rados/thrash: rejection backfill reservations sometimes while thrashing

rados/singleton-nomsgr: add multi-backfill reservation rejection test

This triggers #7922.

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

Merge pull request #23 from ceph/wip-7771-wusui

FIxed dumpling - emperor upgrade suites

upgrade/dumpling-x: run dumpling version of s3tests

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

FIxed dumpling - emperor upgrade suites

Reformatted many of the yaml files.
Fixed duplication of workunit definitions.
Do not do install.upgrade on clients.
Included branch: definition in final workload files.

Signed-off-by: Warren Usui <warren.usui@inktank.com>

upgrade/dumpling-emperor-x: do not warn about tunables

We do the same for dumpling-x.

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

upgrade/dumpling-emperor-x: ignore 'scrub mismatch' from mon

The encoded full osdmaps can differ due to a series of issues that were
fixed well after emperor.  We also whitelist this for dumpling-x.

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

multimds: test ceph-fuse and kclient

Add client as a variable on the basic collection.  Move the other stuff
(libcephfs plus a weird restart test) to libcephfs collection.

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

Merge pull request #21 from ceph/wip-thrash-affinity

krbd, kcephfs: trash osd primary-affinity

Merge pull request #22 from ceph/wip-fuse-permission

multimds: fuse_default_permissions = 0 for kernel build test

multimds: fuse_default_permissions = 0 for kernel build test

This can reduce the test time becuase it avoids sending getattr
request whenever the kernel checks inode permission.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>

krbd, kcephfs: trash osd primary-affinity

libceph.ko support is in testing and should be in 3.15.  This commit
effectively reverts commit 66ffaa65657b ("kcephfs, krbd: do not thrash
primary-affinity").

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>

rados/monthrash: shorten mon pings so that freezing triggers reconnects

This exercises (tho it doesn't necessarily verify) the mon ping reconnect
code when the thrasher freezes the mons.

See #7888.

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

rgw_pool_type: remove accidentally added empty file

Signed-off-by: Josh Durgin <josh.durgin@inktank.com>

upgrade/dumpling-x/parallel: restart after cuttlefish->dumpling upgrade

Fixes: #7824
Signed-off-by: Sage Weil <sage@inktank.com>

rados/thrash/workloads: make cache-agent-big use an ec base pool

Signed-off-by: Samuel Just <sam.just@inktank.com>

rgw: add erasure coded data pool variant

Signed-off-by: Josh Durgin <josh.durgin@inktank.com>

rgw: set max-entries to 10 for data sync test

This way it tests the agent generating more than max-intries writes to
a bucket.

Signed-off-by: Josh Durgin <josh.durgin@inktank.com>

upgrade/dumpling-x/stress-split: set fastcgi idle timeout to 2 min

30 seconds isn't enough when combined with osd thrashing.

Fixes: #7702
Signed-off-by: Sage Weil <sage@inktank.com>

upgrade/dumpling-x/parallel: test cuttlefish->dumpling starting point

Maybe this will trigger #7519.  And anyway, it's good coverage.

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

distros: add rhel 6.4

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

distros: add wheezy

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

Merge pull request #20 from ceph/wip-distros

symlink all distros facets to a common set of 'supported' targets

distros: add 14.04 to supported list

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

distros: test rhel/centos 6.5 instead of 6.4

symlink all distros facets to a common set of 'supported' targets

This makes it easier to maintain the supported/tested targets.

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

knfs: make it a list

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

knfs: async export

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

rados/monthrash: add pool create/delete tests

This is sufficient to trigger #7736 when combined with mon thrashing and
mon msg delay injection.

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

nkfs: 6h timeout of kernel build

Yes, NFS reexport is slow.  But at the moment appears to be reliable.

