Permalink
Commits on May 3, 2017
  1. Merge pull request #9267 from mneedham/3.2-unknown-config-load-balancing

    Only do prefix checking for the load balancing config
    digitalstain committed on GitHub May 3, 2017
Commits on May 1, 2017
  1. Merge pull request #9261 from digitalstain/3.2-follower-pin-multi-dc-…

    …license
    
    refuse_to_be_leader is controlled by multi_dc_license
    digitalstain committed on GitHub May 1, 2017
Commits on Apr 26, 2017
  1. refuse_to_be_leader is controlled by multi_dc_license

    The CC setting refuse_to_be_leader is deemed to be a multi dc
     specific feature and as such must be controlled by the multi_dc_license
     switch. The implemented behaviour is that unless the license is set to
     true, refuse_to_be_leader will always default to false, regardless
     of the actual value. A warning is also logged on the first access
     by ConsensusModule, in case a misconfiguration is suspected.
    digitalstain committed Apr 25, 2017
Commits on Apr 7, 2017
  1. Merge branch '3.1' into 3.2

    Conflicts:
    	integrationtests/src/test/java/org/neo4j/causalclustering/scenarios/BoltCausalClusteringIT.java
    digitalstain committed Apr 7, 2017
  2. Merge pull request #9162 from digitalstain/3.1-bolt-it-fix

    Attempt to stabilize  BCCIT.shouldPickANewServerToWriteToOnLeaderSwitch
    digitalstain committed on GitHub Apr 7, 2017
Commits on Apr 6, 2017
  1. Attempt to stabilize BCCIT.shouldPickANewServerToWriteToOnLeaderSwitch

    It looks that the occasional failures of this test come from concurrent session
     creation and switching of leaders. This is not a problem in real deployments,
     but the fixed timeouts and specific requirements for leader election results
     along with the constant composition of the cluster violate some assumptions
     the test makes. This change ensures the leader switch that is needed completes
     before any further action is taken, and also makes sure that no additional
     elections are triggered during the session creations.
    digitalstain committed Apr 6, 2017
Commits on Mar 13, 2017
  1. Merge branch '3.1' into 3.2

    digitalstain committed Mar 13, 2017
  2. Merge branch '3.0' into 3.1

    Conflicts:
    	enterprise/cluster/src/main/java/org/neo4j/cluster/protocol/atomicbroadcast/multipaxos/context/ClusterContextImpl.java
    	enterprise/cluster/src/test/java/org/neo4j/cluster/protocol/atomicbroadcast/multipaxos/context/ClusterContextImplTest.java
    digitalstain committed Mar 13, 2017
  3. Merge pull request #8986 from digitalstain/3.0-2-instance-quorum

    The master of a two instance cluster will now serve writes on slave failure
    digitalstain committed on GitHub Mar 13, 2017
  4. Merge pull request #9007 from digitalstain/3.0-mismatching-initial-hosts

    Cluster formation decision depends on initial_hosts contents
    digitalstain committed on GitHub Mar 13, 2017
  5. Introduces switch for initial hosts checking, off by default

    The strict enforcement of initial hosts is a slight departure from
     the way things happened so far. As such, it can cause unexpected
     behaviour in certain scenarios with unintentional mismatching
     initial hosts settings. This change introduces a feature toggle
     which is off by default, allowing the known behaviour to continue.
    digitalstain committed Mar 13, 2017
  6. Cluster formation decision depends on initial_hosts contents

    This change makes it so only instances listed in initial_hosts are
     considered for purposes of forming a cluster. This allows mismatching
     initial_hosts contents to be manipulated to make it harder for
     instances to start a cluster, for example slave_only instances to
     depend on the presence of a master but the master to be allowed to
     form the cluster regardless of the presence of other instances.
    The functionality is achieved with the introduction of a
     (backwards compatible) discovery header in the configurationRequest
     messages which keeps track of which instance has contacted which
     during the initial discovery round. Since sending of these messages
     depends only on the initial_hosts content, this effectively allows
     for filtering of configurationRequests based on initial_host content.
    digitalstain committed Mar 10, 2017
Commits on Mar 9, 2017
  1. The master of a two instance cluster will now serve writes on slave f…

    …ailure
    
    A two instance cluster cannot tolerate any failures, since quorum is the cluster
     size. This can be a problem in master+slave_only deployments, where the slave_only
     instance is meant as simply a read only replica. For this particular scenario,
     if the slave dies then it makes sense for the master to maintain write capabilities.
     This commit changes things to make it so that only for two instance clusters, if
     the slave dies (or a partition happens), the master will continue serving
     writes
    digitalstain committed Mar 9, 2017
Commits on Feb 7, 2017
  1. PropertyRecord.getType() now returns PropertyType instead of int

