[
https://issues.apache.org/jira/browse/HADOOP-18855?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17825254#comment-17825254
]
Steve Loughran commented on HADOOP-18855:
-----------------------------------------
I'm thinking of adding a way to pass down the byte byffer release method
alongside the allocator (overloaded methods...) so that on read failures s3a &c
can
* only allocate buffers when submit() returns a range. that is if we think it
is better to do this, a
* release the buffers on failure -which doesn't happen today
* line us up for doing retries better, especially on combined reads, where we
need to continue from where we left off, without rereading from the start. it
may be better here to just split into individual ranges for retrying.
> VectorIO API tuning/stabilization
> ---------------------------------
>
> Key: HADOOP-18855
> URL: https://issues.apache.org/jira/browse/HADOOP-18855
> Project: Hadoop Common
> Issue Type: Improvement
> Components: fs, fs/s3
> Affects Versions: 3.4.0, 3.3.9
> Reporter: Steve Loughran
> Assignee: Steve Loughran
> Priority: Major
>
> what do do we need to do to improve the vector iO experience based on
> integration and use.
> obviously, we cannot change anything incompatibly, but we may find bugs to
> fix and other possible enhancements
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
