# TiFlash duplicate MPP task dispatch

Use this topic only when the ticket contains the specific TiFlash MPP dispatch
error `Failed to register MPP Task ... task is already registered`, or an
equivalent exact duplicate-registration message. Do not apply it to a generic
TiFlash failure, an unrelated `Error 1105`, or resource pressure alone.

- TiDB dispatches MPP tasks to TiFlash. A coordinator retry after an RPC
  timeout or transient network problem can dispatch the same MPP task more
  than once.
- TiFlash can accept the first dispatch and then answer the duplicate dispatch
  with `task is already registered`. Current TiDB handling can treat that
  response as a dispatch failure and abort the query even though TiFlash had
  already accepted and started the task.
- This establishes a known product mechanism that directly explains the exact
  error and resulting query failure. It does not prove that a particular
  network fault occurred in the current incident unless aligned RPC or network
  evidence is available.
- Do not replace this exact mechanism with a generic apply, storage, CPU, or
  memory hypothesis merely because broad resource metrics are available.
  Resource evidence remains relevant only when it is entity- and time-aligned
  and supports an independent mechanism.
- For the reported failure, explain the known dispatch/retry/duplicate-
  registration chain and cite the public tracking issue. Keep the trigger
  boundary precise: RPC timeout or transient network trouble is an example
  cause of the retry, not a confirmed fact for the current occurrence.
- The tracked change is to handle this specific duplicate-registration
  response as idempotent success while leaving other errors unchanged. Do not
  promise a release date or claim the fix is already shipped without separate
  version-specific evidence.

Public tracking issue:

- https://github.com/pingcap/tidb/issues/65864 (OPEN as checked on 2026-08-20)