Rotorsoft / act-root / 36911576578

100%

push

github

fix(act-otel): keep act_streams_blocked at its last value when no read answered (#1716)

The per-scrape collect() swallows each provider rejection so one degraded
store cannot blind every other metric, then called this.set(total)
unconditionally. A scrape where every provider rejected therefore published
0 — and max(act_streams_blocked) > 0 is a documented page, so it resolved
while the streams were still blocked, during the very store incident that
broke the read. The operator was told the poison messages had cleared.

Set the gauge only when something actually measured, so prom-client keeps
serving the last good sample and the alert stays latched. A partial failure
across bridges still publishes the reduced sum — the surrounding comment
chose "a failing app contributes nothing this scrape" deliberately, and this
does not relitigate it. An empty registry has nothing to preserve, so it
still publishes 0: nothing to measure is not a failed measurement.

Two comments in the same function disagreed about this, and the code
implemented the wrong one. The public doc-comment already promised the gauge
keeps its last value; now it does, and it says which half is which.

Closes #1702

Co-authored-by: rotorsoft <rotorsoft@outlook.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

4384 of 4384 branches covered (100.0%)

Branch coverage included in aggregate %.

3 of 3 new or added lines in 1 file covered. (100.0%)

9521 of 9521 relevant lines covered (100.0%)

1222.86 hits per line