server/internal/services/settings.go SaveSettings takes alerts and
email as required (non-pointer) values and writes them unconditionally
- absent fields would blank stored settings, not just leave them
alone. onLocalLoginChange was building its payload from the stale
loaded settings object instead of the in-progress form state
(thresholdMinutes/logRetentionDays) that handleSubmit uses, so editing
the offline threshold and then flipping the toggle silently reverted
the edit. Both paths now submit the same in-progress values.
ackAuthProviderNotice mutated callback_notice with no audit event; it
now writes auth_provider.ack_notice like create/update/delete.
guardProviderChange's callers turned any error from
CountEnabledAuthProviders into a 409 last_provider, so a transient
Mongo error was reported to the operator as an unremovable lockout.
Only services.ErrLockout now produces the 409; anything else is a 500.
auth_providers was missing from ScopedCollections, so reap.go's
scopedCollectionsForPurge() (derived from that list) never deleted an
instance's providers, leaving orphaned rows holding encrypted client
secrets forever. Verified migration 0004's $rename over org_id->instance_id
is a no-op here since auth_providers never carried org_id.
HandleLocalLogin and HandleListPublicProviders each computed their own
answer to whether password sign-in must stay available, and they could
disagree: an instance with local login off and a licence that lapses
loses its only provider and its password form in the same moment, with
no endpoint left to recover. services.LocalLoginPermitted is now the
one predicate both call.
InitRedis now takes a username and password, read from REDIS_USERNAME and
REDIS_PASSWORD, matching what admin has always done. Both empty keeps an
unauthenticated Redis working; a password with an empty username is what a
legacy requirepass instance needs, since go-redis then sends AUTH with one
argument instead of two.
This is what lets a Kubernetes install point at a managed Redis instead of
the bundled one.
Session.Close now closing its own accepted conn (from the prior fix wave)
made net.ErrClosed on the guacd-side reader indistinguishable from a real
remote failure, so a normal browser-tab close could race the handler's
defer and intermittently log console.proxy_failed on a healthy session.
Add a closing flag, set before Close's sync.Once body actually tears
anything down, that setReason respects -- a deliberate local teardown can
no longer produce or race in a failure reason, while Close's own explicit
reason argument still wins normally.
- Arm the unclaimed-relay watchdog in NewSession rather than Serve, so an
agent that never opens its ProxyStream is bounded to 10s and reports
reason "agent_timeout", per the design spec's failure-mode table.
- Session.Close now also closes the accepted net.Conn (stored via setConn),
so ConsoleProxy.Close() is an unconditional kill of the whole relay chain
instead of only closing an already-idle listener.
- Emit console.proxy_failed and end the console session from a defer in
consoleTunnel guarded on relay.Reason(), since guac's OnDisconnect never
runs when the connect callback errors -- which is the path every relay
failure this feature introduces takes. Update the two docsite
troubleshooting rows to match what the audit event can now actually show.
Every console session now rides the agent's outbound gRPC connection
instead of a direct guacd-to-target dial, so it works for servers
behind NAT and now requires a live agent (409 agent_offline
otherwise). Documents PROXY_ADVERTISE_HOST / PROXY_LISTEN_HOST and
corrects reachability claims across the docsite and CLAUDE.md.
mrhid6