kubeflow / mcp-server / 37048655871
83%

DEFAULT BRANCH: main
Ran
Jobs 3
Files 38
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 81.608%. Remained the same
37048655871

push

github

web-flow 
fix(core): drive the clock in two flaky resilience tests (#290)

test_half_open_max_calls_limit and test_tokens_refill_over_time waited with a
real time.sleep() and relied on resilience.py measuring that sleep through
time.monotonic(). On Windows that clock has a resolution of 15.625 ms, which is
coarser than the margin either test allowed, so a sleep could read shorter
than it was:

- the breaker test slept 60 ms against a 50 ms recovery_timeout, and 60 ms
  could read as 47 ms, leaving the breaker OPEN (about 12% of runs)
- the limiter test slept 10 ms at 1000 tokens/s, and 10 ms could read as 0 ms
  when both reads fell in one tick, so nothing refilled (about 31% of runs)

Both appeared with the time.monotonic() switch in #222. That switch is correct;
only these two tests were not sized for it. CI runs on Ubuntu, where the clock
has nanosecond resolution, so this only affected local runs on Windows.

Add a clock fixture that patches time.monotonic and returns a function to
advance it, and use it in both tests. The slow marker comes off since neither
sleeps any more.

The limiter test now also asserts the bucket is empty before advancing the
clock. Without that check the old test still passed when acquire() never
consumed a token, since it only asserted that a later acquire() succeeded.

Fixes #289

Signed-off-by: Abhishek <abhikokadwar2@gmail.com>

2436 of 2985 relevant lines covered (81.61%)

2.45 hits per line

Jobs
ID Job ID Ran Files Coverage
1 python-3.10 - 37048655871.1 38
81.61
GitHub Action Run
2 python-3.11 - 37048655871.2 38
81.61
GitHub Action Run
3 python-3.12 - 37048655871.3 38
81.61
GitHub Action Run
Source Files on build 37048655871