Permalink
Commits on Sep 3, 2012
  1. @tinwelint

    Fixes a problem introduced by efc01d4 regarding recovering index crea…

    …tions
    
    (cherry picked from commit 82f1810)
    tinwelint committed Sep 3, 2012
  2. @tinwelint

    Fixes an issue where recovery on a log which contained changes to an …

    …index followed by deletion of that index. During recovery an exception was thrown previously when trying to do the changes to the index as it didn't exist at that time.
    
    (cherry picked from commit efc01d4)
    tinwelint committed Sep 3, 2012
Commits on Sep 1, 2012
  1. @tinwelint

    Fixes a problem where some elements in the array could be null

    (cherry picked from commit 934a3a7)
    
    Conflicts:
    
    	lucene-index/src/main/java/org/neo4j/index/impl/lucene/LuceneDataSource.java
    tinwelint committed Sep 1, 2012
  2. @tinwelint

    Imports

    tinwelint committed Sep 1, 2012
Commits on Jun 21, 2012
  1. @neo-technology-release-persona
  2. @neo-technology-release-persona
  3. @tinwelint @digitalstain

    Solves a problem where some IndexReader instances would be detached b…

    …ut never closed
    
    due to race conditions. Also some reduction of synchronized blocks.
    
    Changed the following:
     - Removed the "optimized" check before synchronizing on the searcher since
       it was causing problems and still called synchronized methods on that instance anyway.
     - Moved IndexWriter instances into IndexSearcherRef so that IndexWriterClockCache
       could be removed and synchronization regarding that. Also the configuration option
       'lucene_writer_cache_size' which is fine because it doesn't make sense to have a searcher
       w/o its writer and vice versa since we use the NRT feature of lucene.
     - Reduced the block where the global read lock is held (in LuceneIndex#query)
    
    Conflicts:
    
    	kernel/src/main/java/org/neo4j/graphdb/factory/GraphDatabaseConfigurationMigrator.java
    	lucene-index/src/test/java/org/neo4j/index/impl/lucene/PerformanceAndSanityIT.java
    tinwelint committed with digitalstain Jun 20, 2012
  4. @digitalstain
Commits on Jun 20, 2012
  1. @neo-technology-release-persona
  2. @neo-technology-release-persona
  3. @neo-technology-release-persona
  4. @neo-technology-release-persona
  5. @neo-technology-release-persona
  6. @neo-technology-release-persona
  7. @neo-technology-release-persona
  8. @neo-technology-release-persona
  9. @digitalstain
  10. @tinwelint @digitalstain
  11. @tinwelint @digitalstain

    Tests uses their own indexes, to avoid windows problems

    Conflicts:
    
    	lucene-index/src/test/java/org/neo4j/index/impl/lucene/TestLuceneIndex.java
    tinwelint committed with digitalstain Jun 15, 2012
  12. @digitalstain
  13. @tinwelint @digitalstain

    Forces the test to fail if any worker encounter an exception

    Conflicts:
    
    	embedded-examples/src/test/java/org/neo4j/examples/GetOrCreateTest.java
    tinwelint committed with digitalstain Jun 13, 2012
  14. @digitalstain

    Reduce synchronization in getIndexSearcher

    Uses the ClockCache to build a IndexSearcher cache that does not require synchonization on get(). Uses
    that to remove need for global lock when getting a searcher which is not closed and not stale.
    Removes the boolean stale flag from the pair stored in LuceneDataSource#indexReaders, folds it into
    IndexSearcherRef instead.
    Removed the incRef argument from getIndexSearcher, added CommitContext#close() instead.
    Increased the thread count in GetOrCreateTest, makes it fail more reliably
    
    Conflicts:
    
    	embedded-examples/src/test/java/org/neo4j/examples/GetOrCreateTest.java
    	lucene-index/src/main/java/org/neo4j/index/impl/lucene/LuceneDataSource.java
    digitalstain committed Jun 14, 2012
  15. @digitalstain

    Add a new cache, ClockCache

    Added a new cache implementation that behaves approximately like LRU but does not lock on get()
    digitalstain committed Jun 14, 2012
  16. @digitalstain

    Revert "Added a new kind of cache, a ClockCache that tries to mimic a…

    …n LRU but with less synchronization."
    
    This reverts commit fc1e67a.
    
    Conflicts:
    
    	lucene-index/src/main/java/org/neo4j/index/impl/lucene/LuceneDataSource.java
    digitalstain committed Jun 20, 2012
Commits on Jun 7, 2012
  1. Merge pull request #588 from systay/1.7-maint

    Fixes problem with optional properties and collection predicates
    committed Jun 7, 2012
Commits on May 31, 2012
  1. @simpsonjulian

    Pom maintenance

    simpsonjulian committed May 31, 2012
Commits on May 21, 2012
  1. @neo-technology-release-persona
  2. @neo-technology-release-persona
  3. @neo-technology-release-persona
  4. @neo-technology-release-persona
  5. @neo-technology-release-persona
  6. @neo-technology-release-persona
  7. @digitalstain
  8. @digitalstain

    Added a new kind of cache, a ClockCache that tries to mimic an LRU bu…

    …t with less synchronization.
    
    Test cases for the above
    Used that to implement IndexSearcher and IndexWriter caches that tend to behave like the current LRU based ones but with better synchronization behavior
    Used the above in LuceneDataSource in place of the existing solution. Based on that, removed some synchronization making the common get paths on index readers and writes lock free.
    digitalstain committed May 21, 2012