Thread pool statistics recommendations

Recommendations that appear for assistance with resolving thread pool issues are described in the table. The recommendations appear in the Recommendations panel when the Performance Service detects known issues in the CQL statement of a query.

Recommendation Description

Recommendation	Description
Mutation stage	Number of pending mutations. Too many pending mutations, which could be related to disk problems, poor tuning, or cluster overload. Consider adding new nodes, tuning the system, and revisiting your data model. If not `CPU` or `IO` bound, try increasing concurrent_writes.
Pending reads	Number of pending reads. Too many pending reads, which could be related to disk problems, poor tuning, or cluster overload. Consider adding new nodes, tuning the system, and revisiting your data model. If not `CPU` or `IO` bound, try increasing concurrent_reads.
ReplicateOnWriteStage Stress	Be careful when using `CL.ONE counter increments` because it has an async task, which also has a read in it, kicked off to run after the increment is completed. Too many processes in this pool will begin to block writes. Reduce the use of `CL.ONE counter increments` or upgrade to `Cassandra` 2.1 or higher.

Mutation stage

Number of pending mutations. Too many pending mutations, which could be related to disk problems, poor tuning, or cluster overload. Consider adding new nodes, tuning the system, and revisiting your data model. If not CPU or IO bound, try increasing concurrent_writes.

Pending reads

Number of pending reads. Too many pending reads, which could be related to disk problems, poor tuning, or cluster overload. Consider adding new nodes, tuning the system, and revisiting your data model. If not CPU or IO bound, try increasing concurrent_reads.

ReplicateOnWriteStage Stress

Be careful when using CL.ONE counter increments because it has an async task, which also has a read in it, kicked off to run after the increment is completed. Too many processes in this pool will begin to block writes. Reduce the use of CL.ONE counter increments or upgrade to Cassandra 2.1 or higher.

Thread pool statistics recommendations

Was this helpful?

Give Feedback