-
Type:
Bug
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: None
-
None
-
DB Integration & Observability
-
Fully Compatible
-
ALL
-
0
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Claude identified this is a race in our external data source test ("external data source" is the thing that ADF uses for their federated queries).
Root cause: named-pipe creation race
_writeTestPipeObjects() only spawns a detached thread and returns immediately — the test calls aggregate() on the very next line, before the pipe is guaranteed to exist.
The mkfifo() call happens inside that detached thread, via NamedPipeOutput's ctor, called from writeToPipeObjectsAsync() — there's no signal back to JS that the FIFO now exists.
mongod's reader side papers over this with a poll/sleep retry loop instead of a blocking wait: NamedPipeInput::doOpen() retries on ENOENT, sleeping 1→2→4→8→16ms, capped at ~31s total. If the writer thread isn't scheduled in time (likely on an oversubscribed try host), the retries exhaust and it throws FileNotOpen (throw site) — exactly this BF's error.
Same test has a long flaky history.
Fix (test-only): poll pathExists() via assert.soon() after each async pipe write, before aggregate(), so the FIFO is guaranteed to exist before the reader tries to open it.
- related to
-
SERVER-133871 Windows NamedPipeOutput leaks handle if writer thread creation fails
-
- Closed
-