Ask AI

Customizing run queue priority#

When using a run queue, you can define custom prioritization rules for your Dagster instance.

By the end of this guide, you’ll:

  • Understand how the run queue works
  • Learn how to define custom prioritization rules
  • Understand how prioritization rules and concurrency limits work together

Understanding the run queue#

The run queue is a sequence of Dagster runs waiting to be executed. Dagster pulls runs from the queue and calls launch_run on submitted runs. It operates as a first-in, first-out priority queue.

For example, if three runs are submitted in the following order:

  1. Run A
  2. Run B
  3. Run C

Then the runs will be launched in the same order: Run A, then B, then C. This will be true unless there are tag concurrency limits or prioritization rules in place, which we’ll cover later in this guide.

By default, all runs have a priority of 0. Dagster launches runs with higher priority first. If multiple runs have the same priority, Dagster will launch the runs in the order they're submitted to the queue.

Negative priorities are also allowed and can be useful for de-prioritizing sets of runs, such as backfills.

Defining queue prioritization rules#

Custom priority is specified using the dagster/priority tag, which can be set in code on definitions or in the launchpad of the Dagster UI.

When defining a priority value, note that:

  • Values must be integers specified as a string. For example: "1"
  • Negative values are allowed. For example: "-1"

In Python#

In this example, the priority is set to -1 with a dagster/priority tag value of "-1":

@job(tags={"dagster/priority": "3"})
def important_job(): ...

    cron_schedule="* * * * *",
    tags={"dagster/priority": "-1"},
def less_important_schedule(_): ...

Understanding prioritization rules and concurrency limits#

Unless tag concurrency limits and/or prioritization rules are in place, queued runs are executed in the order they’re submitted to the queue. However, a run blocked by tag concurrency limits won’t block runs submitted after it.

Let’s walk through an example to demonstrate. In this example, three runs are submitted in the following order:

  1. Run A, tagged as team: docs
  2. Run B, tagged as team: docs
  3. Run C, which isn’t tagged

Without configured limits, these runs will be launched in the order they were submitted, or run A, then B, then C.

Before any more runs are launched, let’s add the following configuration to our instance’s settings:

  - key: "team"
    limit: 1

Now, runs A and B can’t execute concurrently, while there isn’t a limit on run C. Assuming each run executes for a minimum of five minutes, the order in which the runs are launched will change.

If the runs are submitted in the same order as before - that is, A, B, C - then the following will occur:

  • Run A launches
  • Run B B is skipped, as run A is in progress and concurrent runs are limited to 1 for the team tag
  • Run C launches
  • Run B launches after run A finishes

To summarize, due to the concurrency limit, this configuration will change the run launching order to A, C, B.