Skip to content


Subversion checkout URL

You can clone with
Download ZIP
Commits on Jul 1, 2015
  1. ns

Commits on Jun 17, 2015
Commits on Jan 13, 2015
  1. ns

Commits on Jan 6, 2015
  1. ns

  2. jmx

  3. jmx

Commits on Dec 19, 2014
  1. ns

Commits on Oct 31, 2014
  1. ns

  2. ns

Commits on Sep 16, 2014
  1. - Added Condition and CondVar

    - Added CondVarTest
    - Converted to use CondVar: Promise, Responses, Request/UnicastRequest/GroupRequest
    - Changed GC flags in
    - TimeService: converted ConnectionMap, TP, UNICAST3, FD_ALL, FD_HOST to use nanosecs rather than msecs
    - TimeService: converted ExpiryCache to use nanosecs rather than msecs
    - Converted LazyRemovalCache to use nanosecs rather than msecs
    - Changed Discovery to use nanosecs rather than msecs
    - Changed ResponseCollector to use nanosecs correctly
    - Changed TimeScheduler3 to use nanosecs correctly
    - Deprecated TimeScheduler2
    - Changed MERGE2 to use nanosecs correctly
    - Changed RATE_LIMITER to use nanosecs correctly
    - Changed FlowControl/MFC to use nanosecs correctly
    - Changed Locking to use nanosecs correctly
    - Removed busy sleeping from Util
    - Changed GMS to use nanosec instead of millisecond time
    - Changed FD to use nanosec instead of millisecond time
    - Deprecated UNICAST and UNICAST2
    - Deprecated NAKACK
Commits on Sep 3, 2014
  1. @rhusar
Commits on Jun 27, 2014
  1. - Replaced tabs with spaces in UDP

    - Added -uuid option to UPerf
    - JDBC now extends Discovery directly rather than FILE_PING. This makes it easier to refactor FILE_PING and all subclasses which access cloud stores
    - Added OneTimeAddressGenerator
    - Added IpAddress(String host_port) ctor
    - FILE_PING: first shot at JGRP-1841
    - Removed DumpPingData as contents are now user readable
    - Correct impl of readAll() and writeAll() in FILE_PING, SWIFT_PING, RACKSPACE_PING and S3_PING
    - FILE_PING: information is written to file if logical address was not in coord's cache or physical address doesn't match (e.g. joiner has a physical address that's different than the one in the cache)
    - Added LargeJoinTest
    - TP: internal thread pool can now be set
    - Changed version to 3.5.0.Beta8
Commits on May 28, 2014
  1. - Fixes for JGRP-1826

