Arbiter.Core.Operations

Insert a job, returning it with its database-generated fields. Nothing when an IgnoreDuplicate key already exists, or a ReplaceDuplicate one names a job that is in flight (claimed under a live lease), force-cancel-flagged, has a different parent, or has children (DLQ children count). Parent and rollup state come from insertJobTree.

Internal tree insertion path for engine-owned parent and suspension state.

Batch-insert direct tree leaves under one parent.

Insert children under a job this worker holds, making it a rollup finalizer. Ack it in the same transaction.

insertJob over a batch in one round trip, returning the jobs inserted or replaced. Jobs sharing a dedup key within the batch collapse the way sequential insertJob calls would.

insertJobsBatch discarding the rows, returning the rows affected, dedup replacements included.

What an insert path puts on its jobs, carrying the ambient trace context.

The stamp for the context in scope. One read covers every job inserted under it.

Upsert a child's result, keyed by (parent_id, child_id). A second write replaces the first. Its foreign key cascades on the parent's ack.

insertResult for several (parent id, child id, result) rows in one statement. Each row is an upsert.

A parent's child results, keyed by child id.

The last errors of a parent's children in the DLQ, keyed by child id. Children with no recorded error are left out.

Snapshot results into parent_state before DLQ move.

Read a job's raw parent_state snapshot, which a DLQ-retried finalizer comes back carrying.

Read child results, DLQ errors, and the parent_state snapshot for a rollup finalizer in a single query. Returns (childId->result, childId->error, parentStateSnapshot, dlqRowId->error).

Merge DLQ errors, child results and the snapshot, left-biased in that order.

Claim up to maxJobs visible jobs, one per group. Stamps the anonymous claimant.

claimNextVisibleJobs under a given worker id.

claimNextVisibleJobs claiming at most maxBatches batches of up to batchSize jobs. Each batch is one group, or ungrouped jobs batched together. Stamps the anonymous claimant.

A pool's statements, rendered once. A call binds only its parameters.

Assemble a pool's statements once. Batch size 1 is the single-job claim.

The mkJobStatements batch size of a claim that takes one job per batch.

The mkJobStatements pool size that stages no claim statement. Each claim renders its own.

claimNextVisibleJobsAs over a pool's staged statements. Returns the rejected rows for deadLetterRejected.

claimNextVisibleJobsBatched over a pool's staged statements.

A claimed row its payload type rejected, with the decode error.

Move a rejected row to the DLQ with its decode error. Returns the number of rows moved. Zero means the claim was voided first.

Decode a row's stored payload, or the reason the type rejects it.

decodeRow, throwing ParsingException for a payload the type rejects.

typedRow over a DLQ entry's job snapshot.

typedRow over an archive entry's job snapshot.

Add tokens to a key's bucket, clamped to [0, max]. A no-op without a policy. For operator top-ups and manually-refilled policies.

Delete full, idle rate-limit buckets. Returns the number pruned.

Refill every bucket under a prefix to full. Returns the number refilled.

Make a prefix's throttled jobs claimable again across the given queue tables, in one statement. Returns the number woken.

Wake one key's throttled jobs across the given tables, in one statement. Returns the number woken.

List every policy with its default/override params, bucket aggregates, and live throttled count across the given queue tables.

One prefix's policy view with bucket aggregates and live throttled count.

Whether a rate-limit policy exists for a prefix.

List a prefix's buckets with effective max and fill fraction, paginated.

Set or clear a policy's override params. Returns rows affected (0 if absent).

List every concurrency policy with its default/override limit and live key and in-flight aggregates.

One prefix's concurrency policy view with live aggregates. Nothing when the prefix has no policy.

Whether a concurrency policy exists for a prefix.

List a prefix's keys with effective cap and fill fraction, paginated.

Apply a concurrency policy's override-limit patch (retunes every key under the prefix). Returns rows affected.

Delete drained concurrency rows with no live job across the given tables. Returns the number pruned. A key whose advisory try-lock is contended is skipped until the next pass.

Lock the count rows, then recount those keys under the lock. A key seeded after the lock pass is left to its triggers. Seeds a count row for a live key that has none. Returns the rows repaired.

Rebuild the counts when a crash truncated the UNLOGGED table. Returns the rows it repaired.

Reconcile then prune, skipped when no concurrency key exists. Returns the rows repaired and pruned.

Ack a completed job. Deletes a childless job, suspends a parent whose children are still running, and wakes a parent whose last child finished. A child's ack takes the parent's advisory lock. Returns 1, or 0 for a job gone or reclaimed.

The pool's ack statements.

A table's ack statements, rendered once. A call binds only its parameters.

ackJob inside the caller's transaction, over staged statements.

ackJobsBatch inside the caller's transaction, over staged statements.

ackJob over a batch in one statement, locking the distinct parents to serialize with concurrent sibling acks.

