Re: [hibernate-dev] Correlating Java 1.8 Streams and ScrollableResults with database cursors
Hi Vlad,
I'm not sure if we should set the fetch size automatically, especially
not globally. Could we not use the API wisely to encourage people to
provide an estimate of a reasonable size, which will be different on
each query?
I'm aware that MySQL requires this strange value to avoid pre-loading
it all in memory; in fact for example in Hibernate Search the "mass
index rebuild" has to scroll on the whole dataset, so the
documentation recommends to set the fetch size..
However in some cases the optimisation might make sense; for example
if the resultset is small, why would you use pagination when you can
load it all in one single roundtrip?
On the other hand, if you all think Hibernate should be able to figure
this out automatically, then maybe I should also change Search to
automatically set MIN_VALUE when running on MySQL, etc..
Thanks,
Sanne
On 19 November 2016 at 10:04, Vlad Mihalcea <mihalcea.vlad(a)gmail.com> wrote:
Hi,
I've noticed this new Jira ticket:
https://hibernate.atlassian.net/browse/HHH-11260
and I think we should correlate the Session#stream and Session#scroll with
database cursors.
While Oracle uses a fetch size of 10, and SQL Server uses an adaptive
cursor, PostgreSQL and MySQL just fetch the whole result set into the
client.
So, for MySQL we could do as suggested in the Jira ticket, by setting the
fetch size to Integer.MIN_VALUE.
My question is about PotgreSQL. Postgres requires setting the fetch size to
some positive value in order to turn the current statement into a database
cursor.
Now, should we set a default value for streaming (e.g. 10), which can be
controlled via a new config property:
hibernate.jdbc.stream.fetch_size
Let me know what you think.
Vlad
_______________________________________________
hibernate-dev mailing list
hibernate-dev(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/hibernate-dev