Refactor time-based tests to use fake and sharp loop time #881
Conversation
nolar
force-pushed
the
faketime
branch
3 times, most recently
from
e0f78f3 to
9c8ab16
Compare
|
This pull request introduces 1 alert when merging 420fa17 into 7b45690 - view on LGTM.com new alerts:
|
Otherwise, some tests may populate the cluster-scoped URLs while consuming the namespace-scoped URLs and thus catching errors, which are cryptic and hard to debug. Signed-off-by: Sergey Vasilyev <[email protected]>
Signed-off-by: Sergey Vasilyev <[email protected]>
Signed-off-by: Sergey Vasilyev <[email protected]>
Signed-off-by: Sergey Vasilyev <[email protected]>
…lds to the implementation Signed-off-by: Sergey Vasilyev <[email protected]>
…h has no meaning Signed-off-by: Sergey Vasilyev <[email protected]>
…tervals The routines that depend only on the internal intervals and durations should use the internal loop time only — which can be later faked by e.g. `looptime`. If the system or process time is used, these routines fall out of time-faking and might take many more cycles to reach the desired real-world timestamp (just by letting the real time flow naturally). This does NOT cover the cases with intervals and durations based on the global time (i.e. UTC clock) — e.g. handlers' timeouts, retries, etc. Signed-off-by: Sergey Vasilyev <[email protected]>
Signed-off-by: Sergey Vasilyev <[email protected]>
…topped/delayed) Signed-off-by: Sergey Vasilyev <[email protected]>
Utilise a new library — https://github.com/nolar/looptime — to have a fake time in event loops. The library does a lot of things, but the most important one is the sharp time with predictable steps with **no code overhead included**, which is typically random. Everything else goes as a side-effect: e.g. the fast execution of such tests with zero real-time for any fake duration of loop time, convenient `looptime` fixture for assertions, etc. Signed-off-by: Sergey Vasilyev <[email protected]>
Signed-off-by: Sergey Vasilyev <[email protected]>
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| await simulate_cycle(event_object) | ||
|
|
||
| assert k8s_mocked.sleep.call_count == 0 | ||
| assert looptime == 1.23 # i.e. no additional sleeps happened |
Check warning
Code scanning / CodeQL
Redundant comparison
| await simulate_cycle(event_object) | ||
| await dummy.wait_for_daemon_done() | ||
|
|
||
| assert k8s_mocked.sleep.call_count == 0 | ||
| assert looptime == 1.23 + 4.56 # i.e. not additional sleeps happened |
Check warning
Code scanning / CodeQL
Redundant comparison
| assert timer.seconds < CODE_OVERHEAD # near-instant | ||
| assert mock.call_args[0][0] == 'cancelled' | ||
| await scheduler.close() | ||
| assert looptime == 0 |
Check warning
Code scanning / CodeQL
Redundant comparison
| assert mock.call_args[0][0] == 'finished' | ||
| await flag2.wait() | ||
| assert looptime == 123 | ||
| assert mock.call_args_list[1][0][0] == 'finished' |
Check warning
Code scanning / CodeQL
Unreachable code
| sleep.return_value = 55 # simulated sleep time left | ||
| async with throttled(throttler=throttler, logger=logger, delays=[123, 234], wakeup=wakeup): | ||
| raise Exception() | ||
|
|
||
| await asyncio.sleep(77) |
Check warning
Code scanning / CodeQL
Unreachable code
| sleep.return_value = 55 # simulated sleep time left | ||
| async with throttled(throttler=throttler, logger=logger, delays=[123, 234], wakeup=wakeup): | ||
| raise Exception() | ||
|
|
||
| await asyncio.sleep(77) |
Check warning
Code scanning / CodeQL
Unreachable code
| sleep.return_value = 55 # simulated sleep time left | ||
| async with throttled(throttler=throttler, logger=logger, delays=[123, 234], wakeup=wakeup): | ||
| raise Exception() | ||
|
|
||
| await asyncio.sleep(1000) |
Check warning
Code scanning / CodeQL
Unreachable code
Utilise a new library — https://github.com/nolar/looptime — to have a fake time in event loops. The library does a lot of things, but the most important one is the sharp time with predictable steps with no code overhead included, which is typically random.
Everything else goes as a side-effect: e.g. the fast execution of such tests with zero real-time for any fake duration of loop time, convenient
looptimefixture for assertions, etc.This PR also demonstrates how the tests can be rewritten to use the provided looptime's features (and is mentioned in its README).
Fixes #212; related: #338