[ 
https://issues.apache.org/jira/browse/HDDS-12834?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18108611#comment-18108611
 ] 

Chia-Chuan Ho commented on HDDS-12834:
--------------------------------------

Hi [~tejaskriya09] 
I looked into HDDS-12834 and propose making the background compaction service 
CPU-aware as follows: * Use the OM JVM process CPU utilization as the load 
signal.
 * Before compacting each column family, compare the current utilization with a 
configurable threshold (proposed default: 0.8).
 * If CPU utilization is above the threshold, retain the remaining column 
families and retry after a configurable interval (proposed default: 1 minute).
 * Keep compaction single-threaded. An already-running RocksDB compaction will 
not be interrupted.
 * Once a delayed cycle completes, wait for the normal run interval before 
starting another cycle; missed cycles will not be replayed.
 * If CPU utilization is unavailable, fail open and preserve the current 
compaction behavior.
 * Do not force compaction after a maximum deferral period, to avoid adding 
load while OM remains busy.
 * This applies only to the background service; on-demand admin compaction 
remains unchanged.

 
Does this match the intended behavior, particularly using OM process CPU rather 
than system CPU and the proposed defaults?  Thanks!

> Improve performance of auto-compaction of tables by the background service.
> ---------------------------------------------------------------------------
>
>                 Key: HDDS-12834
>                 URL: https://issues.apache.org/jira/browse/HDDS-12834
>             Project: Apache Ozone
>          Issue Type: Bug
>            Reporter: Tejaskriya Madhan
>            Priority: Major
>
> HDDS-12819 aims to introduce a background service that bottommost-compacts a 
> set of column families. This process can be optimized if it is aware of the 
> CPU utilization of the service that that point in time. 



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to