20 May, 2008

1 commit


19 May, 2008

1 commit

  • 3242: SNMP: no error displayed when the security-agent-file property contains a wrong value
    3241: SNMP: traps are not sent to the correct trap community
    
    
    
    
    git-svn-id: https://svn.forgerock.org/opendj/trunk@4265 41b1ffd8-f28e-4786-ab96-9950f0a78031
    fguigues
     

16 May, 2008

1 commit

  • - use cursors for DB access
    - speed up substring buffer flush at end of load
    - fix incorrect handling of undefined index counts
    - increase default cache size to 60 %
    
    Related to issue 3161.
    
    git-svn-id: https://svn.forgerock.org/opendj/trunk@4261 41b1ffd8-f28e-4786-ab96-9950f0a78031
    dugan
     

14 May, 2008

1 commit


12 May, 2008

1 commit


07 May, 2008

5 commits


06 May, 2008

1 commit

  • …eration with a new superior entry thats in the subtree to be moved. An informative error message is added so the user can fix the mistake. 
    
    This patch also enables the debug logger to print the cause of an exception message as well as all the stack frames by default. It also removes the deadlock retry configuration parameter from the JE backend configuration definition since deadlocks can no longer occur in the JE backend.
    
    Fix for issue 3236
    
    
    git-svn-id: https://svn.forgerock.org/opendj/trunk@4231 41b1ffd8-f28e-4786-ab96-9950f0a78031
    boli
     

05 May, 2008

2 commits


02 May, 2008

6 commits


30 Apr, 2008

6 commits


29 Apr, 2008

1 commit


28 Apr, 2008

1 commit


25 Apr, 2008

2 commits


23 Apr, 2008

2 commits


22 Apr, 2008

2 commits

  • … in heavy add/mod load. An add operation takes a read lock on the parent DN and a write lock on the target DN in that order. However, mod operations first takes an write lock on the target DN then inadvertantly tries takes an read lock on the parent DN through the entryExists method. This could cause a deadlock in the following case with the fair ordering reentrant read write lock:
    
    Thread 1 is performing a modify operation on cn=dep2,cn=dep1,dc=example,dc=com (entry 2)
    Thread 2 is performing an add operation on cn=dep2,cn=dep1,dc=example,dc=com
    Thread 3 is performing an modify operation on cn=dep1,dc=example,dc=com (entry 1)
    
    Thread 1 takes a write lock on target entry 2
    Thread 2 takes a read lock on parent entry 1
    Thread 3 tries to acquire write lock on target entry 1
    Thread 2 blocks trying to acquire write lock on target entry 2
    Thread 1 blocks trying to acquire read lock on parent entry 1
    
    Threads 1, 2, and 3 deadlocks since thread 3's write lock request is before thread 1's read request in the wait queue.
    
    The entryExists method in Backend.java does not need to acquire an read lock on the DN before checking for the entry's existance in the DB, much like the getEntry method. Removing the lock acquisition in entryExists ensures all locks are acquired down the DIT.
    
    The DependencyTest unit-test now passes and is enabled.
    
    Fix for issue 2852
    
    
    git-svn-id: https://svn.forgerock.org/opendj/trunk@4185 41b1ffd8-f28e-4786-ab96-9950f0a78031
    boli
     
  • 
    git-svn-id: https://svn.forgerock.org/opendj/trunk@4182 41b1ffd8-f28e-4786-ab96-9950f0a78031
    boli
     

21 Apr, 2008

1 commit

  • … a fixed lock timeout for subtree delete and mod DN operations. Previously, any index modifications to subordinate entries of the affected operations will be performed with dn2id and id2entry modifications. This creates multiple random access to index database keys which could cause deadlocks in face of multiple parallel operations. With this fix, all index modifications are buffered up until the end of the operation so that each key of each index will be accessed once and in order. This maintains the DB access ordering in the JE backend of dn2id, id2entry, dn2uri, indexes in config order, VLV indexes in config order, and finally id2children and id2subtree. Since deadlocks should no longer occur in the JE backend, JE lock timeouts are now disabled at the JE environment level instead of the txn level. With this change, the performance of subtree deletes and mod DN operations have increased dramatically.
    
    In order to add buffering capabilities to the VLV index, the format of the index records had to be changed. Previous DBs with VLVs configured will no longer be compatible with this new revision. 
    
    Cursors operations for subtree mod DN and delete operations are now taken with READ_COMMITTED JE isolation level to avoid locking un-affected entries and possibility causing deadlocks. Write operations affecting the DN2ID and ID2ENTRY databases will aquire an write lock directly as early as possible to avoid deadlocks.
    
    A issue is also fixed where the debug log genereated during unit tests does not include stack traces.
    
    Fix for issues 2980, 2186, 2979
    
    
    git-svn-id: https://svn.forgerock.org/opendj/trunk@4180 41b1ffd8-f28e-4786-ab96-9950f0a78031
    boli
     

18 Apr, 2008

1 commit


17 Apr, 2008

1 commit

  • …rror logger used during server startup, command line tools, and un
    it-tests. Some categories of error messages can never be displayed on server startup, during unit-tests, and from command line tools (ie. verify-i
    ndex). 
    
    Fix for issue 3102
    his line, and those below, will be ignored--
    
    M    TextErrorLogPublisher.java
    
    
    git-svn-id: https://svn.forgerock.org/opendj/trunk@4170 41b1ffd8-f28e-4786-ab96-9950f0a78031
    boli
     

16 Apr, 2008

2 commits


15 Apr, 2008

2 commits