controller.git
6 years agoBug 8444 - Persistent prefix-based shard cannot load its snapshot 75/57075/5
Jakub Morvay [Mon, 15 May 2017 15:13:53 +0000 (17:13 +0200)]
Bug 8444 - Persistent prefix-based shard cannot load its snapshot

Since the name is URL-encoded, we have to make sure it does not get
double-encoded -- hence we need to make a pass of URL-decoding before
we use the result.

Change-Id: I20fe8702ad7e405a8b68d8bda2f9ce4522f2dfd0
Signed-off-by: Jakub Morvay <jmorvay@cisco.com>
6 years agoFix logging format/argument mismatch 61/57061/2
Robert Varga [Mon, 15 May 2017 13:10:26 +0000 (15:10 +0200)]
Fix logging format/argument mismatch

Two debug sites fail to pass down shardName, leading to mal-formatted
log messages.

Change-Id: I5521539c54c2e1f7ef5ef25d9a47fbc6d6d0a27c
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoBUG-8056: place an upper bound on number of transactions processed 48/56948/2
Robert Varga [Fri, 12 May 2017 13:24:07 +0000 (15:24 +0200)]
BUG-8056: place an upper bound on number of transactions processed

When transactions complete their preCommit step immediately we end
up scheduling the next transaction immediately in the call stack,
hence if that completes immediately we end up eating away our stack
until we hit StackOverflowError.

Limit the number of transactions we process as a reaction to a single
message so that stack usage is under control. Should we hit this
limit, schedule a contiuation, which will deal with the rest of
the transactions.

Change-Id: Iad2812c823bd8e91ad45020ac50f6a8626654afb
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoBUG-8422: Propagate enqueue time 47/56747/7
Robert Varga [Tue, 9 May 2017 21:55:31 +0000 (23:55 +0200)]
BUG-8422: Propagate enqueue time

When we are replaying requests onto a connection we really want
to leave their enqueue times intact, so they time out properly.
This codepath is specific for the replay case, hence we do not
want to incur any waiting, either.

This patch introduces enqueueRequest() which does not wait for
the queue duration and audits code paths so they end up talking
to the right method -- either enqueueRequest() or sendRequest().

Change-Id: Ibf97dcc11e32d9ffa911c78ccf0448d6891a9cac
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoBUG 8318: Add a separate dispatcher for cluster actors 28/56928/4
Tomas Cere [Fri, 12 May 2017 09:59:55 +0000 (11:59 +0200)]
BUG 8318: Add a separate dispatcher for cluster actors

When the system is under load it seems like there can
be a missed heartbeat leading to false positives for unreachable
nodes. Run the actors responsible for heartbeats on a separate
dispatcher to avoid this.

Change-Id: Ib4f4225bf69e99d93e3c7010d6fbe1163b96a5a2
Signed-off-by: Tomas Cere <tcere@cisco.com>
6 years agoFix testLeaderAndFollowerEntityOwnersReassignedAfterShutdown failure 09/56909/2
Tom Pantelis [Thu, 11 May 2017 11:32:39 +0000 (07:32 -0400)]
Fix testLeaderAndFollowerEntityOwnersReassignedAfterShutdown failure

14:19:15 Failed tests:
14:19:15   DistributedEntityOwnershipIntegrationTest.testLeaderAndFollowerEntityOwnersReassignedAfterShutdown:439->lambda$testLeaderAndFollowerEntityOwnersReassignedAfterShutdown$1:440 Raft state expected:<[Leader]> but was:<[Candidate]>

After the leader is shut down, member-2 is supposed to be elected leader but,
in this case, it didn't get the vote from member-5 b/c member-2 had not yet
received the MemberUp for member-5 and thus did not have its peer actor address.
So I made changes to ensure member data stores are ready and members are up.

I also saw a failure where member-3 or member5 didn't grant the vote for member-2
b/c it's last log index was greater. This can happen is member-2 didn't was a bit
behind when the leader was shut down. So I changes it to obtain the leader's last
index and verify the remaining follower's last log index is up-to-date.

Change-Id: Ib5ad2e135bb3809e1c62a432a029a5a56109190d
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
6 years agoBug 8424: Don't output data tree and tree candidates wih debug 06/56906/1
Tom Pantelis [Thu, 11 May 2017 16:31:27 +0000 (12:31 -0400)]
Bug 8424: Don't output data tree and tree candidates wih debug

Data trees and tree candidates can get quite large and fill up the
log files when debug is on and, worst case, cause OOM errors. For
debug logging, only print the whole tree/candidate with trace logging.
In cases where an Optional data tree is outputted via toString, only
output if the data tree is present or not.

