[ 
https://issues.apache.org/jira/browse/IMPALA-6121?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Tim Armstrong resolved IMPALA-6121.
-----------------------------------
       Resolution: Fixed
    Fix Version/s: Impala 2.11.0


IMPALA-6121: remove I/O mgr request context cache

This simplifies the lifecycle of the request contexts and eliminates
some code. The comments claim that request context cache improves
performance when allocating smallish the objects. But allocating
from TCMalloc's thread caches should scale much better than a
global object pool protected by a lock.

I needed to move the definition to a non-internal header file so that it
was visible to clients that manage it by unique_ptr.

We also do not need to transfer the request contexts to the RuntimeState
since I/O buffers do not leave scanners now.

Testing:
Ran exhaustive tests.

Change-Id: I91414eceaa4938fccd74686fe6bebede6ef36108
Reviewed-on: http://gerrit.cloudera.org:8080/8408
Reviewed-by: Tim Armstrong <tarmstr...@cloudera.com>
Tested-by: Impala Public Jenkins


> Remove RequestContext cache in DiskIoMgr
> ----------------------------------------
>
>                 Key: IMPALA-6121
>                 URL: https://issues.apache.org/jira/browse/IMPALA-6121
>             Project: IMPALA
>          Issue Type: Improvement
>          Components: Backend
>    Affects Versions: Impala 2.11.0
>            Reporter: Tim Armstrong
>            Assignee: Tim Armstrong
>             Fix For: Impala 2.11.0
>
>
> I'm planning to do some work on the I/O manager buffer management. I think 
> the lifecycle of request contexts is more complex than necessary because of 
> the recycling. That recycling is unlikely to save CPU cycles or improve 
> scalability relative to TCMalloc since the cache is behind a single global 
> lock.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Reply via email to