[
https://issues.apache.org/jira/browse/PHOENIX-3064?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15372687#comment-15372687
]
Lars Hofhansl commented on PHOENIX-3064:
----------------------------------------
Where this does not work are queries like {{SELECT * FROM T}}. That would
simply miss columns not in the cache.
We could detect {{SELECT c FROM T}} and then update the cache lazily if {{c}}
is not found.
This seems like a pretty important optimization, hitting the servers a 2nd time
for every query by default is very wasteful, especially since the CATALOG will
likely have a single region located on a single region server that will be hit
heavily.
> Reduce client time spent in createTableRef and ARRQ for point queries
> ----------------------------------------------------------------------
>
> Key: PHOENIX-3064
> URL: https://issues.apache.org/jira/browse/PHOENIX-3064
> Project: Phoenix
> Issue Type: Bug
> Reporter: Lars Hofhansl
> Attachments: PhoenixGetTableAndARRQ.png
>
>
> Did some profiling for point queries (I actually wanted to test the HBase
> server side).
> I found that a _lot_ of client time is spent in
> FromCompiler.getResolverForQuery (in createTableRef) and also in
> AbstractRoundRobinQueue.offer.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