Change-Id: I6cb5f9a5da9e3cc3218c83bb103b673db0fb1d80
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
(cherry picked from commit 5083d20c7f49788e64ca9080aebcded623d7aa44)

6 years agoBUG-8392 RpcRegistry has it's buckets populated by unreachable nodes 83/56783/4
Tomas Cere [Wed, 10 May 2017 12:04:37 +0000 (14:04 +0200)]
BUG-8392 RpcRegistry has it's buckets populated by unreachable nodes

In a situation when a member(f.ex member2) is isolated and the rpc registrations
are removed from the node(member1) we can still have our bucket store populated
by buckets from the remaining node(member-3) which might not have received
the memberUnreachable message yet leadig to stale routing of an rpc to
member-2.
This patch adds bucket filtering based on the currently present peers
so that we only accept Buckets that we can see.

Change-Id: I92c1e063f4754aca829bd73df4518f859e1d8497
Signed-off-by: Tomas Cere <tcere@cisco.com>
6 years agoBug 8385: Fix testMultipleRegistrationsAtOnePrefix failure 74/56774/2
Tom Pantelis [Tue, 9 May 2017 12:29:07 +0000 (08:29 -0400)]
Bug 8385: Fix testMultipleRegistrationsAtOnePrefix failure

The previous patch added a callback on the Future returned by
gracefulStop on shard removal. The timout was set to 3 * election timeout
which is 30 s in production by default. For the tests the election
timeout is 500 ms so the timeout is 1500 ms. However, if the timing is right,
the leader may not be able to transfer leadership on shutdown if the other
member was already shutdown. On shutdown there's a 2 sec wait to hear from
a new leader - this is greater than the 1500 ms shutdown timeout which
leads to test failure. To alleviate this, I made 10 s the minimum for the
shutdown timeout.

Another problem was that, after the stop future failed, the OnComplete
callback for PrefixShardCreated was repeated many times before the
OnComplete callback queued the message to remove the Future from the map.
To alleviate this, I added a CompositeOnComplete containing a list of
deferred OnComplete tasks. This allows the control to remove the entry
from the map before the deferred tasks run.

Change-Id: I899518e6d7e92533d2c4008a978ac772b02863cf
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
(cherry picked from commit 6ef8a6b4e403d5908e7090a5bd387f81c10c91c6)

6 years agoFix testTransactionForwardedToLeaderAfterRetry failure 69/56769/1
Tom Pantelis [Wed, 10 May 2017 07:00:09 +0000 (03:00 -0400)]
Fix testTransactionForwardedToLeaderAfterRetry failure

java.util.concurrent.ExecutionException: ReadFailedException{message=Error executeRead ReadData for path /(urn:opendaylight:params:xml:ns:yang:controller:md:sal:dom:store:test:cars?revision=2014-03-13)cars/car, errorList=[RpcError [message=Error executeRead ReadData for path /(urn:opendaylight:params:xml:ns:yang:controller:md:sal:dom:store:test:cars?revision=2014-03-13)cars/car, severity=ERROR, errorType=APPLICATION, tag=operation-failed, applicationTag=null, info=null, cause=org.opendaylight.controller.md.sal.common.api.data.DataStoreUnavailableException: Shard member-1-shard-cars-testTransactionForwardedToLeaderAfterRetry currently has no leader. Try again later.]]}

The test submits transactions and deposes the current leader so it forwards the
pending transactions to the other member-2 that assumes leadership. However it calls

   Cluster.get(followerSystem).leave(MEMBER_1_ADDRESS);

which may result in an untimely MemberExited message sent to the ShardManager that
clears the peer address, causing the FindPrimary message to fail to find the leader.
I'm not clear why this was call was put in but it's unnecessary and may cause a
failure if the timing is right.

I also saw a failure due to a timeout when forwarding a pending transaction. This is
b/c it takes some time for member-2 to switch to candidate and become leader due to
the checking of current leader availability via the akka cluster on ElectionTimout.
If it takes too long the pending transaction forwarding may time out. To alleviate
this, I forced the swicth to candidate by sending an immediate TimeoutNow message.

