highperfocused/multica
1
0
Fork 0
mirror of https://github.com/multica-ai/multica.git synced 2026-08-02 18:13:27 +02:00
Code Issues Packages Projects Releases Wiki Activity

fix(migrations): remove historical skill-bundle backfill (MUL-5400) (#6045)

Browse Source 
* fix(migrations): skip historical skill bundle backfill (MUL-5400)

Co-authored-by: multica-agent <github@multica.ai>

* fix(migrations): drop unused skill bundle migration

Co-authored-by: multica-agent <github@multica.ai>

---------

Co-authored-by: Eve <eve@multica-ai.local>
Co-authored-by: multica-agent <github@multica.ai>
This commit is contained in:
Multica Eve
2026-07-28 18:36:38 +08:00
committed by GitHub
parent 79cc3abc8d
commit c0f397b8d2
2 changed files with 0 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
server/migrations/233_backfill_skill_bundle_unavailable.down.sql
View File

@@ -1,16 +0,0 @@
-- Reverse the backfill: return the re-classified rows to agent_error.unknown,
-- the single bucket the up migration took them from. Restricting the filter to
-- the old wrapper prefix keeps the inverse exact — rows this migration never
-- touched (a new daemon writes skill_bundle_unavailable directly, with a
-- "skill bundle unavailable: ..." error string) are left alone rather than
-- being relabelled into a bucket they were never in.
UPDATE agent_task_queue
SET failure_reason = 'agent_error.unknown'
WHERE status = 'failed'
AND failure_reason = 'skill_bundle_unavailable'
AND error LIKE 'resolve skill bundles:%';
UPDATE chat_message
SET failure_reason = 'agent_error.unknown'
WHERE failure_reason = 'skill_bundle_unavailable'
AND content LIKE 'resolve skill bundles:%';

33
server/migrations/233_backfill_skill_bundle_unavailable.up.sql
View File

@@ -1,33 +0,0 @@
-- Backfill the skill_bundle_unavailable reason introduced in MUL-5370.
--
-- The daemon now labels "could not download the agent's skill bundles"
-- structurally instead of letting the wrapped transport error fall through
-- taskfailure.Classify into agent_error.unknown. That change is forward-only:
-- historical rows still sit in the unknown bucket, which makes the failure
-- invisible on the Usage page's Errors breakdown — exactly the observability
-- gap that let this bug survive in the wild.
--
-- The witness is the old wrapper prefix these rows were written with
-- ("resolve skill bundles: ..."), which is stable and unambiguous: no other
-- code path produced it.
--
-- Scoped to agent_error.unknown alone, and that is exhaustive rather than
-- conservative: the wrapper string first shipped 2026-06-23 (MUL-3281,
-- 5038c983c), three weeks after the in-flight classifier landed 2026-06-03
-- (MUL-2946, 10afd1af1). Every row that can carry this prefix was therefore
-- written through taskfailure.Classify and landed in agent_error.unknown --
-- none can hold the pre-MUL-1949 coarse 'agent_error'. Keeping the source
-- bucket single-valued is what makes the down migration an exact inverse.
UPDATE agent_task_queue
SET failure_reason = 'skill_bundle_unavailable'
WHERE status = 'failed'
AND failure_reason = 'agent_error.unknown'
AND error LIKE 'resolve skill bundles:%';
-- chat_message mirrors the task's failure_reason for chat turns, and the chat
-- bubble reads it directly — leaving these behind would keep showing the
-- generic "something went wrong" copy on historical turns.
UPDATE chat_message
SET failure_reason = 'skill_bundle_unavailable'
WHERE failure_reason = 'agent_error.unknown'
AND content LIKE 'resolve skill bundles:%';