describe-tests: add options for json or csv output

For simple consumption, use objects with fields in the json output,
and list the headers separately in case the original order is useful.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

describe-tests: add subsuite columns for combo output

This gives more context when looking at larger suites as a whole,
e.g. rbd or rados.

Sort the rows as well, so tests within subsuites, e.g. basic, thrash,
singleton, etc. are grouped together.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

describe-tests: add options to show the combinations of a suite

Match the behavior of teuthology-suite and generate the full product
of all fragments by default. Include the same filtering and subset
options as well. Just print an asii table for now.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

test-describe-tests: remove unnecessary vars

These tests use the class-initialized attributes instead.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

describe-tests: tolerate non-dict files

Some yamls are empty to represent the absence of a setting, and parse
into None.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

teuthology-describe-tests: change input format to yaml

Rather than using comments, read description metadata from a yaml
'description' section. Make it a list so that teuthology-suite
aggregates the descriptions in order, and allow multiple fields by
making the list contain a dict.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

teuthology-describe-tests: rename from teuthology-tree

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

teuthology-tree: show facets for each file by default

Suggested by Loic, this provides more context without
needing to remember or look back at the parent directory.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

teuthology-tree: remove whitespace from metadata

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

teuthology-tree: add unit tests

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

fake_fs: add fake open() support

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

test: move fs fakes to a common file

This will be useful for other tests shortly.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

teuthology-tree: remove cruft from usage

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

teuthology-tree: support multiple lines with the same prefix

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

teuthology-tree: output in a table

This is easier to read since it aligns the metadata.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

scripts: teuthology-tree tool to examine a qa suite

An initial prototype, with basic formatting printing out the directory
tree and comments from the yaml fragments.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>

Merge pull request #742 from ceph/wip-os-version

Remote: only submit X.Y for os_version

Copying comment for future documentation:

The opsys.OS class passes the operating system's version according to
the actual host; in some cases e.g. CentOS is reporting things like
'7.1.1503'. In these cases, let's make
Remote.inventory_info['os_version'] only use the '7.1' - since we don't
care about the rest. This will allow us to continue scheduling against
CentOS smoothly.

Reviewed-by: Dan Mick <dmick@redhat.com>

Remote: only submit X.Y for os_version

The opsys.OS class passes the operating system's version according to
the actual host; in some cases e.g. CentOS is reporting things like
'7.1.1503'. In these cases, let's make
Remote.inventory_info['os_version'] only use the '7.1' - since we don't
care about the rest. This will allow us to continue scheduling against
CentOS smoothly.

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

Merge pull request #727 from dachary/wip-nuke-stale-openstack

nuke: --stale-openstack must cope with disappearing volumes / instances

nuke: --stale-openstack must cope with disappearing volumes / instances

"openstack server/volume show" typically takes a few seconds to
complete due to the provider API slowness/throttling. It is possible that
a volume/instance found with "openstack server list" no longer exists when
running "openstack server show" to get more information about it.

Signed-off-by: Loic Dachary <loic@dachary.org>

Merge pull request #741 from ceph/wip-apt-mirror

Stop hardcoding apt-mirror's IP

Stop hardcoding apt-mirror's IP

We will point to apt-mirror.sepia.ceph.com, which is
publicly-accessible.

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

Merge pull request #740 from smithfarm/wip-bootstrap-opensuse

bootstrap: fixes for openSUSE

bootstrap: fixes for openSUSE

When a package is found not to exist, we need to also check if there
is another package installed that provides a capability of the same
name.

Also, the python-devel package must be installed, or the pip phase
fails.

http://tracker.ceph.com/issues/14002 Fixes: #14002

Signed-off-by: Nathan Cutler <ncutler@suse.com>

Merge pull request #737 from ceph/wip-config-port

Use the newer config API

nuke: Use newer config API

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

nuke: Linter fixes

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

misc: Use newer config API

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

Implement YamlConfig.get()

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

Drop misc.read_config()

Replace with config.set_config_attr(). This is mainly for
backward-compatibility in tasks contained in ceph-qa-suite.

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

Merge pull request #723 from dachary/wip-nuke-stale-openstack

nuke: cleanup stale OpenStack resources

Merge pull request #709 from ceph/wip-epel-remove

Remove epel packages during downstream testing if present on the node

Merge pull request #710 from dachary/wip-internal-buildpackages

internal: the buildpackages task must come first

Merge pull request #719 from dachary/wip-openstack-parallel-delete

openstack: run 20 server/deletion at a time

Merge pull request #720 from dachary/wip-openstack-instance-size

openstack: teuthology for 100 or more need 32GB

Merge pull request #722 from dachary/wip-openstack-parallel

openstack: run one job at a time by default

nuke: cleanup stale OpenStack resources

Implement the --stale-openstack action to cleanup OpenStack resources
that are known to be stale because:

* instances or volumes have been created a long time ago
* instances or volumes with REMOVE-ME in the name
* nodes that are locked but did not get an instance

Only the instances and volumes known to belong to the teuthology cluster
because they have ownedby=IP in the name are considered, where IP is the
IP address of the machine running the teuthology clusters.