Change-Id: I2dd228964779e2b755b1740a518e2c400b5cb88d
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
(cherry picked from commit e345c2a17f737d537cda45b0f737dff417e3b359)

6 years agoFix timing issue in PartitionedCandidateOnStartupElection*Test 99/56699/2
Tom Pantelis [Mon, 8 May 2017 16:49:51 +0000 (12:49 -0400)]
Fix timing issue in PartitionedCandidateOnStartupElection*Test

If the initial AppendEntries sent by the leader (member 1) to member 3
is delayed enough such that the behavior field in MemberActor is already
set by the test code, the AppendEntries message will be forwarded to the
Candidate behavior and yield incorrect results for the test. To prevent this,
we really shouldn't set and access the behavior field directly but instead
do so via messages to maintain actor encapsulation.

Change-Id: If497583ce648e62e3279e5abff19cb8702943c17
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
6 years agoBug 8391 - Close producer in become-prefix-leader rpc implementation 98/56698/1
Jakub Morvay [Mon, 8 May 2017 11:58:57 +0000 (13:58 +0200)]
Bug 8391 - Close producer in become-prefix-leader rpc implementation

MdsalLowLevelTestProvider's become-prefix-leader rpc implementation
creates CDSDataTreeProducer to try to move shard leadership. However,
the producer is not closed after leadership change request. This
prevents any subsequent invocations of become-prefix-leader rpc with
same prefix parameter to be successful. Subtree specified by the prefix
is attached to still opened producer and creation of any new producer
for this subtree fails. Close producer once we don't need it.

Change-Id: I3827e425082c35a43ec18dac1ef0f2dbd19b291f
Signed-off-by: Jakub Morvay <jmorvay@cisco.com>
6 years agoBUG-8372: fix AbstractProxyTransaction.replayMessages() 36/56636/6
Robert Varga [Fri, 5 May 2017 12:35:32 +0000 (14:35 +0200)]
BUG-8372: fix AbstractProxyTransaction.replayMessages()

This method made assumptions on whan requests can be present in the
queue -- notably that local requests are never encountered. This is
not true, as local requests can be present here due to being in-flight
when reconnect occurs.

Change-Id: Ia5b6ec442c014329046bf384a0f5ea97666a2c4a
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoBug 8385 - @Ignore testMultipleRegistrationsAtOnePrefix 60/56660/2
Jakub Morvay [Mon, 8 May 2017 08:01:29 +0000 (10:01 +0200)]
Bug 8385 - @Ignore testMultipleRegistrationsAtOnePrefix

DistributedShardedDOMDataTreeRemotingTest.testMultipleRegistrationsAtOnePrefix
is failing intermittently - set it to ignore for now.

Change-Id: I3e8aec2bfbe97559525051805170203574472aab
Signed-off-by: Jakub Morvay <jmorvay@cisco.com>
6 years agoBUG-8372: fix abort message confusion 96/56596/3
Robert Varga [Fri, 5 May 2017 11:46:22 +0000 (13:46 +0200)]
BUG-8372: fix abort message confusion

Immediate transaction aborts need to use the appropriate message,
not 3PC's TransactionAbortRequest.

Change-Id: I9e25e3f20ed62fc520853685af17accef35c1bb4
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoBUG-8372: improve forward/replay naming 93/56593/3
Robert Varga [Fri, 5 May 2017 10:04:12 +0000 (12:04 +0200)]
BUG-8372: improve forward/replay naming

There is a bit of confusion between 'replay' and 'forward' methods.
They serve two distinct purposes:
- 'replay' happens during reconnect, i.e. for requests that have
           already entered the connection queue and have paid
           the delay cost, so they should not pay it again.
- 'forward' happens after reconnect for requests that have raced
            with the reconnect process, i.e. they need to hop from
            the old connection to the new one. These need to enter
            the queue and pay the delay cost.

This patch cleans the codepaths up to use consistent naming, making
it clearer that the problem we are seeing is in the 'replay' path.

Change-Id: Id854e09a0308f8d0a9144d59f41e31950cd58665
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoBug 8385: Fix testMultipleRegistrationsAtOnePrefix failures 34/56634/1
Tom Pantelis [Fri, 5 May 2017 10:41:59 +0000 (06:41 -0400)]
Bug 8385: Fix testMultipleRegistrationsAtOnePrefix failures

The test quickly creates/removes the prefix shard in iterations which
can result in an InvalidActorNameException if the shard actor from the prior
iteration hadn't been destroyed yet. To alleviate this I modified the
removal in the ShardManager to utilize Patterns.gracefulStop to store the
Future and block a subsequent create until the Future completes.

Change-Id: Ica98de3cc17c2d87195840bdf052d81ed3b9dd10
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
6 years agoBUG-8372: add dataTree information 39/56539/4
Robert Varga [Thu, 4 May 2017 15:56:33 +0000 (17:56 +0200)]
BUG-8372: add dataTree information

We are attempting to send a local message to a remote actor,
which seems to be a mixup with data tree presence. Add dataTree
to toString() output so we know which connections resolve to
being local and which to being remote.

