[
https://issues.apache.org/jira/browse/JCR-3722?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Thomas Mueller updated JCR-3722:
--------------------------------
Description:
Currently, for queries that can use an index, the index is always preferred
over traversing over the nodes. For joins, sometimes is much faster to
traverse, specially the exact path is known. Example:
/jcr:root/testroot/b[@jcr:uuid]
This query will use the index on "jcr:uuid", and traverse over all nodes with
the property "jcr:uuid", even thought in fact only one node would need to be
read when using the TraversingIndex.
The problems are: Currently, the TraversingIndex is not even asked to estimate
the cost. Also, the cost estimation of the TraversingIndex is incorrect for
such cases.
We will also need to verify and fix the cost estimation of the other indexes
(for example the PropertyIndex) conforms the QueryIndex API (getCost is
supposed to return the estimated number of entries to traverse).
was:
Currently, for queries that can use an index, the index is always preferred
over traversing over the nodes. For joins, sometimes is much faster to
traverse, specially the exact path is known. Example:
{noformat}
/jcr:root/testroot/b[@jcr:uuid]
{noformat}
This query will use the index on "jcr:uuid", and traverse over all nodes with
the property "jcr:uuid", even thought in fact only one node would need to be
read when using the TraversingIndex.
The problems are: Currently, the TraversingIndex is not even asked to estimate
the cost. Also, the cost estimation of the TraversingIndex is incorrect for
such cases.
We will also need to verify and fix the cost estimation of the other indexes
(for example the PropertyIndex) conforms the QueryIndex API (getCost is
supposed to return the estimated number of entries to traverse).
> Query: an index is used even where traversing is faster
> -------------------------------------------------------
>
> Key: JCR-3722
> URL: https://issues.apache.org/jira/browse/JCR-3722
> Project: Jackrabbit Content Repository
> Issue Type: Bug
> Components: query
> Reporter: Thomas Mueller
> Assignee: Thomas Mueller
>
> Currently, for queries that can use an index, the index is always preferred
> over traversing over the nodes. For joins, sometimes is much faster to
> traverse, specially the exact path is known. Example:
> /jcr:root/testroot/b[@jcr:uuid]
> This query will use the index on "jcr:uuid", and traverse over all nodes with
> the property "jcr:uuid", even thought in fact only one node would need to be
> read when using the TraversingIndex.
> The problems are: Currently, the TraversingIndex is not even asked to
> estimate the cost. Also, the cost estimation of the TraversingIndex is
> incorrect for such cases.
> We will also need to verify and fix the cost estimation of the other indexes
> (for example the PropertyIndex) conforms the QueryIndex API (getCost is
> supposed to return the estimated number of entries to traverse).
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
