Discussed with @tinwelint. I had misunderstood the lucene APIs and this change looks sensible.

that should be up to the SearcherManager. Closing the IndexReader manually
and then trying to release the searcher holding will fail and consecutive
calls to SearcherManager#acquire will get stuck in busy-loop.

This solves a problem that caused the index population phase to hang after
the node scan was complete, when flipping to ONLINE state and there were
one or more updates to apply before flipping. Having a constraint creation
hang caused the schema write lock be held indefinitely.

effectively removing an (almost) duplicated line. Introduced by
46c41198

1.9 merge

Conflicts:
	community/kernel/src/main/java/org/neo4j/kernel/InternalAbstractGraphDatabase.java
	community/kernel/src/main/java/org/neo4j/kernel/impl/transaction/TxLog.java
	community/kernel/src/main/java/org/neo4j/kernel/impl/transaction/TxManager.java
	community/kernel/src/main/java/org/neo4j/kernel/impl/transaction/xaframework/InterceptingXaLogicalLog.java
	community/kernel/src/main/java/org/neo4j/kernel/impl/transaction/xaframework/XaFactory.java
	community/kernel/src/main/java/org/neo4j/kernel/impl/transaction/xaframework/XaLogicalLog.java
	community/kernel/src/main/java/org/neo4j/kernel/impl/transaction/xaframework/XaResourceManager.java
	community/kernel/src/test/java/org/neo4j/kernel/impl/index/TestIndexCommand.java
	community/kernel/src/test/java/org/neo4j/kernel/impl/nioneo/store/TestNeoStore.java
	community/kernel/src/test/java/org/neo4j/kernel/impl/nioneo/store/TestXa.java
	community/kernel/src/test/java/org/neo4j/kernel/impl/transaction/TestXaFramework.java
	community/kernel/src/test/java/org/neo4j/kernel/impl/transaction/TxManagerTest.java
	community/kernel/src/test/java/org/neo4j/kernel/impl/transaction/xaframework/TestStandaloneLogExtractor.java
	community/kernel/src/test/java/org/neo4j/kernel/impl/transaction/xaframework/TestUpgradeOneDotFourToFiveIT.java
	community/kernel/src/test/java/org/neo4j/kernel/impl/transaction/xaframework/XaLogicalLogTest.java
	community/lucene-index/src/test/java/org/neo4j/index/impl/lucene/TestLuceneDataSource.java
	community/lucene-index/src/test/java/org/neo4j/index/impl/lucene/TestRecovery.java
	enterprise/backup/src/main/java/org/neo4j/backup/BackupServer.java
	enterprise/backup/src/main/java/org/neo4j/backup/BackupService.java
	enterprise/backup/src/main/java/org/neo4j/backup/RebuildFromLogs.java
	enterprise/cluster/src/main/java/org/neo4j/cluster/protocol/atomicbroadcast/multipaxos/MultiPaxosContext.java
	enterprise/com/src/main/java/org/neo4j/com/Client.java
	enterprise/com/src/main/java/org/neo4j/com/Server.java
	enterprise/com/src/test/java/org/neo4j/com/MadeUpClient.java
	enterprise/com/src/test/java/org/neo4j/com/MadeUpServer.java
	enterprise/com/src/test/java/org/neo4j/com/ServerTest.java
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/BranchDetectingTxVerifier.java
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/HaRequestType196.java
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/HighlyAvailableGraphDatabase.java
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/MasterClient196.java
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/cluster/zoo/ZooClient.java
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/com/master/MasterServer.java
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/com/slave/MasterClient153.java
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/com/slave/MasterClient17.java
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/com/slave/MasterClient18.java
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/com/slave/MasterClientResolver.java
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/com/slave/SlaveServer.java
	enterprise/ha/src/test/java/org/neo4j/ha/TransactionConstraintsIT.java
	enterprise/ha/src/test/java/org/neo4j/test/ha/ClusterManager.java

Introduction of monitoring in the kernel and HA network operations 

Fixes an issue with phase2Timeout on proposer state

Adds two tests that make use of TransactionMonitor to monitor transaction
 propagation in clusters. One test counts events by committing and pushing
 from both slaves and the master, while the other counts transactions
 pulled from slaves.

Introduces TransactionMonitor to facilitate measurement of transaction
 events, current commits and commit entry injections.
Adds support for that in XaResourceManager, tagged by datasource name.
Introduces a sample counting implementation meant for testing,
 EideticTransactionMonitor.
Introduces tests for basic functionality in kernel and lucene-index

TaggedMonitorListenerInvocationHandler was not getting the tags that
 it was created with. The result was that listeners added for specific tags
 would get called for all invocations of the observed interface regardless.
 This commit fixes the issue and adds a test for it.

ByteCounterMonitor is generic enough so it does not have to be
 in o.n.k.i.t.xaframework anymore

BlockLogBuffer now updates a ByterCounterMonitor with the bytes it
 writes to the underlying buffer. This way all operations that
 serialize transactions to the network can be monitored.
Master and Client instances now make use of that BlockLogBuffer
 for streaming results and transactions.
Introduces a RequestMonitor which logs HA requests. It supports
 request context via a hashmap which can be set to contain various
 information about the request logged.
Client and Server now update a RequestMonitor with requests sent
 and received, respectively.

