[
https://issues.apache.org/jira/browse/JCR-2892?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Christopher Elkins reopened JCR-2892:
-------------------------------------
I do have a reproducible case in my application, but it is not something I can
easily reduce to a test case suitable for attachment here.
The actual number of rows returned is irrelevant. From the aforelinked PDF,
"Since the buffers are allocated when the SQL is parsed, the size of the
buffers depends not on the actual size of the row data returned by the query,
but on the maximum size possible for the row data. After the SQL is parsed, the
type of every column is known and from that information the driver can compute
the maximum amount of memory required to store each column. The driver also has
the fetchSize, the number of rows to retrieve on each fetch. With the size of
each column and the number of rows, the driver can compute the absolute maximum
size of the data returned in a single fetch. That is the size of the buffers."
The specific context is the same as that of JCR-2832, adding a new node to a
cluster. However, any statement that returns a BLOB column given the current
fetch size of 10,000 is going to create large buffers.
> Large fetch sizes have potentially deleterious effects on VM memory
> requirements when using Oracle
> --------------------------------------------------------------------------------------------------
>
> Key: JCR-2892
> URL: https://issues.apache.org/jira/browse/JCR-2892
> Project: Jackrabbit Content Repository
> Issue Type: Bug
> Components: jackrabbit-core, sql
> Affects Versions: 2.2.2
> Environment: Oracle 10g+
> Reporter: Christopher Elkins
>
> Since Release 10g, Oracle JDBC drivers use the fetch size to allocate buffers
> for caching row data.
> cf. http://www.oracle.com/technetwork/database/enterprise-edition/memory.pdf
> r1060431 hard-codes the fetch size for all ResultSet-returning statements to
> 10,000. This value has significant, potentially deleterious, effects on the
> heap space required for even moderately-sized repositories. For example, the
> BUNDLE table (from 'oracle.ddl') has two columns -- NODE_ID raw(16) and
> BUNDLE_DATA blob -- which require 16 b and 4 kb of buffer space,
> respectively. This requires a buffer of more than 40 mb [(16+4096) * 10000 =
> 41120000].
> If the issue described in JCR-2832 is truly specific to PostgreSQL, I think
> its resolution should be moved to a PostgreSQL-specific ConnectionHelper
> subclass. Failing that, there should be a way to override this hard-coded
> value in OracleConnectionHelper.
--
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
