JBossDeveloper

2 - The main issue is that I don't want to remove my segment (remember I'm using Directory.copy(), I already spent an hour loading them in my store;-))......When the InifinispanIndexInput instance is closed the DistributedSegmentReadLocker.deleteOrReleaseReadLock method is called and in my case I expect it to release my lock (not delete my segment!!!). The delete is assumed because the FileReadLockKey has been evicted during the long copy process (remember my maxEntries param is low).... Then in the DistributedSegmentReadLocker.realFileDelete method all my chunks (68 chunks of 30MB) are removed transactionnaly (thru batch) so it means that I end up having my 68*30MB in memory leading to OOM. (hopefully the OOM occurs before the real remove on the store)

(...My understanding is that the endBatch() triggers the cache transaction commit, only then the store can save the entries.....)

it's opened : https://jira.jboss.org/browse/ISPN-680 however the assignee is Manik Surtani as I don't have necessary rights on this field.thx

I've downloaded the new fresh 4.2.0.ALPHA3 and it's working. Performance aren't good but I will next orient myself toward a cassandra store as it seems to be more adequate to handle big binary data chunk.

Here after the config (a mix of xml and programmatic):

<?xml version="1.0" encoding="UTF-8"?>
<infinispan xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
    xsi:schemaLocation="urn:infinispan:config:4.0 http://www.infinispan.org/schemas/infinispan-config-4.1.xsd"
    xmlns="urn:infinispan:config:4.0">
    <global>
        <transport clusterName="lucene-cluster"
            transportClass="org.infinispan.remoting.transport.jgroups.JGroupsTransport" />
    </global>
    <default>
        <loaders passivation="false" shared="true" preload="false">
            <loader
                class="org.infinispan.loaders.jdbc.stringbased.JdbcStringBasedCacheStore"
                fetchPersistentState="false" ignoreModifications="false"
                purgeOnStartup="false">
                <properties>
                    <property name="stringsTableNamePrefix" value="LUCENE" />                    
                    <property name="key2StringMapperClass" value="org.infinispan.lucene.LuceneKey2StringMapper" />
                    <property name="idColumnName" value="ID_COLUMN" />
                    <property name="dataColumnName" value="DATA_COLUMN" />
                    <property name="timestampColumnName" value="TIMESTAMP_COLUMN" />
                    <property name="timestampColumnType" value="BIGINT" />
                    <property name="connectionFactoryClass"
                        value="org.infinispan.loaders.jdbc.connectionfactory.PooledConnectionFactory" />
                    <property name="connectionUrl"
                        value="jdbc:as400://as400.url;database name=infinispan;libraries=infinispan;block size=512;lob threshold=16777216;extended dynamic=true;package=jbpkg;package cache=false;package library=infinispan;errors=full" />
                    <property name="userName" value="jboss" />
                    <property name="driverClass" value="com.ibm.as400.access.AS400JDBCDriver" />
                    <property name="idColumnType" value="VARCHAR(256)" />
                    <property name="dataColumnType" value="BLOB(40M)" />
                    <property name="dropTableOnExit" value="false" />
                    <property name="createTableOnStart" value="true" />
                    <property name="databaseType" value="DB2" />
                </properties>
                <async enabled="true" flushLockTimeout="15000"    threadPoolSize="4" />
            </loader>
        </loaders>
        <invocationBatching enabled="true" />
        <jmxStatistics enabled="true"/>
    </default>
    <namedCache name="luceneChunk">
        <eviction wakeUpInterval="1000" maxEntries="1" strategy="UNORDERED" />
    </namedCache>
</infinispan>

and the pertinent code:

    public InfinispanDirectoryFactory(String indexName, String namedCache, int chunkSizeInMB)
    {
        if(StringUtils.isEmpty(indexName))
            throw new IllegalArgumentException("indexName must be significant.");
        try
        {
            cacheManager = new DefaultCacheManager(INFINISPAN_CONFIGURATION);
            
            checkCacheNameExists(cacheManager,namedCache);
            
            Cache<String, Object> chunkCache =  cacheManager.getCache(namedCache);//define in xml
            
            //special caches
            Cache<String, Object> metadataCache = this.getReplImmortalCache(cacheManager, "metadata_" + namedCache, true);
            Cache<String, Object> lockCache = this.getReplImmortalCache(cacheManager, "lock_" + namedCache, false);
            

            this.infinispanDirectory = new InfinispanDirectory(metadataCache , chunkCache, lockCache , indexName, chunkSizeInMB * INFINISPAN_LUCENE_CHUNK_SIZE_IN_MB);
        }
        catch (IOException e)
        {
            throw new RuntimeException("Problem loading cache configuration file "+ INFINISPAN_CONFIGURATION, e);
        }

    private Cache<String, Object> getReplImmortalCache(EmbeddedCacheManager cacheManager, String newCacheName, boolean needStore)
    {
        Configuration newConfig = new Configuration();
        newConfig.setCacheMode(CacheMode.REPL_SYNC);
        if(!needStore)
            newConfig.setCacheLoaderManagerConfig(new CacheLoaderManagerConfig());//no cache loader/store if null is passed then NPE is thrown
        //[the previous has no effect, It seems the override features is not working to remove feature, it only changes or enriches a base configuration but cannot restrict:-(]
        newConfig.setExpirationLifespan(-1);//never expired
        newConfig.setEvictionMaxEntries(-1);//never evict
        newConfig.setEvictionStrategy(EvictionStrategy.NONE);
        cacheManager.defineConfiguration(newCacheName, newConfig);//override default config with newConfig
        return cacheManager.getCache(newCacheName);
    }

==> The one little issue with this is that it seems I cannot override the default configuration to remove the loader completely on the lock cache.(see my comment on the code)

The result is that a table is created in the db for the lock cache....

(And if I define all caches in xml I'll have to repeat the same loaders definition for both metadata and chunk cache....)

Next step for me will be to test the search phase in a distributed cluster.

I have 3 caches to configure. 2 of them will need the cache store (metadata and chunk) one of them won't (lock). So I decided to declare my store in the default cache and strip it programatically for the lock cache (lockConfig.setCacheLoaderManagerConfig(new CacheLoaderManagerConfig()) but the problem is that the configuration overrides mechanism seems not to work when restricting attributes (maybe I'm wrong here but the visitor pattern my friend is too much for me).

If I don't do like this I have to declare the cache store twice in xml for my 2 caches (which is duplicate code)...or maybe I should declare everything programmatically....I think the lockCache should be encapsulated in the InfinispanDirectory and not visible to the user (myself)...IMHO It now becomes quite cumbersome to configure 3 caches.