Change-Id: If1ed3cfdea24148456a4d310949fb480791c1ffa
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoMake the last submit timeout after 30 seconds 89/56589/3
Tomas Cere [Fri, 5 May 2017 10:11:17 +0000 (12:11 +0200)]
Make the last submit timeout after 30 seconds

The low level test was waiting indefinetly for submits
to finish, change this to block and timeout after one minute
in case there's an unrecoverable failure on the backend which
doesnt propagate to the frontend.

Change-Id: I3df2465b56c701c88341ab6cc7fa37a015f1c893
Signed-off-by: Tomas Cere <tcere@cisco.com>
6 years agoFix DistributesShardedDOMDataTree.ProxyProducer's getShardAccess 99/56399/5
Jakub Morvay [Tue, 2 May 2017 16:54:52 +0000 (18:54 +0200)]
Fix DistributesShardedDOMDataTree.ProxyProducer's getShardAccess

DistributesShardedDOMDataTree.ProxyProducer's getShardAccess works only
for subtrees that are rooted at some registered prefix based shard.
Moreover subtree has to be one of the subtrees specified in
DistributedShardedDOMDatatTree's createProducer method.

This is way more strict than what is required by CDSDataTreeProducer's
API. Pass ProxyProducer's implementation current shard layout, so
producer can lookup corresponding shard for specified subtree in
getShardAccess method. One-to-one mapping between shards and subtrees
is no longer required.

Change-Id: I765567d34c803a85b4be8a6e10fd81b6f64a1610
Signed-off-by: Jakub Morvay <jmorvay@cisco.com>
6 years agoFix logger formatting strings 36/56536/2
Robert Varga [Thu, 4 May 2017 15:51:43 +0000 (17:51 +0200)]
Fix logger formatting strings

Fix %s/{} mixups.

Change-Id: I916996e17839a61802a83ddff31d162ac662f934
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoMove initial list creation to create-prefix-shard. 68/56468/3
Tomas Cere [Wed, 3 May 2017 13:21:00 +0000 (15:21 +0200)]
Move initial list creation to create-prefix-shard.

This move the initial list population of produce-transactions
to create-prefix-shard rpc with 3 hardcoded prefixes(prefix-1,prefix-2,prefix-3)
so that csit suites can populate the id-int list just once when the shard is created
and produce-transactions can now run parallely on multiple entries from
multiple nodes.

Change-Id: If70990c0e217cd68027ae960a7545c69acf52cdb
Signed-off-by: Tomas Cere <tcere@cisco.com>
6 years agoFix Eclipse warnings in config-manager 09/56509/2
Robert Varga [Wed, 3 May 2017 13:34:13 +0000 (15:34 +0200)]
Fix Eclipse warnings in config-manager

Change-Id: I0ed9bc52d4cf4e5ee7a4da8bd53355191326cba6
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
(cherry picked from commit 405c97ddf445b6057a2654545dd8072d28eedcce)

6 years agoBug 8328 - Create prefix shards with correct peers 77/56177/6
Jakub Morvay [Thu, 27 Apr 2017 15:21:42 +0000 (17:21 +0200)]
Bug 8328 - Create prefix shards with correct peers

Change-Id: I068b38bb275d23d27559aec3f336a6b9081fb732
Signed-off-by: Jakub Morvay <jmorvay@cisco.com>
6 years agoBug 8336 - Fix NPE in DistributedShardedDOMDataTree's ProxyProducer 52/56352/3
Jakub Morvay [Tue, 2 May 2017 08:32:20 +0000 (10:32 +0200)]
Bug 8336 - Fix NPE in DistributedShardedDOMDataTree's ProxyProducer

Change-Id: If0060e6e2696674bc5418d2f2a80ad0d01327e29
Signed-off-by: Jakub Morvay <jmorvay@cisco.com>
6 years agoBUG 8301: Convert queue to a local variable 56/56456/3
Tomas Cere [Wed, 3 May 2017 11:12:31 +0000 (13:12 +0200)]
BUG 8301: Convert queue to a local variable

There's a possibility that this might race and an
actor can have it's queue overwritten by another thread, so convert
this to a local variable.