Whether acking this job tees it into the archive (positive archiveFor).

Extend a job's visibility timeout. Matches the claim token only.

setVisibilityTimeout over a batch, reporting each row through VisibilityUpdateInfo. Each row must also match its holder. A row whose row lock or group lock is busy is not extended.

What a batch visibility update found for one job.

Park a failed job for its retry backoff, recording the error. Returns 0 for a job another worker holds or a suspended job.

Soft-nack a job. Hands back the attempt the claim consumed and records no failure. A delay parks the job for that long, rounded up to whole seconds. Without one, it is reprocessed once its lease lapses. Returns 0 for a job another worker holds, a suspended job, or a released claim.

nackJob over a batch in one statement, returning the ids nacked. Jobs another worker holds, suspended jobs and released claims are absent.

Take the advisory lock of every distinct parent named, ascending, before any row lock the caller goes on to take.

Take every advisory lock a cascade cancellation will acquire, in one ordered pass before locking its rows. A branch root may itself have a parent.

Lock every job named and all of its descendants, in one descending pass, then their groups and their parents' groups.

Apply lockJobTrees to the complete tree of each named job. Use these locks before tree cancellation.

Proof that this transaction ran lockJobTrees or lockJobTreesFromRoot, so it holds the rows and their groups before it wakes a parent.

Whether the caller already took the parent and tree locks over its whole set.

Move a job to the DLQ, cascading a rollup parent's descendants with it and waking the parent of a child. A held job is moved too. Returns 0 when the job's claim token changed or the job was deleted after the read.

moveToDLQ driven by scalar fields, for callers without a typed JobRead.

moveToDLQ over a batch, each job under its own error message. Jobs another worker reclaimed are skipped. Returns the number moved.

Retry the whole DLQ tree that holds a DLQ row, with a fresh attempt count. The dedup key is left behind. Returns the restored job for that row. Nothing when the row is gone or the root's parent no longer exists.

retryFromDLQ with an optional replacement payload. The columns the payload derives are derived again from it. Nothing retries the stored payload.

Whether a DLQ row with the given DLQ primary key exists.

List DLQ jobs, most recently failed first.

Delete a DLQ job, resuming its parent when no child of it is left in the main queue.

Delete multiple jobs from the dead letter queue, resuming any parents left childless. Returns the total number of DLQ jobs deleted.

Delete one batch of DLQ jobs per given queue that failed longer ago than the queue's DLQ retention (setQueueDLQRetention), resuming any parents left childless. A queue without a retention is skipped. Returns the rows deleted and the queues that failed.

retryFromDLQ over several DLQ rows, each tree in its own transaction. A row restored with an earlier row's tree is skipped. Returns the number of rows whose retry restored a tree.

retryFromDLQBatch over every DLQ row the filters match when it starts, oldest first.

List archived jobs (most recent first).

List archived (completed) jobs with composable filters and a typed sort (defaulting to most recent first).

Fetch a single archived job by its original job id.

List archived jobs in a group, most recent first.

Count archived jobs with composable filters.

Purge one batch of expired archived jobs per given queue. Each row uses its archive_expires_at value. Returns the rows purged and the queues that failed.

Delete one archived job by its archive primary key. Returns rows deleted.

Delete archived jobs by archive primary key. Returns rows deleted.

Re-enqueue an archived job as a fresh standalone job, keeping the archive row. Returns the new job, or Nothing when the archive row no longer exists.

reEnqueueFromArchive with an optional replacement payload. The columns the payload derives are derived again from it. The archive row keeps its payload. Nothing re-enqueues the stored payload.

Store a completed root job's result on its archive row. No-ops when the job was not archived. Returns rows updated.

updateArchiveResult for several (job id, result) pairs in one statement.

A narrowing predicate on a job listing. Filters combine with AND.

The table a JobFilter narrows.

Sort direction.

A sort direction as a query parameter.

Sortable columns on the main jobs table.

Underlying SQL column name for a JobSortColumn.

Sortable columns on the DLQ table. DlqId is the DLQ primary key and DlqJobId the original job id.

The DLQ column a sort key names.

Sortable columns on the archive table. ArchiveId is the archive primary key and ArchiveJobId the original job id.

The archive column a sort key names.

The WHERE clause a set of filters narrows a listing of one table by. Empty for no filters.

List filtered jobs, newest first.

List filtered jobs under an explicit sort spec. Nothing for both sort arguments orders by id DESC.

listJobsFilteredOrdered that also returns each job's derived status.

Count jobs with composable filters.

List filtered DLQ jobs, most recently failed first.

List filtered DLQ jobs under an explicit sort spec. Nothing for both sort arguments orders by failed_at DESC.

Count DLQ jobs with composable filters.

List a parent's children in the DLQ, most recently failed first.

Count DLQ jobs matching a parent_id.

List jobs, newest first.

Fetch a job by id.