Problem
The shipped daemon has no production path that publishes a Remote Brain writer authority. Because of that, tracedecay remote failover (and any replay that needs the current writer) cannot get past the writer lookup on a real daemon.
RemoteSqliteStorageV1::publish_authority (crates/tracedecay-rusqlite-runtime/src/remote/mod.rs, around line 231) is the only code that INSERTs into the RemoteNode remote_authorities table. On master 9b79b48 it has no non-test caller:
$ rg -n "publish_authority" crates --type rust | rg -v "fn publish_authority"
crates/tracedecay-rusqlite-runtime/src/remote/recovery_authority.rs:230: self.publish_authority(&authority, frontier_sequence) # a different fn: RemoteRecoverySqliteAuthorityV1::publish_authority, which writes remote_recovery_authorities
crates/tracedecay-rusqlite-runtime/src/remote/tests.rs:660/709/735/1298 # tests only
Every production reader expects that row to exist:
RemoteSqliteStorageV1::recovery_writer / recovery_writer_for_lineage (remote/recovery_authority.rs, lines 268 and 290), which failover uses through DaemonRemoteRecoveryPhysicalEffectsV1::current_authority / promoteremote/replay_authority.rs, lines 30 and 92remote/recovery_authority/journal.rs::promote_primary_writer_in, which only UPDATEs an existing row
Consequence
On a real daemon, /failover → RemoteRecoverySqliteAuthorityV1::promote → ensure_authority_seeded → effects.current_authority → storage.recovery_writer finds no row and returns RemoteRecoveryOperationErrorV1::Unavailable. The rest of the promotion (writer-fence read, fence install, sink publication) cannot run from the shipped CLI. The only coverage is in-process tests that call publish_authority directly, for example daemon::store_runtime_tests::remote_failover_reports_corrupt_fence_text_naming_cancel_as_corruption_not_cancellation, added in the root-cause-remote-recovery lane.
Expected
Either a real production journey (enrollment, capture, or an operator authority-seeding operation) publishes the current writer authority with typed validation, or failover and replay report a typed "no writer authority published" state instead of generic Unavailable. Any fix needs a production-daemon journey test.
Found while typing the remote-recovery fence errors (lane root-cause-remote-recovery).
Problem
The shipped daemon has no production path that publishes a Remote Brain writer authority. Because of that,
tracedecay remote failover(and any replay that needs the current writer) cannot get past the writer lookup on a real daemon.RemoteSqliteStorageV1::publish_authority(crates/tracedecay-rusqlite-runtime/src/remote/mod.rs, around line 231) is the only code that INSERTs into the RemoteNoderemote_authoritiestable. On master 9b79b48 it has no non-test caller:Every production reader expects that row to exist:
RemoteSqliteStorageV1::recovery_writer/recovery_writer_for_lineage(remote/recovery_authority.rs, lines 268 and 290), which failover uses throughDaemonRemoteRecoveryPhysicalEffectsV1::current_authority/promoteremote/replay_authority.rs, lines 30 and 92remote/recovery_authority/journal.rs::promote_primary_writer_in, which only UPDATEs an existing rowConsequence
On a real daemon,
/failover→RemoteRecoverySqliteAuthorityV1::promote→ensure_authority_seeded→effects.current_authority→storage.recovery_writerfinds no row and returnsRemoteRecoveryOperationErrorV1::Unavailable. The rest of the promotion (writer-fence read, fence install, sink publication) cannot run from the shipped CLI. The only coverage is in-process tests that callpublish_authoritydirectly, for exampledaemon::store_runtime_tests::remote_failover_reports_corrupt_fence_text_naming_cancel_as_corruption_not_cancellation, added in the root-cause-remote-recovery lane.Expected
Either a real production journey (enrollment, capture, or an operator authority-seeding operation) publishes the current writer authority with typed validation, or failover and replay report a typed "no writer authority published" state instead of generic
Unavailable. Any fix needs a production-daemon journey test.Found while typing the remote-recovery fence errors (lane root-cause-remote-recovery).