[Data] Fix flaky test_streaming_split_stats

[goutamvenkat-anyscale]

Why

test_streaming_split_stats is flaky. The task_output_backpressure_time* metrics on the running MapBatches operator are wall-clock timers for output backpressure, which is a timing race against the single, slower streaming_split(1) consumer. The value is genuinely nondeterministic across runs (sometimes 0, usually positive), but the test hardcoded it to zero (Z), so it only passed on the occasional fast run (~1/5 locally; also seen failing in CI).

Empirically, output backpressure is the normal state for this pipeline (nonzero ~every run), so the hardcoded Z was the less-representative outcome. There is no deterministic value to assert for a wall-clock timer, so the honest fix is to stop asserting its magnitude rather than force a value.

What

Scoped, test-only change inside test_streaming_split_stats:

• Mark just the three timer values (average_task_output_backpressure_time_s, task_output_backpressure_time, task_output_backpressure_time_s) as not-asserted (a <varies> sentinel applied to both the expected and produced stats) — but only for the running MapBatches operator (first occurrence).
• Everything else stays strictly checked:
  • task_submission_backpressure_time remains N.
  • The idle split operator's backpressure timers remain strictly asserted as 0.
  • All other fields and all other tests are untouched (no change to canonicalize() or gen_expected_metrics()).

This mirrors the existing precedent for obj_store_mem_used (collapsed to A because it's "zero or positive depending on the run").

Verification

• test_streaming_split_stats: 10/10 pass (was 1/5).
• test_dataset__repr__, test_dataset_stats_basic: still pass (strict, unchanged).
• pre-commit: clean.

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request updates test_streaming_split_stats in python/ray/data/tests/test_stats.py to handle nondeterministic task_output_backpressure_time metrics. It replaces these metrics with a <varies> sentinel in both the expected and produced stats to prevent flaky test failures caused by timing races. There are no review comments, so no feedback is provided.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.