feat: improve test speed

[kian99]

Description

This PR aims to improve the speed of several slow tests (tested using LXD, Microk8s specific tests have not been worked on). The current largest offenders are below (tests that takes >= 100s) based on running them locally on my laptop:

TestAcc_ResourceOffer (300.84s)
TestAcc_DataSourceOffer (146.42s)
TestAcc_ResourceIntegrationWithMultipleConsumers (298.13s)
TestAcc_ResourceIntegrationWithViaCIDRs (325.72s)
TestAcc_ResourceApplication_UpdatesRevisionConfig (139.34s)

The offer based tests are slow because they deploy the postgres charm and the provider then waits for the application storage to be ready which takes a while. Here we can opt to use the juju-qa-dummy-source and juju-qa-dummy-sink charms since the goal is to test offers rather than the postgres charm and storage (if there is a need to test the postgres charm it would likely be better suited for a separate test).

The second offender for the slow tests was the very lenient code in the client code offers.go which would wait up to 5m for an offer's connections to drop to 0. If the connections didn't drop to 0 within these 5 minutes, it would destroy an offer with the force flag set to true. I've dropped this to a more aggressive 30s.

Type of change

• Change in tests (one or several tests have been changed)

[hmlanigan]

@kian, I just landed a PR, #557, which helps with the waiting for storage situation. Please review the times for tests to complete again. Using real charms in production is helpful to find bugs. The two charms being suggested are very simple test charms.

[kian99]

@hmlanigan I've tested TestAcc_ResourceOffer with your changes included and it doesn't seem to have made a difference locally (I'll test the full suite later as it takes some time), which intuitively makes sense, waiting for the postgres charm to provision storage takes time and I don't believe was necessarily impacted by the bug that was fixed.

Looking at the integration test results of PR #557 vs the integration test results of this PR, you can compare and see a large difference in the test times for LXD tests (approx. 35m vs 17m)

Interestingly, the Microk8s based tests are quite quick taking 10-12 min.

Existing times

New Times

Finally, I agree that it's worthwhile testing real charms but these tests are testing the creation of offers and integrations, testing a Postgres charm that creates storage is imo stretching the goal of the test.

[hmlanigan]

I'm happy to update the charms for this 1 test.

Reducing the wait time when destroying an offer is a no go.

[kian99]

@hmlanigan It's only 2 tests that have been updated now.

TestAcc_ResourceOffer
TestAcc_DataSourceOffer

[hmlanigan]

Thank you. Let's merge once the tests all pass.

[kian99]

@hmlanigan Please take another look when you get a chance, all the tests that usually pass are passing.