[
https://issues.apache.org/jira/browse/JDO-624?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12675669#action_12675669
]
Christiaan commented on JDO-624:
--------------------------------
The request is basically that "things should not change", so I doubt whether I
can be more specific about something that should not change. I would assume
that if I set "NonTransactionalRead" to true (reading is not depending on
transactions), and I execute a query to read objects, this reading would not be
affected by a commit on another thread for the same pm.
However, 2 of 3 jdo implementations I tried so far changed the behaviour of the
query, so I think it should get some attention in the spec. One simply closed
the query, the other loaded all instances into memory and then closed the
query. From an implementation point of view this probably can be solved by
using ResultSet.HOLD_CURSORS_OVER_COMMIT or may be have separate connection for
queries when NonTransactionalRead, but I am no jdo implementation expert. I am
just describing this from a user point of view. Anyway, I will look at the spec
to see if I can discover the areas which are affected by this, but I am hoping
that someone with more indepth knowledge of the spec can help out here.
> Query behaviour independent of transactional boundaries
> -------------------------------------------------------
>
> Key: JDO-624
> URL: https://issues.apache.org/jira/browse/JDO-624
> Project: JDO
> Issue Type: Bug
> Affects Versions: JDO 2 maintenance release 2
> Reporter: Christiaan
>
> (see also
> http://www.nabble.com/NonTransactionalRead---Query-Fetchplan-to19875446.html)
> I have a question about the behaviour of Queries after a commit when
> "NonTransactionalRead" is set to true. Should a Query behave like it did
> before the commit? More specifically, when defining a fetchsize in the
> FetchPlan I would like to see that the Query is still honouring this
> fetchsize after a commit is given.This behaviour is not explicitly described
> in the specification.
> The scenario I am working with is as such:
> I have set NonTransactionRead=true and Multithreaded=true. I have 1 pm, and 2
> threads. Thread 1 performs reads (executing queries with large result set)
> and Thread 2 performs writes. Now when I am iterating over a large query in
> Thread 1 and Thread 2 is doing a commit, I don't want this to affect the
> iterating over the query in Thread 1.
> My proposal would be that the behaviour of the query should be unchanged by a
> commit(). For me this currently a portability issue, since our application
> depend on the fact that the query behaves the same. May be the following note
> to 14.3 Architecture of Query would be sufficient:
> "The behaviour of a query should not depend on transactional boundaries"?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
