chore(fix): unflake MetricsTracerTest

[kolea2]

Fixes #310

[codecov]

Codecov Report

Merging #313 into master will not change coverage.
The diff coverage is n/a.

@@            Coverage Diff            @@
##             master     #313   +/-   ##
=========================================
  Coverage     80.69%   80.69%           
  Complexity     1049     1049           
=========================================
  Files            99       99           
  Lines          6541     6541           
  Branches        344      344           
=========================================
  Hits           5278     5278           
  Misses         1082     1082           
  Partials        181      181           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 50a19a6...b667001. Read the comment docs.

[igorbernstein2]

I dont think this will fix the flakiness. I think the issue is that there is a race between the server thread (the emulator) and the client recording the duration:

client: starts 2 timers (first response, overall op) and sends request
server: sleeps(beforeSleep), sends response

race starts now
server: sleeps (afterSleep)
client: stops first response timer

server: finishes the call
client: stops overall timer

if the client won the race then timer will not be greater than overall timer - afterSleep, however if the server won the race then this condition will not be true.

I think the easiest fix is to add some slop to:
assertThat(firstRowLatency).isIn(Range.closed(beforeSleep, elapsed - afterSleep));

something like:
assertThat(firstRowLatency).isIn(Range.closed(beforeSleep, elapsed - afterSleep/2));

[igorbernstein2]

lgtm