Quotas and limits

This document lists the quotas and system limits that apply to Batch.

  • Quotas have default values, but you can typically request adjustments.
  • System limits are fixed values that can't be changed.

Batch jobs run on Compute Engine resources, so Batch is also subject to Compute Engine quotas and limits. If your Batch job also uses other Google Cloud services, then it is subject to the quotas for those services too. Attempts to exceed a quota or limit return an error. For more information, see the Cloud Quotas documentation.

In addition to the quotas and limits from other Google Cloud services, Batch has the following quotas and limits, which cannot be changed.

Rate quotas

The following limits apply to Batch requests. These limits appear as quotas–for example, in the Google Cloud console–however, they can not be increased.

Limit Value
Agent requests per minute per region 30,000
Mutation requests per minute per region 6,000
Read requests per minute per region 15,000

Job limits

The following limits apply to Batch jobs.

Limit Value
Tasks per task group 100,000
Parallel tasks per job 5,000
Concurrent VMs per single-zone job1 2,000
Concurrent VMs per multi-zone job1 4,000
Size of the job definition2 1 MB
Wait time (duration of QUEUED job state) before automatic failure 2 days
Run time (duration of RUNNING job state) before automatic failure3 14 days

1 The terms single-zone job and multi-zone job here refer to the number of zones allowed for a job's VMs, which might be different than the zones for a job's location. For more information, see Batch locations.

2 For example, roughly the size of the REST request to the jobs.create method or gcloud CLI --config file. You can work around this limit by the dividing the tasks across multiple jobs—for example, learn how to schedule dependent jobs (Preview).

3 This limit has the following exceptions:

  • If you set the provisioningModel field for a job to FLEX_START, then the maximum run time before automatic failure is 7 days instead.
  • If a job uses a machine type from the A3 series, sets the reservation field to NO_RESERVATION, and has the provisioningModel either undefined or set to STANDARD, then the maximum run time before automatic failure is 7 days instead.
  • Optionally, you can reduce the maximum run time before automatic failure for a job by using timeouts.