You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
mcp2cli cache warm <service> deadlocks on the daemon for a daemon-routed service:
mcp2cli cache warm open-brain
# -> open-brain: failed (Failed to communicate with daemon: Daemon failed to start within 10000ms)
cache warm routes its refetch THROUGH the daemon (discoverServiceSchemas → discoverServiceSchemasViaDaemon), which self-calls the same running daemon process and times out. This makes the documented refresh step in open-brain/docs/downstream-rollout.md non-functional on CT216.
Split out of #58 / PR #59 by decision: the obvious fix (route cache warmdirect, bypassing the daemon via the existing discoverServiceSchemasDirect path) partially reverses PR #52 "[codex] route schema refresh through daemon", which deliberately made these commands daemon-routed (enforced by tests/cli/daemon-routed-discovery.test.ts, which mocks direct transports to throw). The direct path does resolve credentials safely (imports the Vaultwarden resolver), so it's not a security bypass — but reversing a prior intentional decision deserves its own focused PR + review.
Note: #58's invalidation core does NOT depend on this — the drift-hook repopulation already routes around the self-call by reusing the already-open pool connection. This is purely the standalone cache warm command.
Keep daemon-routed; fix the self-call: detect the running daemon and reuse its own connection for the warm fetch instead of dialing a fresh daemon. Preserves [codex] route schema refresh through daemon #52 fully; larger daemon-internals change.
Acceptance
mcp2cli cache warm open-brain succeeds on the CT216 daemon (no "Daemon failed to start" timeout).
open-brain/docs/downstream-rollout.md refresh step works as documented (coordinate with the OB-side known-issue note).
Activity