fix(issues): pair the trace id with a timestamp it agrees with, and scope the

logs tab to the trace instead of the page window

Two defects with one root. Every trace lookup on this page pins a +/-5min window
to the timestamp carried alongside the trace id, so the two have to be a matched
pair — and they were not. `updated_at` is bumped by activity that does not set a
new `recent_trace_id`: on the reference issue the recent trace is from 09-02
08:04 while `updated_at` reads 09-04 00:15, and the trace exists nowhere near it.
That mismatch is what made the waterfall report "couldn't load this trace" for an
issue whose trace is present and readable. The pairing is now exact whenever the
recent trace is the first one, which is the common case; when they genuinely
differ the old timestamp is kept, no worse than before.

Separately, the Logs tab scoped a trace-scoped query to the *page's* range rather
than the trace's. A trace happens at an instant, and `virtualTable` fetches on
connect rather than on tab activation, so every runtime-exception page load paid
for it. Measured on the reference issue: the same 14 rows in 0.49s scoped to the
trace, 35.9s over the page's +/-2h. The fallback branch already narrowed this way
and documented why; the trace branch just never did.

Also drops the errM.recentTraceId / traceRef disagreement — traceRef is the trace
the rest of the page shows, it honours First/Recent, and it carries the timestamp
the window needs, which a bare id does not.
This commit is contained in:
Anthony Alaribe 2026-09-04 11:46:26 +02:00
commit 356968f5e5