Change-Id: Ic84922c6d109d8361a48debbf971fddd9cee1d3e
Signed-off-by: Tomas Cere <tcere@cisco.com>
6 years agoBug 8342: Add info logging to ConfigManagerActivator 06/56406/2
Tom Pantelis [Tue, 2 May 2017 19:16:44 +0000 (15:16 -0400)]
Bug 8342: Add info logging to ConfigManagerActivator

Change-Id: I7b01961910dd2ba7ed9a421ee52e0aec29c68ade
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
6 years agoBUG-8342: force config-manager startup 37/56437/1
Robert Varga [Wed, 3 May 2017 08:51:32 +0000 (10:51 +0200)]
BUG-8342: force config-manager startup

config-manager needs to be pretty much the first thing that comes
up due to historic reasons. Assign it a low start level so it
activates before the blueprint extension.

Change-Id: I2d0a3706843409e8a22f9064f27e47cc0df46c95
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoDowngrade most info messages in benchmarks 18/56118/2
Vratko Polak [Wed, 26 Apr 2017 18:37:43 +0000 (20:37 +0200)]
Downgrade most info messages in benchmarks

They create spam during CSIT,
making real errors less noticable in log.

Change-Id: Icf00389526919751e88189ffef1be70e16e806e8
Signed-off-by: Vratko Polak <vrpolak@cisco.com>
6 years agoBUG-8327: deprecate sal.core.api.model.SchemaService 05/56305/2
Robert Varga [Fri, 28 Apr 2017 15:03:32 +0000 (17:03 +0200)]
BUG-8327: deprecate sal.core.api.model.SchemaService

This interface is deprecated in favor of the DOMSchemaService
for the MD-SAL project.

Change-Id: Icff2cced791bc9fbf5bfadbe2f1cf2b949ff2d58
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
(cherry picked from commit 84f6c81afe8c53307dc9be8c39824ca4e4191819)

6 years agoImprove test logging in DistributedEntityOwnershipIntegrationTest 04/56304/2
Tom Pantelis [Sun, 30 Apr 2017 02:44:26 +0000 (22:44 -0400)]
Improve test logging in DistributedEntityOwnershipIntegrationTest

Some of the tests in DistributedEntityOwnershipIntegrationTest set the
datastore type to "test" which isn't helpful in identifying the output
in jenkins log archives. Use the name of the test method instead as is
done with other tests.

Change-Id: I25e40df5139a4d9f8c46d03c0f2c9c8a52fd15ee
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
(cherry picked from commit 56af1b2a45b4a567813de5314c31aaf26a2e4052)

6 years agoBUG-8327: GlobalBundleScanningSchemaServiceImpl should be a proxy 00/56300/2
Robert Varga [Fri, 28 Apr 2017 14:56:48 +0000 (16:56 +0200)]
BUG-8327: GlobalBundleScanningSchemaServiceImpl should be a proxy

We are currently running to separate services which assemble
the GlobalSchemaContext, which hurts our startup performance and
leads to wasted memory. This is an artefact of the mdsal split,
hence we should be getting the service from the MD-SAL and
just proxy to old interfaces.

This lowers the startup time for

feature:install odl-restconf odl-bgpcep-bgp
    odl-bgpcep-data-change-counter odl-netconf-topology

from 86s down to 67s (22%). Final retained heap size is also
lowered from 217MiB to 181MiB (16%)

Change-Id: I549e9512538bd83d86cfd2164d03e34bc9130c1e
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoBug 8303: BP odl:clustered-app-config initial/*-config.xml testability 23/56123/3
Michael Vorburger [Tue, 25 Apr 2017 22:32:15 +0000 (00:32 +0200)]
Bug 8303: BP odl:clustered-app-config initial/*-config.xml testability

DataStoreAppConfigDefaultXMLReaderTest illustrates usage.

Change-Id: I342fca4583c90802238e63262871e33b4b713438
Signed-off-by: Michael Vorburger <vorburger@redhat.com>
(cherry picked from commit 821944277049bbb3949021626844ef7a80101f70)

6 years agoAdd more debug logging for DTCL registration/notification code paths 93/56293/1
Tom Pantelis [Thu, 27 Apr 2017 11:23:34 +0000 (07:23 -0400)]
Add more debug logging for DTCL registration/notification code paths

Added logging so the listener instance and actor can be traced
end-to-end from FE registration to the BE publisher actor.

Also added log context to some classes to identify which shard it
belongs to.

Change-Id: I1a26fb8775a57e0fc563eceec919d50395f4ceb1
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
6 years agoBug 8337: Ignore testMultipleShardLevels 85/56285/1
Tom Pantelis [Sat, 29 Apr 2017 01:43:34 +0000 (21:43 -0400)]
Bug 8337: Ignore testMultipleShardLevels

DistributedShardedDOMDataTreeTest.testMultipleShardLevels is
failing intermittently - set it to ignore for now.

