UpdateHub / updatehub / 30379631623

66%

Build Type

push

github

Committed by otavio

Commit Message

test: mem_drain: cover the bounds that keep an offline device's log in check

An offline device records one probe failure per retry, and what keeps that from
growing without bound is counting consecutive repeats, with eviction as the
backstop for when the failures differ. Both were only lightly covered, so add
tests for what each of them promises:

- Memory stays flat over a hundred thousand retries of the same failure, which
  is the regression proper: the drain must hold no more after days offline than
  it did after the first few retries.
- Only a run at the tail is counted, and records differing in level or in their
  key values are not taken for repeats of one another.
- A counted repeat is not a new entry, so it must not move the first index and
  tell a reader there is something new to read.
- The first index grows across both eviction and the start of a new operation,
  so neither can be mistaken for the other.
- An entry larger than the bound is kept rather than evicted into nothing, and
  gives way once there is a newer one to keep instead.
- The repeat count reaches the JSON the SDK reads, not just the text form.
- Stopping recording keeps what was recorded and resumes without discarding it,
  which is what reporting a probe failure out of scope relies on.

Each was checked against a drain mutated back to the behaviour it guards, so
they fail when it regresses rather than merely passing today.

Coverage Stats

112 of 115 new or added lines in 1 file covered. (97.39%)

4 existing lines in 1 file now uncovered.

2765 of 4179 relevant lines covered (66.16%)

1191.83 hits per line