outbox

Generated by tinox doc

tinox.core.outbox: transactional outbox over tinox.core.db and AMQP 0-9-1 publisher confirms.

Module	Version	Group
`db`	`1.2.2`	tinox.core
`amqp091`	`1.3.0`	tinox.core

class Outbox

* Writing side of the transactional outbox. add() stores the message in table `tinox_outbox` with * the ORM, i.e. on the connection of the caller's @Transactional method: the message is committed * or rolled back together with the business data, never separately. Nothing is sent here.

Methods

fnc add(destination: String, messageId: String, payload: String) → Nothing

* @param destination queue name (routing key on the default exchange) * @param messageId stable key of this logical message (the consumer deduplicates on it); must be * non-empty and at most 255 bytes * @param payload the message body as text (encode structured data yourself, e.g. JSON)

fnc backlog() → Int64

Records not yet confirmed by the broker and not parked

fnc parked() → Int64

Records parked after maxAttempts failures: they need an operator (fix the cause, set status back to 0)

fnc lagMs() → Int64

Age in ms of the oldest pending record, 0 if there is none (how far the dispatcher lags behind)

class OutboxDispatcher

* Publishing side of the outbox: reads pending records (oldest first, 50 per batch), publishes each to * its queue with the record's messageId as AMQP message-id using publisher confirms, and marks it sent * only after the broker confirmed it. * * Delivery is AT-LEAST-ONCE. If the process dies after the broker confirmed but before the record is * marked sent (or the confirm is lost), the record is published again after a restart, with the same * messageId: consumers must deduplicate on it (IdempotencyWindow091 or a unique key in their database). * No committed record is lost: a record stays pending until a confirm was seen. Failures are retried with * exponential backoff (backoffMs * 2^attempts, capped at 60 s) and parked after maxAttempts, so nothing * retries forever. Order is by id, but a record in backoff is overtaken by later ones.

Fields

`host`	`String`
`port`	`Int64`
`vhost`	`String`
`user`	`String`
`pass`	`String`
`maxAttempts`	`Int64`
`backoffMs`	`Int64`
`confirmTimeoutMs`	`Int64`
`conns`	`List<AmqpConnection091>`
`chans`	`List<AmqpChannel091>`
`connected`	`Bool`
`stopRequested`	`Bool`
`confirmed`	`Int64`
`failures`	`Int64`

Methods

fnc create(host: String, port: Int64, vhost: String, user: String, pass: String, maxAttempts: Int64, backoffMs: Int64) → OutboxDispatcher

fn connect() → Bool

Opens the connection and enables publisher confirms; false (and nothing open) on failure

fn close() → Nothing

fn runOnce() → Int64

Publishes one batch; returns how many records the broker confirmed

fn publishOne(rec: OutboxRecord) → Bool

Publish + wait for the confirm of one record; settles the record either way

fn run(pollMs: Int64) → Nothing

* Runs until stop() (or until the owning task scope is cancelled), draining the current batch * before it returns; sleeps pollMs between empty rounds. Always closes the connection.

fn stop() → Nothing

@Entity@Tableclass OutboxRecord

* One message waiting to be published. Written by Outbox::add inside the application's own * transaction, published (and marked sent) later by the OutboxDispatcher. * * Table `tinox_outbox`; create it in your changelog, see docs/outbox.md. * status: 0 pending, 1 sent (broker confirmed), 2 parked (maxAttempts failed -- needs an operator).

Fields

@Id@GeneratedValue`id`	`Int64`
@Column`destination`	`String`
@Column`messageId`	`String`
@Column`payload`	`String`
@Column`status`	`Int64`
@Column`attempts`	`Int64`
@Column`lastError`	`String`
@Column`createdAt`	`Int64`
@Column`nextAttemptAt`	`Int64`
@Column`sentAt`	`Int64`