Change-Id: Ib7f86166fd85cd54e6ec8cac106c993e9407ffea
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
6 years agoBug 8301: Fix some issues with testProducerRegistrations 58/56158/2
Tom Pantelis [Thu, 27 Apr 2017 10:45:11 +0000 (06:45 -0400)]
Bug 8301: Fix some issues with testProducerRegistrations

The LogicalDatastoreType.CONFIGURATION type was being used for both data
stores - modified the IntegrationTestKit to set the logicalStoreType
appropriately.

Fixed a synchronization issue in DistributedShardedDOMDataTree#lookupShardFrontend
where it accessed shards unprotected.

Change-Id: I628add86667e4a812f8e7516bac59f9b66fe4033
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
(cherry picked from commit 64b9093c2c9ce670b92f2d0ea44c76dc9a385b5a)

6 years agoBUG-7927: stop scanning bundles on framework stop 35/56235/2
Robert Varga [Thu, 27 Apr 2017 15:36:37 +0000 (17:36 +0200)]
BUG-7927: stop scanning bundles on framework stop

Monitor framework bundle for STOPPING event and when it triggers
flag us as stopping: all bundles are about to shut down, so there
is no point in trying to update the schema context anymore.

Change-Id: I1a55169fce1705c19a139063cf632674fc256701
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
(cherry picked from commit 6a79e55d2b6462cd609ab8cd5766fd4222c18c4f)

6 years agoFix read-only abort message mismatch 64/56064/5
Robert Varga [Wed, 26 Apr 2017 12:39:23 +0000 (14:39 +0200)]
Fix read-only abort message mismatch

Testing has revealed:

WARN  | FrontendReadOnlyTransaction | Rejecting unsupported request ModifyTransactionRequest{target=member-2-datastore-config-fe-0-txn-2-0, sequence=1, replyTo=Actor[akka.tcp://opendaylight-cluster-data@10.29.15.184:2550/user/$a#585956314], operations=[], protocol=ABORT}

This is a thinko on the part of which message does what:

TransactionAbortRequest is dedicated for 3PC doAbort phase, hence
it is never seen for read-only transactions.

The message corresponding to an abort is either
AbortLocalTransactionRequest or ModifyTransactionRequest with protocol
set to ABORT.

Change-Id: I3238ade7b9f7933e6538742354888d182f599412
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoPropagate shard name to FrontendClientMetadataBuilder 73/56073/5
Robert Varga [Wed, 26 Apr 2017 13:24:37 +0000 (15:24 +0200)]
Propagate shard name to FrontendClientMetadataBuilder

Prefixing log message with shard name is useful to track things
down. Pass the shard name down from FrontendMetadata, so we can
emit such messages.

Change-Id: Ie6a2cd218e1a2686f8cc14f67574f245e3de914b
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoBug 8116 - Make DistributedShardChangePublisher agnostic to data tree change events... 61/56061/2
Jakub Morvay [Mon, 24 Apr 2017 13:21:29 +0000 (15:21 +0200)]
Bug 8116 - Make DistributedShardChangePublisher agnostic to data tree change events ordering

DistributedShardChangePublisher allows for registering DCTLs on
DistributedShardFrontend. Internally, DistributedShardChangePublisher
sets up DataTreeChangeListenerProxies on respective backend shard and
also on all of its backend subshards. Upon receiving data tree change
events from backend shards, DistributedShardChangePublisher updates
its own data tree. With the help of this tree, it finally constructs
data tree chnage events for registered DCTLs.

DistributedShardChangePublisher relies on specific ordering of backend
shards data tree change events. If it receives subshard's data tree
change event prior to current shard data tree change event, updating
internal data tree can fail. Subshard's data tree change event can
expect some changes from its parent shard.

Clearly, we don't have control on ordering of these events. Do not rely
on this. If we cannot apply subshard's change to data tree, cache it
and try to apply it once we have also its parent's change.

Change-Id: I3bd9b2d217d01974bce02465529c6cdbf8c3d633
Signed-off-by: Jakub Morvay <jmorvay@cisco.com>
(cherry picked from commit b73d37a30e750c6ef7a6f6614f3b00a01b1fdd4c)

6 years agoRemove artifacts entries for long-gone RESTCONF 14/56114/2
Robert Varga [Wed, 26 Apr 2017 18:34:01 +0000 (20:34 +0200)]
Remove artifacts entries for long-gone RESTCONF

RESTCONF has been moved to its own project, hence these
artifacts entries are duds. Remove them.

