Queue Ingestion Task

Overview

A queue ingestion task routes matching traces, spans, or threads into an annotation queue for human review, and assigns them to reviewers as they arrive. No one has to add items to the queue by hand. On the Workflows page, these tasks are listed under Queue Ingestion; each queue also lists its own under Automations.

Create a Queue Ingestion Task

Click New ingestion task under Queue Ingestion in the Workflows panel (after picking Traces, Spans, or Threads), or open a queue, pick Automations in its left navigation and click Add ingestion task. To start it from a classifier, click Add action under the classifier in the Workflows graph; the task then only picks up traces with the labels you choose (see Chain Tasks).

1. Pick an ingestion mode

   Choose how the task runs:

   • Continuous — ingest matching data as it arrives, every 5 minutes.
   • Recurring — ingest a sample of each period on a Schedule you set; each run samples across the period since the last one.

2. Pick the queue

   Select the Annotation Queue to ingest into. A recurring task can leave it blank to create a new queue each period.

3. Name and describe

   Pick a clear Name (e.g. Production billing complaints) and an optional Description. The Data Model is locked to the queue type — a thread queue ingests threads, a trace queue ingests traces.

4. Configure filters

   Add Filters to narrow which items the task ingests. Filters use the same syntax as the Traces page and Dashboards, so you can match by environment, tag, metadata field, score, latency, or any combination. When a trace's or thread's evaluation lands after the task's sweep, a continuous task that filters on metrics checks it again. To ingest by classifier label, chain the task after the classifier instead.

5. Tune sample rate and max items

   Set the Sample Rate between 0 and 1 to ingest only a fraction of matches (e.g. 0.2 ingests 20% of matching items). Optionally set Max Items to cap how many items the task can add to the queue total (Max Items Per Period for a recurring task, where it's required) — leave blank for no limit.

6. Pick an assignment strategy

   Decide who each ingested item gets routed to with the Assignment Strategy:

   • Single user — every ingested item is assigned to the project member in Assign To.
   • Round robin — rotates ingested items across the Reviewers you list, weighted by fewest assignments first (with lastAssignedAt as a tiebreaker). Even distribution is the goal — reviewers don't end up lopsided just because the task ran more often during their off-hours.
   • Random — picks one of the Reviewers at random for each ingested item. Useful when you want statistical coverage rather than even distribution.

   Whichever strategy you pick, every assignee receives an in-app notification and an email summarizing how many items were routed to them. Notifications are batched per ingestion run, so a tick that assigns ten items to one reviewer fires one notification, not ten.

7. Save

   Save the task. New items start arriving on the next run — every five minutes for a continuous task, on its schedule for a recurring one.

Each task row shows its name, target queue, data model, and how many items have been ingested so far. Disable a task with its switch when you're done — the queue and any already-ingested items are untouched.

Worked Examples

Single reviewer — a queue of low-feedback billing traces from production, with light sampling, that auto-routes to one reviewer:

Once enabled, the task ingests roughly one in five matching traces every five minutes (capped at 200 total) and assigns them to the named reviewer.

Round robin across a team — a queue of all production threads, distributed evenly across three on-call reviewers:

Each ingested thread goes to whichever of the three has the fewest assignments so far on this task. Each reviewer gets a single batched notification per tick listing how many items they picked up.

Next Steps

Last updated on October 6, 2026