VERDOCK

Held queue and exceptions

Risky actions wait in a held queue while everything else keeps moving. The exceptions list is the only thing a person needs to look at.

On this page

When a result is unclear or a send is outside policy, only that action waits. Everything else keeps moving. Nothing runs twice.

What is held#

SourceWhy it waitsHow it leaves the queue
Offline queueThe egress gateway refused the action before any byte reached the provider, so it is known not to have happened.The caller reads the provider record first, then resumes the entry with the same request id.
Exchange storeAn exchange outside the send policy. Nothing leaves.A named approver approves, then the receiver accepts. See authorized exchange.
GuardA call that ends UNKNOWN keeps its reservation.Reconcile through the outcome log. The guard never retries on its own.

The held queue keeps the stable request id of each entry. Holding the same key again keeps the first entry. A provider that saw an action anyway cannot apply it twice, because the resume uses the same request id.

Exceptions#

exceptions() returns held items, operations whose latest outcome is MISMATCH, and operations that have been UNKNOWN for longer than a set number of minutes. VERIFIED and PENDING are never listed. An UNKNOWN younger than the threshold is not listed yet, because it is still being reconciled. Items are sorted by age, oldest first.

Each item is {"kind", "id", "since", "age_seconds", "detail"} with kind held, mismatch or unknown. Held items have no timestamp in the queue table, so their since and age are None and they sort first.

Terminal
verdock exceptions --outcomes outcomes.sqlite3 --held held.sqlite3 --older-than-minutes 15
verdock exceptions --exchange

The command prints JSON. --older-than-minutes defaults to 15. In the fleet scene, 100 agents run and the owner sees exactly 3 exceptions: one held, one MISMATCH and one old UNKNOWN.

Terminal
verdock demo --scene fleet