Change-Id: I72d918567a04841784b0a8061ec655fe79af6ae4
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
(cherry picked from commit 09f44218611fb5a1439d1b2c7ffef401449c354b)

6 years agoBUG-7390: fix dsbenchmark 08/56108/4
Robert Varga [Wed, 26 Apr 2017 17:56:30 +0000 (19:56 +0200)]
BUG-7390: fix dsbenchmark

Benchmark tests were not consistent as to what data store they were
using, leading to flooded logs in read case because of this and
irrelevant results in the delete case.

This patch corrects the mistakes, adding at least some consistency
and hope for relevant results.

Change-Id: I0528eb42cb38eacd5e0525c0a78ada111b1edb55
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoAdd ClientBackedTransaction allocation recording 77/55977/9
Robert Varga [Tue, 25 Apr 2017 10:55:44 +0000 (12:55 +0200)]
Add ClientBackedTransaction allocation recording

This patch adds a very simple recording of where a transaction
was allocated, aiding identification of callers who fail to close
transactions.

Change-Id: I9a8743c7a38e83c855102a3a25adecfea8599dfe
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoLower AbstractNormalizedNodeDataOutput debugs to trace 05/56005/3
Robert Varga [Tue, 25 Apr 2017 14:19:17 +0000 (16:19 +0200)]
Lower AbstractNormalizedNodeDataOutput debugs to trace

Setting debug to org.opendaylight.controller.cluster.datastore
also catches the clustering-commons, leading to a lot of logs
from serialization. Lower its logging to trace.

Change-Id: Ic0e9f9c60020675c45e79c7638dcb500d6de5091
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoBUG-8309: Add message identity information 44/56044/5
Robert Varga [Wed, 26 Apr 2017 08:46:15 +0000 (10:46 +0200)]
BUG-8309: Add message identity information

We have encountered an attempt to serialize a local request across
a remote connection. Since this is hit by the akka serializer, we
have lost the identity of the call site and of the message, because
all akka is seeing is the Envelope and the exception's stack trace,
which only indicates class hierarchy up to and including
AbstractLocalTransactionRequest.

This patch enriches the exception message so we know what the actual
request was, hopefully pinpointing the offending call site. Since
the problem revolves around the reconnect process, bump critical
transitions to info instead of debug.

Change-Id: I6d6d6e702d4b5baff7b707242583e923708e7637
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoNest id-ints list inside a container 95/55995/2
Tomas Cere [Tue, 25 Apr 2017 10:50:52 +0000 (12:50 +0200)]
Nest id-ints list inside a container

Needs to be nested to be able to refer to the whole list via restconf
and instance-identifier yang element, so update the model and the handlers
to account for this change.

Change-Id: Idf50de5e6faa9757f45ec68e9b796ae0742f6aa9
Signed-off-by: Tomas Cere <tcere@cisco.com>
6 years agoBug 8301: Disable DistributedShardedDOMDataTreeRemotingTest for now 20/56020/2
Tom Pantelis [Tue, 25 Apr 2017 17:56:50 +0000 (13:56 -0400)]
Bug 8301: Disable DistributedShardedDOMDataTreeRemotingTest for now

Change-Id: I24068c5ee92533cdc23174d17cc1805328df7c4d
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
6 years agoFix intermittent failure in testLeadershipTransferOnShutdown 17/56017/2
Tom Pantelis [Tue, 25 Apr 2017 17:26:27 +0000 (13:26 -0400)]
Fix intermittent failure in testLeadershipTransferOnShutdown

10:03:06 java.util.concurrent.ExecutionException: ReadFailedException{message=Error executeRead ReadData for path /(urn:opendaylight:params:xml:ns:yang:controller:md:sal:dom:store:test:cars?revision=2014-03-13)cars/car, errorList=[RpcError [message=Error executeRead ReadData for path /(urn:opendaylight:params:xml:ns:yang:controller:md:sal:dom:store:test:cars?revision=2014-03-13)cars/car, severity=ERROR, errorType=APPLICATION, tag=operation-failed, applicationTag=null, info=null, cause=org.opendaylight.controller.md.sal.common.api.data.DataStoreUnavailableException: Shard member-2-shard-cars-testLeadershipTransferOnShutdown currently has no leader. Try again later.]]}
10:03:06  at org.opendaylight.yangtools.util.concurrent.MappingCheckedFuture.wrapInExecutionException(MappingCheckedFuture.java:64)
10:03:06  at org.opendaylight.yangtools.util.concurrent.MappingCheckedFuture.get(MappingCheckedFuture.java:92)
10:03:06  at org.opendaylight.controller.cluster.datastore.DistributedDataStoreRemotingIntegrationTest.verifyCars(DistributedDataStoreRemotingIntegrationTest.java:215)
10:03:06  at org.opendaylight.controller.cluster.datastore.DistributedDataStoreRemotingIntegrationTest.testLeadershipTransferOnShutdown(DistributedDataStoreRemotingIntegrationTest.java:928)

