[
https://issues.apache.org/jira/browse/HIVE-12636?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Eugene Koifman updated HIVE-12636:
----------------------------------
Attachment: HIVE-12636.02.patch
fixed most test failures
> Ensure that all queries (with DbTxnManager) run in a transaction
> ----------------------------------------------------------------
>
> Key: HIVE-12636
> URL: https://issues.apache.org/jira/browse/HIVE-12636
> Project: Hive
> Issue Type: Improvement
> Components: Transactions
> Affects Versions: 1.0.0
> Reporter: Eugene Koifman
> Assignee: Eugene Koifman
> Priority: Critical
> Attachments: HIVE-12636.01.patch, HIVE-12636.02.patch
>
>
> Assuming Hive is using DbTxnManager
> Currently (as of this writing only auto commit mode is supported), only
> queries that write to an Acid table start a transaction.
> Read-only queries don't open a txn but still acquire locks.
> This makes internal structures confusing/odd.
> The are constantly 2 code paths to deal with which is inconvenient and error
> prone.
> Also, a txn id is convenient "handle" for all locks/resources within a txn.
> Doing thing would mean the client no longer needs to track locks that it
> acquired. This enables further improvements to metastore side of Acid.
> # add metastore call to openTxn() and acquireLocks() in a single call. this
> it to make sure perf doesn't degrade for read-only query. (Would also be
> useful for auto commit write queries)
> # Should RO queries generate txn ids from the same sequence? (they could for
> example use negative values of a different sequence). Txnid is part of the
> delta/base file name. Currently it's 7 digits. If we use the same sequence,
> we'll exceed 7 digits faster. (possible upgrade issue). On the other hand
> there is value in being able to pick txn id and commit timestamp out of the
> same logical sequence.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
