Add a work around for Bug #1964117. To ensure that this is only
a temporary measure the fix is tied to specific OpenStack releases
and package versions.
This change drops the use of AssertionError in favor of
LoadBalancerUnexpectedState that was introduced by commit acaeb62 which
stopped retrying on AssertionError.
A missing 's' on importing the zaza.openstack.utilities.exceptions in
the octavia tests are causing gate tests to fail. This fixes the import
to point at the right module.
Fixes#721
Signed-off-by: Billy Olsen <billy.olsen@gmail.com>
* Stop retrying if LB provisioning status reached ERROR
The method `wait_for_lb_resource()` retries for 15m while the load
balancer could have reached to ERROR during the provisioning in the
first few minutes, this approach makes the testing take longer for no
reason.
This change makes the ERROR state in provisioning_status final and
abort raising a ValueError() exception.
More details of the provisioning_status possible states can be found at:
https://docs.openstack.org/api-ref/load-balancer/v2/#provisioning-status-codes
* Use LoadBalancerUnexpectedState and LoadBalancerUnrecoverableError.
Drop the (re-)use of AssertionError and ValueError to identify when a
load balancer status is in a state where the test needs to retry or
break and fail respectively.
This change introduces 2 new exceptions to be explicit of what the code
is trying to do.
- LoadBalancerUnexpectedState is raised when the status of the load
balancer is in a state different from the one requested by the caller,
but said state can be considered as transitory.
- LoadBalancerUnrecoverableError is raised when the status of the load
balancer is in ERROR state and said state is final for the
proviniong_status property, hence retrying only delays the failure.
test_800_remove_leader is getting a reference to the mysql lead
unit, removing the unit and then attempting to retrieve the ip.
This used to work but no longer does, perhaps due to changes
in the way zaza interfacts with libjuju. This change stores
the lead ip address before removing the unit.
* Make run_package_change_test more robust
The run_package_change_test disables auto restarts and then
simulates a packages update. It then checks the service has been
marked as in need of restarting. However bug #1955498 shows that
if a new service is introduced that restarts a disabled service
then this is not picked up by the tests.
This change queries the last active time of a service proir to
the package being updated. If this changes after the package
reconfiguration then the service has incorrectly been restarted.
Depends-On: https://github.com/openstack-charmers/zaza/pull/478
* Log success too
There are situations where the DNS infrastructure hasn't been stable and
many CI jobs have failed with the error:
urllib.error.URLError: <urlopen error [Errno -3] Temporary failure in name resolution>
This change decorates the function find_cirros_image() to retry and give
the job more chances of recovering from the DNS failures.
When pgrep_full=True is used, the resulting pgrep can match
more binaries than expected if the search pattern isn't specific
enough. This change makes the search patterns more specific
in order to only match the expected binaries.
Closes-Bug: #1933338
The keystone charm allows multiple VIP addresses to be configured via
the 'vip' config parameter. Unfortunately, get_keystone_ip() didn't
support this. This patch adds that support, plus the missing tests.
Closes: #689
The current rabbit tests post a message on one unit and then
consumes the first message on another unit. If the two do not match
then the test fails. This means that a single pre-existing message
can break all these tests as the consumer always gets the wrong
message. This change updates the tests to search for the target
message rather than assuming that the first message is the target
message.
Messages that are not the target message are not reposted so any
pre-existing messages are consumed and in effect thrown away but
given these are all test messages in a test queue and the tests do
not run in parallel this should be fine.
Add zaza tests for charm glance when cinder is configured
as storage backend.
Related-Bug: LP#1905042
Co-authored-by: Hemanth Nakkina <hemanth.nakkina@canonical.com>
When doing a series upgrade from bionic to focal, the percona-cluster
charm needs to be avoided. This filters the applications for percona if
the target is focal. Note if percona is placed on the same unit as
something else that needs to be upgraded (bad idea) then it may still
get 'upgraded' and the operation will fail.
This PR tests that a new device is succesfully added to the OSD
charm, and that it's done so via the 'bcache' mechanism. The
backing storage is provided via a loopback device, whereas the
caching storage is provided by Juju itself.
