Zhengchao An e3d7892404 test(io-metrics): assert what the remaining smoke tests only called (#6238) ...

Eight tests in this crate called a recorder and asserted nothing. Five of them were worse than that: every `record_*` in `list_objects_metrics` returns early unless `get_stage_metrics_enabled()` is true, and that flag defaults to false, so those tests only ever exercised the early return — never the code their names describe.

They now run against a local `DebuggingRecorder` with the flag on, and each asserts the boundary it is named for: an empty page reports the scan count as its amplification instead of dividing by zero, a zero read quorum is recorded rather than skipped, index serving divides verification attempts by returned objects, and the `-1` whole-directory sentinel reaches the limit histogram unclamped.

`msgpack_json_fallback_counter_records_without_panicking` has no in-struct total to check, so it now asserts the emission: two direction/message pairs must land in two separate series, which a dropped label would collapse into one.

The two process-sampler tests discarded their snapshots. They now assert what cannot differ between callers — a process has one start time and one descriptor limit regardless of which entry point or which sampler observed it, and the status enum must match its numeric projection.

This clears io-metrics from the census (`scripts/find_assertless_tests.py`), taking the tree from 61 candidates to 53.

Refs backlog#1836

2026-08-19 10:32:09 +08:00

..

mod.rs

fix(obs): isolate process sampler windows (#4492)

2026-07-08 15:51:34 +00:00

process.rs

test(io-metrics): assert what the remaining smoke tests only called (#6238)

2026-08-19 10:32:09 +08:00

system.rs

refactor(obs): migrate metrics runtime/schema and tighten migration guards (#2584)

2026-04-18 07:51:15 +00:00