mirror of
https://github.com/mastra-ai/mastra.git
synced 2026-11-17 08:44:46 +00:00
fix(mcp): address review of background-task MCP tasks
- Cancelling or expiring an 'auto' task whose row waits for a slot now aborts the attempt already running under its request; the manager never started it, so it could not abort it. - The sweeper dispatches unowned pending rows older than a lease: a task queued on an instance that died was otherwise stranded until a restart, since the manager only dispatches such rows at start-up. - Rows carry the tool as mcp-server:<id>:<tool>, so core's bare-name executor fallback can never run a Mastra tool named like the MCP tool with the task's arguments; an unserved row fails instead. - tasks/update errors instead of acknowledging an answer storage refused to store. - A failed store choice (storage unreachable) is no longer memoized as in-memory; the next request chooses again. - The sweeper starts on the first task request, not at registration. - Warn when tasks are stored under an id generated for this process. - Docs: what recover: true retries, and what ttlMs: null means in storage. Co-Authored-By: mastracode <284800079+mastra-platform[bot]@users.noreply.github.com>
This commit is contained in:
parent
66dec45811
commit
67d678751c
6 changed files with 276 additions and 27 deletions
Loading…
Add table
Add a link
Reference in a new issue