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

multimds: fix misc -> fs/misc paths

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

upgrade/dumpling-x/stress-split: tolerate legacy crush tunables

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

Merge pull request #19 from ceph/wip-print

added print task
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>

added print task

upgrade/dumpling-x: more mon scrub whitelist

There are two log messages; whitelist them both

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

knfs: test v3 and v4

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

Merge pull request #18 from ceph/wip-knfs-fsstress

knfs: add fsstress and kernel build

knfs: add fsstress and kernel build

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>

knfs: restructure a bit

Separate cluster buildup from tasks.

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

nfs -> knfs

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

workunits/misc -> fs/misc

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

upgrade/dumpling-x: whitelist mon scrub mismatch

A mixed dumpling/firefly cluster will have each mon independently encode
the full map.  Firefly now puts feature bits in the Incremental so that
we encode the "right" way, but dumpling doesn't have that capability
yet, and we don't plan to backport it.

Fixes: #7704
Signed-off-by: Sage Weil <sage@inktank.com>

upgrade/dumpling-x: upgrade final client node for final test

This means the final workloads are running firefly code.

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

upgrade/dumpling-x: full librados tests after full upgrade

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

multimds: new (separate) suite for multi-mds tests

Eventually we can merge this into the fs suite, but for now it is useful
to segregate the test cases.

Start with a dup of basic and verify sections; add others later.

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

Revert "fs/basic: multimds"

This reverts commit aea501beb744dd6f6ec2b60678cefc7fae6761b0.

Let's put this in a separate subdir for now.  It won't pass on firefly because
the libcephfs changes aren't there.

Merge remote-tracking branch 'gh/firefly'

upgrade/dumpling-x/stress-split: use dumpling workunits

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

Merge pull request #16 from ceph/wip-hint-tests

rados/singleton-nomsgr: add allocation hint test

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

Merge pull request #17 from ceph/wip-final_load_fix

fixed final_load problem

fixed final_load problem

upgrade/dumpling-x: upgrade mon before osd

The MDS needs to see the TMAP2OMAP feature bit in the osdmap.  THe mons
have to encode the feature bits passed by the new osds in order for that
to happen.  Ergo, upgrade mons first.

This is only important for the MDS and for users who want to use erasure
coded pools or tiering without restarting the osds *again* after the mons
get upgraded.

Fixes: #7679
Signed-off-by: Sage Weil <sage@inktank.com>

rgw: add data sync test

Signed-off-by: Josh Durgin <josh.durgin@inktank.com>

rgw: use different keys for different system users

Only one of these was being used before, due to a bug in
teuthology. This is fixed, and trying to create a second user with the
same access key fails.

Signed-off-by: Josh Durgin <josh.durgin@inktank.com>

upgrade/dumpling-x: fix order

Make it match the filename.

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

upgrade/dumpling-x: do not thrash primary-affinity

We can't do it on a mixed cluster

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

stress-split: use dumpling version of rbd.py tests

The firefly version has a test that depends on new behavior.

Fixes: #7643
Signed-off-by: Josh Durgin <josh.durgin@inktank.com>

fs/basic: multimds

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

rados/singleton-nomsgr: add allocation hint test

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>

Remove the mds-mon-osd upgrade sequence

It wasn't actually supported.

Signed-off-by: Zack Cerza <zack.cerza@inktank.com>

Don't warn on legacy crush tunables

Signed-off-by: Zack Cerza <zack.cerza@inktank.com>

upgrade: fix typo in subsection name

It's referred to as workload, not workunit, by the parallel task earlier.

Signed-off-by: Josh Durgin <josh.durgin@inktank.com>

upgrade: fix indentation for final-workload

Signed-off-by: Josh Durgin <josh.durgin@inktank.com>

Revert "The parallel task expects dicts here..."

This reverts commit 79e348320f3695d085693b7ce0c538a9dfdd182b.

The parallel task expects dicts here...

We should fix that at some point.

Signed-off-by: Zack Cerza <zack.cerza@inktank.com>

get test-upgrade-firefly.sh from the dumpling branch

it doesn't exist elsewhere

Revert "Update rados_api_tests.yaml"

This needs to reference dumpling now.