LogExtractor now updates a ByteCounterMonitor with bytes read while
 extracting transaction entries from a log
XaLogicalLog updates a ByterCounterMonitor when deserializing
 prepared transactions from disk for sending over the network
XaLogicalLog now updates a ByteCounterMonitor when deserializing
 a transaction stream over the network

Ha protocols from versions 1.5.3 and 1.7 are no longer supported for
 rolling upgrades. Hence they are no longer necessary to be present
 in the codebase.

It is clear after some usage that log buffer monitoring is really
 a case of counting bytes, functionality that needs to be replicated in
 other places as well. It makes sense therefore to have LogBufferMonitor
 strictly count bytes, rename it to ByterCounterMonitor and reuse it
 where necessary.

Uses the Monitors functionality to add bytes written counting
 facilities to DirectMappedLogBuffer. Achieved via introducing
 a LogBufferMonitor interface that is used by DMLB.
XaLogicalLog and TxLog require access to Monitors to
 create the proper monitor and pass it to their DMLB.

There was only one implementation of LogBufferFactory which
 was a straightforward call to DMLB constructor. This did
 not call for the need for a factory abstraction and since
 it was complicating upcoming work on monitoring it has now
 been removed.

Fixes #1876

Also fix false index collisions for online indexes

The story so far:

* Indexes store all numbers as doubles, and that means that large long values
  can coerce to the same double values and cause collisions in the index.
  We fixed this for indexes that are populating, by making them read colliding
  values out from the property store to double check if the collision is real
  or not.
* We also added a similar "exact match" filtering to the index lookup
  operations on StateHandlingStatementOperations, to fix index lookups.
* What we didn't fix in the first round, was adding new stuff to indexes that
  are online. This is the bug that Mattias found.
* Specifically, the UniquePropertyIndexUpdater verifies that the constraint is
  still holding in its close method.
* This is super bad, because the close method is called in commit - not in
  prepare, not during the transaction - in commit. Huge no-no.
* Initially you'd think that this puts recovery at risk, but that turns out not
  to be the case, because during recovery, we use a normal index updater
  instead of the unique one.
* So this means that the following scenario was possible: Add two nodes with
  similar but distinct long values to the index. The
  ConstraintEnforcingEntityOperations is fine with this because the values are
  really different. However, the UniquePropertyIndexUpdater trashes the
  transaction in commit. Now suppose the entire database server crashes for
  whatever reason, and we must do recovery. We recover the index using the
  normal updater that just adds the values without checking any constraints.
  And this is actually totally fine, because we filter for exact matches in the
  StateHandlingStatementOperations. And so only then does the index work as
  advertised. We had to crash the database in order to add a valid value to the
  index. Except, there is a window between the crash of the transaction, and
  the crash of the database, where the transaction could have been retried with
  the same exact value and encounter the same exact crash. Now we recover both
  transactions, and boom: we got a duplicate in the unique index!
* The solution I'm going for currently is to just not verify the constraint in
  the UniquePropertyIndexUpdater.close method... just accept the value as
  given, and trust that the ConstraintEnforcingEntityOperations has done it's
  job.
* Implementing this solution is made a bit harder by a number of tests
  operating directly with the IndexUpdaters, instead of checking the publicly
  observable behaviour.
* I intend to rephrase these tests in terms of higher level APIs, but in the
  mean time (because there are a lot of cases to cover) we will have to ignore
  all the now failing tests, and see what the robustness suite says.

When an instance times out on phase 2, if it is closed or delivered
 it should have its payload retried. This patch adds this functionality.

Removes the ComException wrapping of server-side BranchedDataException

Making the reporting of BiasedWinnerStrategy a bit easier to

to be instantiable where a Logging dependency is accepted, instead of
printing to System.out

o This is a 1.9 version - lots of classes moved around so I've
  submitted a 2.0 version as well
o Changed wording so it describes it in words that will be easier
for the field team to understand
o Added tests around the BiasedWinnerStrategy

Merge branch 'neo4j/1.9-maint' into master

which is no longer needed due to recent changes in exception streaming
from server to client. In fact the wrapping prevented a slave to properly
react to branched data and would make that slave unable to ever join the
cluster.

Conflicts:
	enterprise/ha/src/main/java/org/neo4j/kernel/ha/cluster/HighAvailabilityModeSwitcher.java

The test MonitorsTest.shouldUnregister() was failing with a ClassCastException.
 The reason was that it was doing the cast on a field different than the one that
 it was performing the instanceof check.

Conflicts:
	community/kernel/src/test/java/org/neo4j/kernel/monitoring/MonitorsTest.java

Adding System.out of which processes gets dumped, for debugging purposes. It's OK since this class is used for in testing/debugging

The test MonitorsTest.shouldUnregister() was failing with a ClassCastException.
 The reason was that it was doing the cast on a field different than the one that
 it was performing the instanceof check. This error is now fixed and the test
 passes.