Asset import is blocking all available Caesium threads

[Stefan Stadler]

Hi all,

I have observed some strange behaviour in our environment.

When an import of Asset objects is happening (this happens quite regulary), there are no incoming emails processed. I managed to get a thread dump while the issue happened and found the following listed for all Caesium threads, that appear to be responsible for email processing:

As all the threads appear to be blocked by Asset importing new objects, I am wondering, how I could prevent that.

Currently we are running 10 workers for importing. The import still takes about 40min to complete in some cases. Could this be too much and reducing it might help here? Or are there any other tweaks for that?

Thanks for your help!

Stefan

[Mercy]

This behavior usually happens when Asset imports consume all available Caesium worker threads in the Jira Data Center node. Caesium powers scheduled jobs like email polling and imports, and the thread pool is shared across all background tasks. When your Asset import runs with 10 workers, each worker may tie up a thread for I/O and indexing, leaving none for other jobs such as mail processing. Atlassian documents that the number of concurrent Asset import workers can significantly affect overall system throughput and recommends tuning it based on available CPU and memory rather than maxing it out ([Jira Data Center scheduled jobs](