    - Changed PingData: one physical_addr instead of a list of phys addrs
    - Changed GossipData: one physical_addr instead of a list of phys addrs
    - Fixed compile error with JDK 8 in MessageDispatcher
    - Added Responses and test
    - Replaced MERGE2 with MERGE3 in all sample config files
    - Removed num_initial_members (deprecated and ignored) from XML config files
    - MERGE2/MERGE3 now use their own protocol to implement view fetching and doesn't depend on discovery anymore. This allows us to remove the view fetching code from the discovery protocols
    - Replaced is_server and is_coord with 1 byte flag in PingData
    - Removed division/zero exception in UPerf
    - Added keySet() to LazyRemovalCache
    - Destroying window on close
    - First shot at CloudBasedDiscovery.txt
    - Using GMS.join_timeout rather than Discovery.timeout for discovery
    - Set TCP.use_async_discovery to true in tcp.xml
    - Protocol: setValue() logs a warning if the field to be set is deprecated
    - PingData carries an optional list of members for which discovery is requested
    - Tests for modified PingData
    - Changed Discovery.findMembers() to accept a members list and populate a Responses object  (
    - ClientGmsImpl.join() uses FIND_INITIAL_MBRS event and then waits for GMS.join_timeout ms on the returned Responses object
    - MERGE2 uses FIND_MBRS event and then waits for discovery_timeout ms on the returned Responses object
    - MERGE3 uses FIND_MBRS event and then waits for discovery_timeout ms on the returned Responses object
    - TCPGOSSIP/RouterStub: changed signature of findMembers()
    - Passing async parameter to findMembers() in case async_discovery should be overwritten (e.g. in up(FIND_MBRS,...)
    - TP: using Responses returned by FIND_MBRS evt sent up
    - Looping back if target is a physical address and local address == target
    - SHARED_LOOPBACK_PING.findMembers() changed to the new method
    - FILE_PING: added a shutdown hook to remove own info in case of a crash
    - FILE_PING: selectively reading info for listed members in findMembers()
    - JDBC_PING.findMembers(): add result to Responses as soon as it has been read
    - RACKSPACE_PING.findMembers(): add result to Responses as soon as it has been read
    - SWIFT_PING.findMembers(): add result to Responses as soon as it has been read
    - S3_PING.findMembers(): add result to Responses as soon as it has been read
    - S3_PING: read members selectively if listed
    - Set TUNNEL.supportsMulticasting() to true
    - Removed timeout from discovery protocols' configs
    - Changed GMS.join_timeout to 2000 in udp.xml and tcp.xml
    - TP: when IP multicast is not available, sending a message to N-1 members by looking up the IP address for each member. We used to send this to *all* physical addresses in the cache, which was a waste
    - Sending a multicast (N-1) only sends to members of current cluster. Needed to take shared transports into account
    - AsciiString: prevent NPE in ctor when string is null
    - SHARED_LOOPBACK_PING return an empty Responses object if it is the first member in a cluster. This allows the member to become coordinator *immediately* rather than having to wait for GMS.join_timeout ms
    - Set GMS.max_join_attempts to 10
    - UNICAST3 only sends a CLOSE to dest if dest is still a member
    - TP: removed warning for incorrect destination as this might happen a few times (correct)
    - Added Responses.contains(Address addr)
    - Fixed Discovery.serializeWithoutView(): didn't add is_coord flag
    - Changed version to 3.5.0.Beta7
Commits on Mar 25, 2014
Commits on Mar 7, 2014
  1. - Changed type in EncryptHeader from short to byte

    - Removed encrypt_entire_msg boolean and folded it into type
    - XOR'ing the type before returning it (bit ENCRYPT_ENTIRE_MSG might be set)
    - Using a byte[] array for EncryptHeader.version rather than a string
    - Added a pool of ciphers, configured via cipher_pool_size
    - Using the same lock for a MessageBatch
    - Removed Observer: only used in unit tests (replaced with Protocol impl)
    - Updated doc on ENCRYPT
    - Moved ENCRYPT (with asymm. keys) on top of GMS
Commits on Nov 5, 2013
  1. Added Dan's UPerf2

Commits on May 2, 2013
  1. - First impl of MPerfRpc (

    - Added custom marshallers
Commits on Mar 19, 2013
  1. - Added logger for log4j2

    - Added log4j (2 JARs)
    - JIRA:
    - Defined correct lib/* inclusion for JARs in ./lib dir
    - Added config file for JUL logging
    - log4j2.xml: added follow=true (to tweak stdout/stderr)
    - JUnitXMLReporter: deleting old test directory before running new test
    - Using String.format() instead of MessageFormat.format()
    - Added vararg methods to Log
    - Implementation of log impls (+vararg methods)
    - Added support for log4j2 to LogFactory (prefers log4j2 over log4j over JUL)
    - Set the dest in UNICAST_MessagesToSelfTest: before it was null, so we didn't really test unicast messages !
    - Using runtest in build.xml
    - Fixed incorrect path for byteman tests
    - Documentation for logging
Commits on Dec 21, 2012
  1. ns

Commits on Mar 26, 2012
  1. - Moved TOA over RSVP in toa.xml

    - Renamed to
Commits on Mar 23, 2012
Commits on Feb 16, 2012
Commits on Feb 9, 2012
  1. Added CountTests

  2. Added DumpData

Commits on Jan 12, 2012
  1. changed max heap size to 600M

Commits on Dec 12, 2011
  1. Removed DTest

  2. - Renamed UnicastTestRpcDist --> UPerf

    - Moved UPerf to perf package
Commits on Dec 8, 2011
  1. Implemented DTest

Commits on Nov 25, 2011
  1. set -Xmx500m

Commits on Nov 21, 2011
Commits on Aug 17, 2011
  1. ns

Something went wrong with that request. Please try again.