This reverts commit 8e88922581f80000784072e381867f0b5448cd7f.

upgrade/dumpling-x: more indentation

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

upgrade/dumpling-x: fix indentation

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

Merge remote-tracking branch 'gh/firefly'

Merge pull request #14 from ceph/wip-final-workload-wusui

Fix parallel workunit issues.

Fix parallel workunit issues.

Fixed the spacing of workunits inside 2-workload and 4-final-workland
files.  Removed parallel final-workload reference from dumpling.yaml.

Fixes: 7633
Signed-off-by: Warren Usui <warren.usui@inktank.com>

upgrade/dumpling-x: skip rados api tests that fail against firefly OSDs

Fixes: #7584
Signed-off-by: Sage Weil <sage@inktank.com>

Merge pull request #9 from ceph/wip-upgrade-dumpling

upgrade/dumpling: two fixes

Merge pull request #13 from ceph/wip-upgrade

a few upgrade fixes

upgrade/dumpling-x/parallel: do all final workloads

Run all of the final workloads in succession instead of in separate tests.
Since we are not concerned about timing in this case, only that the
cluster is functioning at the end, this cuts down our test matrix by a
factor of 4.

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

upgrade/dumpling-x: do not upgrade client host

If we are running against a mixed cluster for the duration of the upgrade,
we need to make sure we are running the *old* client tests against the
cluster.  Do not upgrade client.0.

Fixes: #7574
Signed-off-by: Sage Weil <sage@inktank.com>

Merge pull request #12 from ceph/wip-final-workload-wusui

Fix dumpling-x upgrade suite.

Fix dumpling-x upgrade suite.

final-workload references were added (replacing second task
reference)

Fixes: 7606
Signed-off-by: Warren Usui <warren.usui@inktank.com>

Removed branch ref to fix http://tracker.ceph.com/issues/7584

Update rados_api_tests.yaml

Removed branch ref to fix http://tracker.ceph.com/issues/7584

Merge pull request #11 from ceph/wip-7581

increase verbosity for OSDs for ceph-deploy tests

increase verbosity for OSDs for ceph-deploy tests

Signed-off-by: Alfredo Deza <alfredo.deza@inktank.com>

Merge pull request #10 from ceph/wip-7485

suites/fs/basic: add mds_creation_retry

Barely-
Reviewed-by: Greg Farnum <greg@inktank.com>

suites/fs/basic: add mds_creation_retry

This invokes the new mds_creation_failure task from
teuthology, ahead of mounting the FS and running the
trivial_sync workload.  It is a regression test
for #7485.

Fixes: #7485
Signed-off-by: John Spray <john.spray@inktank.com>

Add missing fedora yamls.

Running a git add first would have been helpfull...

Run ceph-deploy on fedora19 not old fedora18.

Signed-off-by: Sandon Van Ness <sandon@inktank.com>

Add missing fedora yamls.

Running a git add first would have been helpfull...

Run ceph-deploy on fedora19 not old fedora18.

Signed-off-by: Sandon Van Ness <sandon@inktank.com>

Merge remote-tracking branch 'gh/firefly'

upgrade/dumpling: add recent dumpling point releases

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

upgrade/dumpling: upgrade to latest dumpling, not emperor

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

rados/thrash/workloads: enable copy_from for the ec workloads

Signed-off-by: Samuel Just <sam.just@inktank.com>

Merge pull request #8 from ceph/yuriw

Added new dumpling-emperor-x upgrade suite

removed leftover dirs again

removed one upgrade-sequence step and re-named the rest of steps

removed unfinished stress-split suite from

removed obsolete suite from upgrades dir and added dumpling-emperor-x parallel suite

Merge branch 'master' of github.com:ceph/ceph-qa-suite into yuriw

Merge pull request #7 from ceph/wip-upgrade

upgrade: dumpling-next -> dumpling-x