[
https://issues.apache.org/jira/browse/HIVE-5843?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13878842#comment-13878842
]
Alan Gates commented on HIVE-5843:
----------------------------------
bq. Therefore do we see these methods being called in tight loops? It seems
that the IO in this situation would dominate any unwrapping of objects?
I don't see this in tight loops, but whenever this is turned on we are adding a
round trip to the metastore on every read query (for lock acquisition) and two
for inserts (txn open plus lock acquisition). I'd just like to keep those as
fast as possible. I'll do some micro-benchmarking and if I can't show that it
costs something extra I'll move the single integer values to a struct. A
question though. Can thrift handle adding new parameters to functions? It
would seem a natural thing to support given thrift's version independence, but
I couldn't find anything one way or another in the docs.
bq. Additionally I would not throw an exceptions as they are problematic as well
What's wrong with exceptions in thrift?
> Transaction manager for Hive
> ----------------------------
>
> Key: HIVE-5843
> URL: https://issues.apache.org/jira/browse/HIVE-5843
> Project: Hive
> Issue Type: Sub-task
> Affects Versions: 0.12.0
> Reporter: Alan Gates
> Assignee: Alan Gates
> Fix For: 0.13.0
>
> Attachments: HIVE-5843-src-only.patch, HIVE-5843.2.patch,
> HIVE-5843.3-src.path, HIVE-5843.3.patch, HIVE-5843.patch,
> HiveTransactionManagerDetailedDesign (1).pdf
>
>
> As part of the ACID work proposed in HIVE-5317 a transaction manager is
> required.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