Signed-off-by: Loic Dachary <loic@dachary.org>

lock: locked_since_seconds converts locked_since

The node locked_since property is converted by locked_since_seconds
into the number of seconds since the present time, which is convenient
to compare with configured delays.

Signed-off-by: Loic Dachary <loic@dachary.org>

openstack: introduce OpenStackInstance

To reduce the number of redundant calls and reduce the code complexity.

Signed-off-by: Loic Dachary <loic@dachary.org>

internal: the buildpackages task must come first

Otherwise the machines will be locked and remain idle while the packages
are building. When there are 20 jobs requiring a package, it means an
average of 40 machines locked idle while the packages are building.

Signed-off-by: Loic Dachary <loic@dachary.org>

openstack: run one job at a time by default

Most jobs in rados/thrash allocate six volumes. The default tenant on
OVH has ten volumes, that's just not enough for two jobs to run in parallel.

Signed-off-by: Loic Dachary <loic@dachary.org>

openstack: teuthology for 100 or more need 32GB

Signed-off-by: Loic Dachary <loic@dachary.org>

openstack: run 20 server/deletion at a time

Signed-off-by: Loic Dachary <loic@dachary.org>

Merge pull request #679 from ceph/wip-unpin-gevent

Unpin gevent

Reviewed-by: Loic Dachary <ldachary@redhat.com>

libevent -> libev for newer gevent

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

Unpin gevent version

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

Merge pull request #716 from ceph/wip-fix-setuptools

Work around setuptools 18.6 regression

Reviewed-by: Dan Mick <dmick@redhat.com>

Work around setuptools 18.6 regression

See
https://bitbucket.org/pypa/setuptools/issues/464/typeerror-in-install_wrapper_scripts

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

Merge pull request #676 from dachary/wip-openstack-timeout

openstack: increase ssh timeout to 240 (part 2)

Merge pull request #696 from dachary/wip-openstack-only-ipv4

openstack: only consider IPv4 subnets

Merge pull request #695 from dachary/wip-openstack-resolver

openstack: add 8.8.8.8 as a last resort resolver

Merge pull request #697 from dachary/wip-openstack-neutron-private-ip

openstack: try to get private IP from neutron

Merge pull request #705 from dachary/wip-glance-raw

openstack: support glance disk-format raw

Merge pull request #706 from dachary/wip-dreamhost-provider

openstack: set dreamhost if OS_AUTH_URL has dream.io

Merge pull request #707 from dachary/wip-openstack-delete-wait

openstack: wait for server delete to complete

Merge pull request #693 from dachary/wip-no-roles

run: do not lock machines if there are no roles

tests: teuthology-* --invalid-option is a sure fail

Providing the INVALID argument may succeed, depending on how the script
interprets arguments.

Signed-off-by: Loic Dachary <loic@dachary.org>

Remove epel packages during downstream testing if present on the node

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>

Merge pull request #708 from ceph/wip-tolerate-giterror

Be more tolerant of GitErrors in repo_utils

fetch_repo(): Call rmtree() when all attempts fail

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

Don't call rmtree() on GitErrors

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

openstack: wait for server delete to complete

Signed-off-by: Loic Dachary <ldachary@redhat.com>

run: do not lock machines if there are no roles

The worker runs the job, even if there are no roles: it is legitimate to
run a job that does not need machines.

Run only locks machines if there are roles and init_tasks is only
populated with tasks that make sense depending on the presence of roles
or not.

Move the internal task added in suite.py to run.py : there is no reason
to not have them with teuthology-run or teuthology-schedule (ansible +
clock).

Signed-off-by: Loic Dachary <ldachary@redhat.com>

openstack: set dreamhost if OS_AUTH_URL has dream.io

Signed-off-by: Loic Dachary <ldachary@redhat.com>

openstack: support glance disk-format raw

For providers such as dreamcompute that only support the glance image
format raw, convert the qcow2 image to raw prior to uploading.

Signed-off-by: Loic Dachary <ldachary@redhat.com>

openstack: add 8.8.8.8 as a last resort resolver

Signed-off-by: Loic Dachary <ldachary@redhat.com>

openstack: only consider IPv4 subnets

Signed-off-by: Loic Dachary <ldachary@redhat.com>

openstack: try to get private IP from neutron

Signed-off-by: Loic Dachary <ldachary@redhat.com>

ceph.conf: filstore {ondisk,apply} finisher threads = 3

Exercise new finisher sharding in qa.