    Reading the property type and converting to a PropertyType value
     requires bit shifting and masking that is not achieved by simply
     using the type int value as read from disk. This can lead to
     improper identification of the record type during recovery, among
     other scenarios.
    This patch changes the PropertyRecord.getType() method to return
     PropertyType, properly parsed. It adds a getTypeAsInt() method
     to allow for access to the raw int value for serialization
     purposes.
    digitalstain committed with tinwelint Feb 6, 2017
Commits on Feb 2, 2017
  1. Merge pull request #8754 from digitalstain/3.1-unbind-close

    Ensures test neostore file is closed before test finishes
    digitalstain committed on GitHub Feb 2, 2017
  2. Ensures test neostore file is closed before test finishes

    This omission caused the test to spuriously fail on Windows,
     with inability to delete a mapped file during test directory
     cleanup.
    digitalstain committed Feb 2, 2017
  3. Merge pull request #8726 from digitalstain/3.1-truncated-log-recovery

    Partially written Raft log entries no longer cause recovery to fail
    digitalstain committed on GitHub Feb 2, 2017
Commits on Jan 31, 2017
  1. Partially written Raft log entries no longer cause recovery to fail

    In scenarios where a Raft log segment had a partial entry appended,
     recovery would complete normally, ignoring it as it should, but then
     subsequent entries would be appended after that, causing recoveries
     that happened before a rotation to fail. This is fixed by rotating
     at startup and after recovery but before any entries are appended,
     ensuring that no misalignment happens while avoiding introducing
     delete code for the Raft log segment files.
    digitalstain committed Jan 30, 2017
Commits on Dec 20, 2016
  1. Merge branch '3.1' into 3.2

    Conflicts:
    	enterprise/causal-clustering/src/main/java/org/neo4j/commandline/dbms/UnbindFromClusterCommand.java
    digitalstain committed Dec 20, 2016
  2. Merge pull request #8515 from digitalstain/3.1-unbind-lock-release

    Ensures all files are closed when running UnbindFromClusterCommand
    digitalstain committed on GitHub Dec 20, 2016
Commits on Dec 14, 2016
  1. Ensures all files are closed when running UnbindFromClusterCommand

    digitalstain committed Dec 14, 2016
Commits on Nov 18, 2016
  1. Merge branch '3.1' into 3.2

    digitalstain committed Nov 18, 2016
  2. Merge pull request #8401 from mneedham/3.1-read-replica-catchup

    Read replica catchup remake
    digitalstain committed on GitHub Nov 18, 2016
  3. Merge pull request #8397 from mneedham/3.1-chunking-store-copy

    Fixing store copy to use netty chunking
    digitalstain committed on GitHub Nov 18, 2016
  4. Merge pull request #8386 from maxsumrall/3.1-hazelcast-cleanup-deprec…

    …ated-usage
    
    Increase frequency which Hazelcast attempts merging/connecting.
    digitalstain committed on GitHub Nov 18, 2016
Commits on Nov 9, 2016
  1. Merge pull request #8330 from digitalstain/3.1

    Ensures IdReuseEligibility is always set in Causal Clustering mode
    digitalstain committed on GitHub Nov 9, 2016
  2. Ensures IdReuseEligibility is always set in Causal Clustering mode

    The field should have been initialized on every subclass of EditionModule,
     otherwise a null value would be passed in the id releasing job killing
     it on the first invocation. The result would be a constant accumulation
     of recycled but never released ids, which while safe would occupy
     increasing amounts of memory. Setting the value to ALWAYS ensures that
     semantics for Id reuse remain the same as for other non HA deployments.
    digitalstain committed Nov 9, 2016
Commits on Nov 8, 2016
  1. Merge pull request #8314 from mneedham/tx-pull-batch

    Tx pull batch
    digitalstain committed on GitHub Nov 8, 2016
Commits on Oct 31, 2016
  1. Merge pull request #8284 from digitalstain/3.1-fix-cc-netty-buffers

    Fix netty buffers leaks in CC message encoders
    digitalstain committed on GitHub Oct 31, 2016
  2. Merge branch '3.0' into 3.1

    Conflicts:
    	community/kernel/src/main/java/org/neo4j/kernel/impl/util/StoreUtil.java
    	enterprise/backup/src/main/java/org/neo4j/backup/BackupService.java
    	enterprise/com/src/main/java/org/neo4j/com/storecopy/StoreCopyClient.java
    	enterprise/com/src/test/java/org/neo4j/com/storecopy/StoreCopyClientTest.java
    	enterprise/ha/src/main/java/org/neo4j/kernel/ha/BranchedDataPolicy.java
    	enterprise/ha/src/main/java/org/neo4j/kernel/ha/cluster/SwitchToSlave.java
    	enterprise/ha/src/test/java/org/neo4j/kernel/ha/cluster/SwitchToSlaveBranchThenCopyTest.java
    digitalstain committed Oct 31, 2016
  3. Merge pull request #8234 from mneedham/3.0-biased-winner-strategy

    3.0 biased winner strategy
    digitalstain committed on GitHub Oct 31, 2016