Fixing loadbalancer e2e test #12838
Fixing loadbalancer e2e test #12838
Conversation
Griffin-Sullivan
added
CherryPick
|
trigger: test-robottelo |
There was a problem hiding this comment.
I think the functionality that you added to this test makes a lot of sense; we should definitely be testing whether the load balancer is correctly proxying yum transactions to a second Capsule if the first Capsule is unavailable. However, I'm curious as to why you removed the existing functionality from the test instead of writing a new test to capture this scenario.
Before your changes, if I'm reading correctly, the test was effectively checking that the HAProxy server's round robin algorithm functions as expected with registration requests. With your changes, we're no longer testing that functionality, since the other test in this module specifies target parameters at registration time that should override the round robin algorithm.
Could you please explain your thinking a bit more regarding removing functionality vs. adding a new test?
| url = f'https://{loadbalancer_setup["setup_haproxy"]["haproxy"].hostname}' | ||
| server_url = f'{url}:8443/rhsm' | ||
| base_url = f'{url}/pulp/content' | ||
|
|
||
| result = rhel7_contenthost.download_install_rpm( |
There was a problem hiding this comment.
Non-blocking: since you're refactoring this test, consider replacing installing the katello-ca-consumer package and registering with sub-man with registering via global registration.
Griffin-Sullivan
force-pushed
the
lb-capsule
branch
from
7829e72 to
604e9d1
Compare
|
trigger: test-robottelo |
This was at my recommendation. I did not feel that our tests job should be to test whether round-robin works for HAProxy and that the steps involved here are not of a real use case nature. And this test was proving flaky. First and foremost, users care that if a host is registered that it can receive it's content no matter what is happening behind the load-balancer. Secondly, that if a host registers it will register no matter what is happening behind the load-balancer. So we could keep the registration test -- but I would opt for the same strategy of taking down a Capsule and ensuring that registration continues to work. |
We do have https://github.com/SatelliteQE/robottelo/blob/master/tests/foreman/destructive/test_capsule_loadbalancer.py#L270 so I can edit that after this. |
Griffin-Sullivan
force-pushed
the
lb-capsule
branch
from
604e9d1 to
65974df
Compare
|
trigger: test-robottelo |
Griffin-Sullivan
force-pushed
the
lb-capsule
branch
from
65974df to
4230751
Compare
|
trigger: test-robottelo |
|
@ehelms That makes sense to me. Thanks for the explanation. |
Griffin-Sullivan
force-pushed
the
lb-capsule
branch
from
4230751 to
f44cb61
Compare
|
trigger: test-robottelo |
Modifying this test to focus more on the content host being able to get content even if the capsule it's registered to is down.