com.bigdata.journal

Class WarmUpTask

java.lang.Object

com.bigdata.journal.WarmUpTask

All Implemented Interfaces:

Callable<Map<String,BaseIndexStats>>

public class WarmUpTask
extends Object
implements Callable<Map<String,BaseIndexStats>>

Helper class to warm up the indices associated with various namespaces on the journal.

Author:

bryan

See Also:

pre-heat the journal on startup TODO The current implementation assigns one thread per index. However, it is also possible to use a reader pool to accelerate the index scans. This would require a change to the {@link com.bigdata.btree.AbstractBTree#dumpPages(boolean, boolean)} implementation to parallelize the IOs over the children of a node. We have experimented with that in the past but not observed a big win for query. It might be more significant for warmup since we are using a full scan of each index.

Constructor Summary

Constructors

Constructor and Description
WarmUpTask(Journal journal, List<String> namespaces, long timestamp, int nparallel, boolean visitLeaves)

Method Summary

Methods

Modifier and Type	Method and Description
Map<String,BaseIndexStats>	call()

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

WarmUpTask

public WarmUpTask(Journal journal,
          List<String> namespaces,
          long timestamp,
          int nparallel,
          boolean visitLeaves)

Parameters:

journal - The journal.

namespaces - A list of zero or more namespaces to be warmed up (optional). When null or empty, all namespaces will be warmed up.

timestamp - The commit time to be warmed up and -or- ITx.READ_COMMITTED to warm up the last commit point on the journal.

nparallel - The #of threads that will be used to scan the pages in the indices associated with those namespaces (GTE ONE). The indices will be scanned with one thread per index. This parameter determines the #of such scans that will execute in parallel. Since the thread will block on any IO, you need a modestly large number of threads here to enqueue enough disk reads to drive enough IOPs for an efficient disk scan. Even a slow disk should use at least 10 threads.

visitLeaves - when true the leaves of the indices will also be read (the recommended warm up procedure does not read the leaves).

Method Detail

call

public Map<String,BaseIndexStats> call()
                                throws Exception

Specified by:

call in interface Callable<Map<String,BaseIndexStats>>

Throws:

Exception