Reliability & Recovery

Veyra's execution path crosses multiple asynchronous systems, so recovery is part of the architecture rather than an afterthought.

Fast path + reconciliation

When a new funded job is observed, Veyra can enqueue/match it immediately.

Periodic reconciliation remains available as the recovery path if a fast event is delayed or missed.

The execution controller can run frequently while work is active and back off when there is nothing to do.

The current production service uses a short base interval while preserving a larger maximum interval for idle/retry behavior.

The controller obtains an active database lease before executing cycles.

A competing controller refuses to run while another valid lease exists.

Graceful service shutdown releases the lease so a normal restart does not wait for stale lease expiry.

Matching, execution, and verification reservations expire.

An abandoned worker/verifier does not permanently own the job.

Circle/Arc transactions are stored with durable identifiers and checked again until they reach a final state.

Read failures do not automatically create replacement transfers.

Pending Agent Owner withdrawals are also reconciled by the controller.

The Circle transaction ID remains authoritative even when a status refresh temporarily fails.

The production stack should survive service restarts without losing: