Job Concurrency Queue not working as expected

[Neha_Gupta]

Hi,

 We have created a Databricks Jobs in Workflows where concurrent runs are set to 10 and the queue is enabled. We were trying to perform concurrent users testing by triggering 100 job runs using Jmeter script. We have observed that the first 10 jobs are run concurrently, and the remaining jobs are queued and when the remaining job from the queue starts running, they run one after other and not in parallel . When one job in queue is running others are still queued and the next only one job starts running only after the previous job finishes.

Request you to clarify us on the functionality and if this is expected behaviour and what changes we need if all the queued jobs need to be picked up in parallel based on count available for concurrent runs of the  jobs that can be run in parallel.

There was similar issue raised here

https://community.databricks.com/t5/data-engineering/job-concurrency-queue-not-working-as-expected/m...

@Kaniz 

[Kaniz]

Hi @Neha_Gupta, If you want to change this behaviour and allow more queued jobs to run in parallel, ensure that your Databricks cluster has sufficient resources (CPU, memory, etc.) to handle the desired concurrency.

• Job Configuration:
  • Check the job configuration settings.
  • Specifically, look at the “Maximum concurrent runs” setting. You might need to increase it beyond 10 to allow more concurrent executions.
  • Keep in mind that certain job types (e.g., Structured Streaming) may have different recommendations for concurrency.

[Neha_Gupta]

@Kaniz We have observed that even if none of the concurrent jobs are running the single job in queued status is only picked up for run even if rest 9 slots are free.

The databricks cluster has sufficient resources to handle the concurrency thats why in initial run all 10 got executed in parallel. But after that the jobs seems to be executed in sequential manner as only single job running with rest of the slots for concurrency still available.