HDDS-8922. Random EC pipeline ID causes XceiverClient cache churn

[guohao-rosicky]

What changes were proposed in this pull request?

The XceiverClient cache is often expired when using ec reads

I am currently using ec for large concurrent reads and found that on the client side, a large number of XceiverClient cache remove logs .

Now each read builds a new pipeline, I think we can use the datanode id, as the pipeline id, so that we can take the client's XceiverClient cache.

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-8922

How was this patch tested?

Existing test

[guohao-rosicky]

hi @ChenSammi @sodonnel What are your thoughts on this change?

[adoroszlai]

Thanks @guohao-rosicky for the patch.

Tested it locally using Freon, reading the same 3MB EC key 1024 times concurrently in 16 threads. Cache stats from XceiverClientManager confirm the fix.

Before:

CacheStats{hitCount=3075, missCount=3075, loadSuccessCount=3075, loadExceptionCount=0, totalLoadTime=3789988382, evictionCount=2819}

After:

CacheStats{hitCount=6147, missCount=3, loadSuccessCount=3, loadExceptionCount=0, totalLoadTime=324556550, evictionCount=0}