stacklok / toolhive / 35070099263

71%

Build Type

push

github

Committed by web-flow

Commit Message

feat(vmcp): stop session init stalling on backends that never service the notification stream (#6633)

* feat(vmcp): make the backend session-init timeout configurable

WithBackendInitTimeout existed but nothing outside tests called it, so
session init was pinned to the 30s default. operational.timeouts only
ever raised it. Add operational.timeouts.backendInit and wire it, and
make an explicitly configured value authoritative so a longer workload
request timeout no longer extends it.

A backend that stalls the handshake instead of answering or failing now
has a bound short enough to lose the race against a client's own connect
timeout, which partialFailureMode: best_effort turns into a usable
session.

Default behaviour is unchanged when backendInit is unset.

(cherry picked from commit 7862764c3)

* test(vmcp): cover the session factory option wiring

Serve is too big to unit test, so the two lines wiring backendInit went
uncovered. Pull the option building into sessionFactoryOptions and test
its branches directly. No behaviour change.

(cherry picked from commit 02f0a0361)

* feat(vmcp): let backends opt out of list_changed propagation

Subscribing to a backend's list_changed notifications opens a standalone
notification stream during session initialization. A backend that accepts
the subscribe and then never services it stalls the handshake until the
init deadline, and clients with their own connect timeout give up first.

The connector already treats a nil sink as "do not subscribe" and its
comment says the stream must stay opt-in because some backends hang on
it, but the server supplies a sink for every session, so in production
that gate was unreachable.

Add operational.listChanged, and a WithListChangedFilter factory option
in the per-workload resolver shape the factory already uses, so a single
misbehaving backend can be excluded while the rest keep live
propagation. An excluded ba... (continued)

Coverage Stats

51 of 75 new or added lines in 3 files covered. (68.0%)

829 existing lines in 10 files now uncovered.

81662 of 115501 relevant lines covered (70.7%)

97.52 hits per line