@ahasenack
> Without the fix, the Ocatvia API is unavailable, and that is not
desired, right? You want it to remain available.
Correct. This is a typo. Good catch.
> Any particular reason why, to test the issue, you chose to edit source
code, instead of setting enabled_provider_drivers to an invalid driver?
No particular reason. I actually did both when developing the patch. I
suppose I'll update the test plan to reflect that since it's barely any
extra effort per test and covers more situations.
** Description changed:
[ Impact ]
Currently, when the enabled_provider_drivers setting is not correctly
configured, e.g., it contains a non-existing driver, or when a provider
driver fails to load, octavia as a whole errors out and the API becomes
unreachable. This means that if there is a transient problem with a
provider, or a misconfiguration, even if that provider is not being
actively used, the other functioning drivers cannot be used.
[ Test Plan ]
1. Deploy openstack with charmed openstack including charmed octavia.
2. Ensure that the octavia endpoint is working by calling `openstack
loadbalancer provider list`
Expected Output
---------------
+---------+--------------------+
| name | description |
+---------+--------------------+
| ovn | Octavia OVN driver |
| amphora | Amphora |
+---------+--------------------+
---------------
3. Manually break the ovn-provider by editing `/usr/lib/python3/dist-
packages/ovn_octavia_provider/driver.py` on the octavia unit to raise an
error in the `__init__` method of `OvnProviderDriver` i.e.
```
--- driver.py
+++ driver.py
@@ -38,6 +38,7 @@
def __init__(self):
super().__init__()
+ raise RuntimeError("internal error")
```
4. Restart the octavia api service with `juju run octavia/leader
restart-services` and wait a little for the unit to settle.
5. Call the API with `openstack loadbalancer provider list`
Expected Output
---------------
+---------+--------------------+
| name | description |
+---------+--------------------+
| amphora | Amphora |
+---------+--------------------+
---------------
- Expected Behavior: The Octavia API is unavailable.
+ Expected Behavior: The Octavia API is available.
Actual Output
-------------
Service Unavailable (HTTP 503) (Request-ID: None)
-------------
Actual Behavior: The Octavia API does not become available.
[ Where problems could occur ]
This patch implements the strategy described in [0], skipping over
drivers which error while allowing others to proceed with
initialization. The Octavia API service now removes the failing
driver(s) from the enabled list, and the other provider drivers remain
functional.
Because driver initialization errors are no longer fatal, it is easier
for user to miss problems with their configuration. Errors are still
logged, however, and misconfigured drivers will be missing from the
provider list.
[ Other Info ]
This patch has been merged upstream into master, stable/*, and
unmaintained/*.
Story for this patch
[0]: https://storyboard.openstack.org/#!/story/2008710
Upstream patches:
[1]: https://review.opendev.org/c/openstack/octavia/+/780215
[2]: https://review.opendev.org/c/openstack/octavia/+/976490
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2141119
Title:
[SRU] Don't fail if a provider driver cannot be loaded in Octavia API
To manage notifications about this bug go to:
https://bugs.launchpad.net/cloud-archive/+bug/2141119/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
