Tasks accumulating in management thread pool

I have a data node that has its management thread pool filled up (like 5 active and hundreds or even thousands of queued jobs). The other nodes are fine (0 active, 0 in queue.)

What kind of tasks get in to this pool? How can I find out which ones keep saturating the pool's queue?

Your best bet is to look at the tasks currently executing in that pool. Some of them perform IO, and an overflowing management pool can be a sign of slow IO, possibly a failing disk. You'd normally use the nodes hot threads API for that, but this probably won't work since it runs on the management threadpool. Instead, you can get a thread dump using the jstack tool that comes with your JVM.

Thanks, I queried the tasks API and most of what I got was either of these two:

"action": "indices:admin/mappings/fields/get[index][s]",
"action": "indices:admin/seq_no/global_checkpoint_sync",

Can you tell me what specific tasks get into the managament pool? (E.g. cluster state updates, node stat queries....) What requests or inner mechanisms generate these types of tasks?

I'm not sure these questions are going to help you with your issue. You need to obtain thread dumps to see what the management threadpool is really spending all its time doing.

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.