Measure retry delays with monotonic clocks in the Python and Ruby conformance runners

[jeremy]

The flake class

delayBetweenRequests assertions compare captured request timestamps against an exact minimum-delay bound. When the capture uses a wall clock, any NTP step or slew between two captures can make a real sleep of min_delay read marginally short, failing the assertion — a nondeterministic flake unrelated to SDK behavior.

#496 reported exactly this in the TypeScript runner; 058fd11 (#488) patched it there with a TIMER_SLACK_MS = 2 tolerance (a libuv early-fire workaround). The same class remained latent in two other runners:

• Python (conformance/runner/python/runner.py): captured time.time(), asserted d < min_delay
• Ruby (conformance/runner/ruby/runner.rb): captured Time.now, same exact bound

The fix: Go's approach, not TS's

The Go runner never had this problem — main.go subtracts time.Time values, which is monotonic in Go. This PR applies the same principle rather than copying the slack-constant hack:

• Python captures time.monotonic()
• Ruby captures Process.clock_gettime(Process::CLOCK_MONOTONIC)

Both SDKs' retry sleeps wait against the monotonic clock, so measuring on the same clock makes the elapsed delta a true lower bound on the sleep — no tolerance constant needed, and the assertion stays exact.

Consumer audit

The captured timestamp's only consumer in both runners is the delay-delta math (delays_between_requests). Nothing exports or reports it as an absolute time — other assertions read method/url/headers/body, and reports carry name/pass/fail only. So the field swaps wholesale, renamed time → monotonic_time so it can't be mistaken for a wall-clock reading. Ruby's now-unused require "time" is dropped.

Replay runners capture no timestamps; Kotlin/Swift runners don't implement delayBetweenRequests. Go and TS are untouched.

Evidence

Both runners green with the change, exit codes pinned:

• make conformance-python → REAL_EXIT=0, 0 failed
• make conformance-ruby → REAL_EXIT=0, 0 failed

The delay-assertion fixtures (GET operation retries on 503, GET operation retries on 429 with Retry-After) PASS through the monotonic path in both; the two DownloadURL delay fixtures remain pre-existing intentional skips in both runners.

Closes the Python/Ruby residual of #496 (TS half already fixed by 058fd11).

---

Summary by cubic

Switch Python and Ruby conformance runners to monotonic clocks for measuring retry delays to eliminate flaky failures from wall-clock adjustments. Keeps exact minimum-delay assertions without adding tolerance.

• Bug Fixes
  • Python: replace time.time() with time.monotonic() and compute delays from monotonic_time.
  • Ruby: replace Time.now with Process.clock_gettime(Process::CLOCK_MONOTONIC) and compute delays from monotonic_time.
  • Rename captured field time → monotonic_time; remove unused require "time" in Ruby.

^{Written for commit 22b1241. Summary will update on new commits.}

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[jeremy]

Bot-review status: Copilot errored on its review attempt and a re-request could not be resolved via API; Codex is at its usage limit today. Zero review threads. Merging on green CI per the standing flaky-reviewer convention.