ss77892 opened a new pull request, #11234:
URL: https://github.com/apache/ozone/pull/11234
## What changes were proposed in this pull request?
In OM serialize the Ratis reply eagerly so the retry cache holds bytes, not
the OMResponse tree
OMRatisHelper.convertResponseToMessage returned a lambda Message that
captured the parsed OMResponse and serialized it on every getContent() call.
Ratis keeps the reply Message of every applied write in its retry cache, on the
leader and on every follower, for the retry cache timeout, and the cache has no
size bound. So under a write burst each OM retained the whole parsed response
tree (KeyInfo, locations, ACLs) per request for that long, and re-serialized it
each time the content was read.
Serialize once, up front, and hand Ratis a Message that holds only the
bytes. Same bytes on the wire and in the cache, so retried requests are still
answered from the retry cache exactly as before.
Comparison to master on freon omkg workload with 1m keys. OM instances have
3gb heap size:
||instance||metric||09\-11 16:22 · HDDS\-16407@c750bd76e8||09\-07 22:11 ·
master@54da65a9d4||
|node1|gc\_count|645.0 (\-33.2%)|965.0|
|node2|gc\_count|1,329 (\+39.2%)|955.0|
|node3|gc\_count|648.0 (\-75.0%)|2,589|
## What is the link to the Apache JIRA
https://issues.apache.org/jira/browse/HDDS-16407
## How was this patch tested?
UT
Benchmark on 3 node cluster using freon omkg.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
