -08:00
Sage Weil
e841f9c713 move snap thrashing back into regression suite 2012-02-11 13:39:46 -08:00
Sage Weil
6028b36376 move kclient_workunit_suites_blogbench.yaml to stress suite
This is consistently failing due to an mds/kclient interaction.
2012-02-10 16:40:03 -08:00
Sage Weil
0b68dbca9a add backfill test 2012-01-31 16:37:57 -08:00
Sage Weil
9da01185a4 make 6-osd-2-machine simpler... single monitor 2012-01-28 21:11:32 -08:00
Josh Durgin
06c8fdc9e6 regression: add admin socket test for objecter requests. 2012-01-27 18:08:45 -08:00
Sage Weil
fe2834f603 remove snap thrashing from regression suite for time being 2012-01-25 14:04:04 -08:00
Samuel Just
5bb9a9d615 Add small cluster thrashing tasks
Signed-off-by: Samuel Just <samuel.just@dreamhost.com>
2012-01-17 15:08:02 -08:00
Sage Weil
b2c07d8a60 add simple thrash workload to regression suite 2012-01-16 15:09:29 -08:00
Sage Weil
8e126db1da mon.0 -> mon.a 2012-01-16 11:08:34 -08:00
Sage Weil
43da161d2b mds.0 -> mds.a 2012-01-16 11:08:19 -08:00
Yehuda Sadeh
2317b9ae91 add rgw readwrite and roundtrip tasks 2012-01-10 15:30:53 -08:00
Sage Weil
fbf791213b do not put monitors on the same nodes as clients
Otherwise, for kernel clients (rbd or kclient), ceph-mon can cause a deadlock when it calls sync(2).
2012-01-07 10:16:39 -08:00
Sage Weil
cfeaef451b move multimon failure thrashing tests into regression
We need to test these nightly.
2012-01-06 15:08:01 -08:00
Sage Weil
4da96ff35b rados load-gen workunits 2011-12-13 08:28:33 -08:00
Samuel Just
6b8588b781 Use btrfs for regression tests
Some of the tests (particularly the s3 tests) use very long filenames
which trigger bugs related to ext4 xattr handling.

Signed-off-by: Samuel Just <samuel.just@dreamhost.com>
2011-12-08 15:35:16 -08:00
Josh Durgin
2918b5014c Move kclient multiple_rsync workunit to stress collection.
Bug #1760 keeps being triggered by this.
2011-11-29 16:20:55 -08:00
Sage Weil
24ee09b05c Revert "more logs (yuck) for #1682"
This reverts commit ea00114f08.
2011-11-21 21:58:13 -08:00
Sage Weil
ea00114f08 more logs (yuck) for #1682 2011-11-20 15:24:17 -08:00
Sage Weil
dcab329b5f fix conf thinko
'int' object has no attribute 'iteritems'
2011-11-19 19:28:26 -08:00
Sage Weil
a08e7f126a regression/basic/tasks/kclient_workunit_misc: turn on mds log
Hopefully will catch #1682
2011-11-19 13:56:17 -08:00
Sage Weil
13c98df920 regression/basic/tasks/cfuse_dbench: turn up client debugging
Hopefully we'll hit #1737...
2011-11-19 13:45:28 -08:00
Josh Durgin
94100ad072 Move collections into separate suites
For now, there are just two suites:
  * regression - tests that should always pass
  * stress     - tests that have problems for one reason or another
2011-11-17 17:57:57 -08:00