[
https://issues.apache.org/jira/browse/HADOOP-18391?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17575387#comment-17575387
]
Mukund Thakur commented on HADOOP-18391:
----------------------------------------
{quote}VectoredReadUtils.readInDirectBuffer should allocate a max buffer size,
.e.g 4mb, then do repeated reads and copies; this ensures that you don't OOM
with many threads doing ranged requests. other libs do this.
{quote}
Yes, makes sense.
{quote}readVectored to call validateNonOverlappingAndReturnSortedRanges before
iterating
{quote}
Can't do this as Checksum FS has to support the overlapping ranges because of
checksum ranges calculation and it uses the default implementation.
> harden VectoredReadUtils
> ------------------------
>
> Key: HADOOP-18391
> URL: https://issues.apache.org/jira/browse/HADOOP-18391
> Project: Hadoop Common
> Issue Type: Improvement
> Components: fs
> Affects Versions: 3.3.9
> Reporter: Steve Loughran
> Assignee: Mukund Thakur
> Priority: Major
>
> harden the VectoredReadUtils methods for consistent and more robust use,
> especially in those filesystems which don't have the api.
> VectoredReadUtils.readInDirectBuffer should allocate a max buffer size, .e.g
> 4mb, then do repeated reads and copies; this ensures that you don't OOM with
> many threads doing ranged requests. other libs do this.
> readVectored to call validateNonOverlappingAndReturnSortedRanges before
> iterating
> this ensures the abfs/s3a requirements are always met, and that because
> ranges will be read in order, prefetching by other clients will keep their
> performance good.
> readVectored to add special handling for 0 byte ranges
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