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

Merge pull request #698 from dachary/wip-openstack-instance-creation-poll

openstack: poll every 30 seconds instead of 2

Merge pull request #703 from dachary/wip-os-client-config

Unpin os-client-config

Unpin os-client-config

The bug found in 1.10.1 was fixed and version 1.10.2 is now current.

https://bugs.launchpad.net/os-client-config/+bug/1513919

Signed-off-by: Loic Dachary <loic@dachary.org>

Merge pull request #702 from ceph/wip-empty-roles

Run jobs even if 'roles' is empty

This was the original behavior pre-181256f3, and still seems better; the job will fail, but
this way worker won't keep it behind

Reviewd-by: Dan Mick <dmick@redhat.com>

Run jobs even if 'roles' is empty

They will fail immediately, but at least we'll know about them.

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

Merge pull request #701 from ceph/wip-suite-tests

TestSuiteMain.test_main(): don't hit the network

TestSuiteMain.test_main(): don't hit the network

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

Merge pull request #700 from ceph/wip-pin-occ

Pin os-client-config to 1.9.0

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Pin os-client-config to 1.9.0

This is to fix 'openstack image list --json' acting as if --debug were
passed. This was breaking things.

https://bugs.launchpad.net/os-client-config/+bug/1513919

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

Merge pull request #649 from dachary/wip-install-doc

install: document tag, branch and sha1 precedence

Merge pull request #699 from ceph/wip-results

teuthology-results: Be aware of all job statuses

More consistent spacing between sections

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

Format durations as HH:MM:SS

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

Increase polling interval

When we used the filesystem, we polled for a run's status every 10s;
that was more aggressive than necessary. Increase to 60s to avoid
overloading paddles.

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

Refactor teuthology.results to remove copypasta

And also stop mixing tenses in the email subject/body.

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

Handle all possible job states

Instead of lumping jobs into a made-up 'hung' state, express that they
are in one of the following: dead, running, waiting, queued

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

Update unit tests for teuthology.results

This commit updates the existing unit tests to match the new
paddles-based behavior.

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

Use paddles instead of the filesystem

We do still need the filesystem to generate coverage reports.

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

Provide a default for --timeout

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

openstack: poll every 30 seconds instead of 2

No cloud provider boots an instance within 30 seconds, no need to poll
aggressively. Also wait more than a total of 1200 seconds (3000 seconds)
because some providers may take more than 20 minutes to complete the
boot sequence.

Signed-off-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #692 from dachary/wip-upload-order

run: upload must happen after archive

Merge pull request #691 from dachary/wip-suite-sha1

suite: add the ceph-qa-suite hash to the job config

Merge pull request #689 from dachary/wip-openstack-user-data

openstack: get user-data path relative to the sources

Merge pull request #688 from dachary/wip-tests-openstack-resources

openstack: tests: use new resources hint style

Merge pull request #690 from dachary/wip-openstack-sha1

openstack: disable teuthology-suite sha1 verification

run: upload must happen after archive

The unwinding order is in reverse: internal.archive must be after
archive_upload so that it actually happens *before*.

Signed-off-by: Loic Dachary <ldachary@redhat.com>

suite: add the ceph-qa-suite hash to the job config

To be able to re-run a job we need:

* the ceph repository and hash
* the ceph-qa-suite repository and hash
* the job description

When reading the job log from the archive directory, we currently only
have the ceph-qa-suite branch name. If and when the branch name moves
on, the job description may cease to be valid and re-running the job
may no longer be possible.

Signed-off-by: Loic Dachary <ldachary@redhat.com>

openstack: disable teuthology-suite sha1 verification

By default teuthology-openstack builds packages dynamically and does not
rely on gitbuilder.ceph.com

Signed-off-by: Loic Dachary <ldachary@redhat.com>

openstack: get user-data path relative to the sources

The user-data files are in the sources, in the teuthology/openstack
directory. By default the template path being used is
teuthology/openstack/openstack-{os_type}-{os_version}-user-data.txt and
assumes teuthology-suite / teuthology-openstack is run from the root of
the source directory.

If the user-data path is relative, make it absolute based on the
directory in which the openstack module is found.

http://tracker.ceph.com/issues/13416 Fixes: #13416

Signed-off-by: Loic Dachary <ldachary@redhat.com>

openstack: tests: use new resources hint style

The memory cannot be lowered by the hint: test that it increases to 10GB
instead of testing that it was lowered under a given limit.

Signed-off-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #637 from dachary/wip-suite-sha1

suite: add config.suite_verify_ceph_hash when no gitbuilder