[
https://issues.apache.org/jira/browse/LUCENE-10143?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17424063#comment-17424063
]
Robert Muir commented on LUCENE-10143:
--------------------------------------
{quote}
The issue here is more specific to the rate limiter and is well known as the
"incomplete delegator" issue.
{quote}
That's fine, we can make a smaller PR just for this rate limiter. But I believe
if you look hard at the prototype I made, you will see there are other
potential performance problems where the slow implementation is currently being
used:
* ByteBuffersDataInput (also used by ByteBuffersIndexInput), how this differs
from ByteBufferIndexInput (without the -s), I do not know. Maybe it only
affects RAMDirectory and not MmapDirectory?
* Lucene90CompressingStoredFieldsReader: I think numeric fields are decoded
from here with readInt(), etc?
* Various FST reader/writers, ByteSliceReader/Writer, PagedBytes, etc: worth
looking into.
It took making the things abstract to at least flush them all out so we can
look and evaluate them.
> RateLimitedIndexOutput should delegate writeShort/writeInt/writeLong
> --------------------------------------------------------------------
>
> Key: LUCENE-10143
> URL: https://issues.apache.org/jira/browse/LUCENE-10143
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Adrien Grand
> Priority: Minor
> Attachments: screenshot-1.png
>
> Time Spent: 3h 10m
> Remaining Estimate: 0h
>
> Otherwise merges are not taking advantage of LUCENE-10125.
> cc [~uschindler]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
