Summary
sentry explore --dataset errors --sort '-count()' prints:
[explore] WARN --sort is only supported on the spans dataset. Ignoring sort for errors.
Combined with --limit, this means an errors aggregate is an arbitrary N rows, not a top N — but the flag combination reads exactly like a top-N request, and the warning does not say that the returned rows are unordered.
The underlying events endpoint does support ordering on errors:
sentry api "/organizations/<org>/events/?field=failure.node&field=count()\
&query=release%3A%22<release>%22&statsPeriod=14d&project=<id>&dataset=errors\
&sort=-count()&per_page=5" --json
returns a correctly descending result:
ReturnMain = 1625
(empty) = 1510
TargetCountStartReplay = 485
controller_initialization_failed = 413
HomeFlag = 289
Environment
sentry 0.40.0
node v24.18.0
Windows 10.0.26200
Request
Either:
- plumb
--sort through to the events endpoint's sort parameter for the errors dataset (the capability is already there); or
- if that is intentionally unsupported, make the warning state the consequence — that the result is an arbitrary subset and must not be treated as a ranking — and mention the events-endpoint alternative.
Why it matters
"Which failures dominate release X" is the most common aggregate question, and the natural formulation is --field ... --field 'count()' --sort '-count()' --limit 25. Today that silently returns 25 arbitrary rows. To rank anything you have to pull the full result set and sort locally, which then runs into #1519 (duplicate rows and a non-reproducible total on the same dataset). Fixing either one alone still leaves errors aggregates unusable for ranking.
Summary
sentry explore --dataset errors --sort '-count()'prints:Combined with
--limit, this means anerrorsaggregate is an arbitrary N rows, not a top N — but the flag combination reads exactly like a top-N request, and the warning does not say that the returned rows are unordered.The underlying events endpoint does support ordering on
errors:returns a correctly descending result:
Environment
Request
Either:
--sortthrough to the events endpoint'ssortparameter for theerrorsdataset (the capability is already there); orWhy it matters
"Which failures dominate release X" is the most common aggregate question, and the natural formulation is
--field ... --field 'count()' --sort '-count()' --limit 25. Today that silently returns 25 arbitrary rows. To rank anything you have to pull the full result set and sort locally, which then runs into #1519 (duplicate rows and a non-reproducible total on the same dataset). Fixing either one alone still leaveserrorsaggregates unusable for ranking.