Repository navigation
[Improvement] Add retention helper to /remove-outputs: keep N newest runs within a tag #309
Description
Problem
A common housekeeping need is "delete the oldest runs until only N remain", optionally scoped to a tag ("keep the N newest runs with tag X, untouch the rest"). This currently requires custom client-side scripting on top of /get-outputs + /remove-outputs. It would be useful to support this natively.
Current behavior
/remove-outputs already supports:
limit-> keep N most recent, but global, ignores tags (_remove_by_limitin database.py).tags-> remove all runs with tag (_remove_by_tagin database.py).
These run independently, so "keep N newest among runs with tag X" cannot be expressed today.
Proposed direction
Make limit respect the tags filter when both are supplied: compute the candidate run set (runs matching any given tag), keep the N newest of that set, remove the rest. No new endpoint needed - extend the existing limit handling so it filters run_starts by tag before slicing.
{"limit": 10}-> unchanged (global, keep 10 newest).{"limit": 10, "tags": ["nightly"]}-> keep 10 newestnightlyruns, delete oldernightlyruns, leave other runs untouched.
Notes
- Pure superset of current behavior -> backward compatible.
- Covers both the no-tag and with-tag retention cases in one call.
- Add tests for: limit-only (unchanged), limit+single tag, limit+multiple tags, limit >= matching count (warn, no-op).
Activity
[-]Add retention helper to /remove-outputs: keep N newest runs within a tag[/-][+][Improvement] Add retention helper to /remove-outputs: keep N newest runs within a tag[/+]Narrow tag-scoped retention to limit only, matching issue #309
timdegroot1996 commentedon Sep 13, 2026
Scoped the implementation in PR #313 down to exactly what this issue asked for:
limit+tagsonly.The branch had grown a parallel
age+tagscombination that this issue never requested. That extra combination doubled the validation/detection logic inarguments.pyandserver.py, and requireddatabase.pyto parse a hand-rolled;-encoded tag suffix in two separate functions instead of one — complexity that wasn't earning its keep for an unrequested feature.Dropped
age+tags: it now behaves exactly like onmaintoday (two independent operations, no scoping).limit+tagsis untouched and still matches this issue's proposed direction and examples exactly:{"limit": 10}→ unchanged, global keep-10-newest.{"limit": 10, "tags": ["nightly"]}→ keep 10 newestnightlyruns, remove older matching runs, leave everything else untouched.Tests cover limit-only, limit+single tag, limit+multiple tags, and limit ≥ matching count (warn, no-op) — exactly what was asked for in the issue's Notes section.
feat(remove-outputs): tag-scoped retention for limit and age (#313)