From the logs it seems member-2 hadn't gotten MemberUp for member-3 after the
leader transfer and by the time it tried to read. I added calls to wait for members
to be up. After the change it ran 333 times w/o failure.

Change-Id: Ifbbf304230292f69429d3086867679effb8db01c
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
6 years agoHandle AbortLocalTransactionRequest 08/56008/1
Robert Varga [Tue, 25 Apr 2017 14:58:47 +0000 (16:58 +0200)]
Handle AbortLocalTransactionRequest

When local transactions are aborted from the frontend, it is done
via a dedicated message which we failed to account for. This can
happen only as an alternative to CommitLocalTransactionRequest,
hence needs to be handled only in FrontendReadWriteTransaction.

Change-Id: I350a103f132da473d397a7d5f7de7e45850911f3
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoImprove logging around transaction lifecycle 76/55976/1
Robert Varga [Tue, 25 Apr 2017 08:39:21 +0000 (10:39 +0200)]
Improve logging around transaction lifecycle

Testing has shown that we have a gap in request handling and we
have a lot of unclosed transactions. Add logging of code paths
which trigger unsupported request.

Change-Id: I013ba8a141d5a1a9e311a8bca7842ac77064d277
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoImprove orphan transaction logging 30/55930/1
Robert Varga [Mon, 24 Apr 2017 19:51:43 +0000 (21:51 +0200)]
Improve orphan transaction logging

This patch improves logging when we perform last-resort cleanup
from garbage collector, so that the type of client handle is also
logged. This allows us to discern snapshots and snapshots.

Also lower the logging level to INFO, as this is something that
should be fixed by whoever is causing it, but it does not pose
serious threat to stability.

Change-Id: Iad55c49de87ca73f9671f04f569be7eae0e4f885
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoBUG-8219: Cleanup CompositeDataTreeCohort 19/55819/6
Robert Varga [Fri, 21 Apr 2017 14:44:10 +0000 (16:44 +0200)]
BUG-8219: Cleanup CompositeDataTreeCohort

This patch reworks the logic so we can track which cohort times
out in case that happens. We also instantiate shortcuts so we do
not go through asynchronous processing if there are no cohorts
at all.

Change-Id: I9493b768c86e8d6b2d0f4f1d13f53b13ff98fe7b
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoFix checkstyle problems not detected by the current version 14/55814/2
David [Thu, 20 Apr 2017 22:42:07 +0000 (00:42 +0200)]
Fix checkstyle problems not detected by the current version

This change is required for overall move to new Checkstyle version, see
https://git.opendaylight.org/gerrit/#/q/topic:bumpCheckstyle-stable/carbon

Most of the changes are redundant "final" modifiers.

Change-Id: I637dd46617ca144f0ed33bd705c6357493b887fe
Signed-off-by: David <david.suarez.fuentes@ericsson.com>
6 years agoBUG-8159: fix local transaction history tracking 39/55739/5
Robert Varga [Thu, 20 Apr 2017 14:37:02 +0000 (16:37 +0200)]
BUG-8159: fix local transaction history tracking

ShardCommitCoordinator needs to make sure ShardDataTree tracks
the histories involved with local transaction being submitted
via ReadyLocalTransaction. This is consistent with what we are
doing for the BatchedModifications message.

Change-Id: I02cc61476b5e02fb45f1482c4a9693bc77335793
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoRelax visibility on FrontendReadWriteTransaction methods 18/55818/2
Robert Varga [Fri, 21 Apr 2017 13:38:03 +0000 (15:38 +0200)]
Relax visibility on FrontendReadWriteTransaction methods

We are invoking these methods from anonymous subclasses, hence
keeping them private forces redirection via synthetic accessors:

 at org.opendaylight.controller.cluster.datastore.FrontendReadWriteTransaction.successfulDirectCanCommit
 at org.opendaylight.controller.cluster.datastore.FrontendReadWriteTransaction.access$300
 at org.opendaylight.controller.cluster.datastore.FrontendReadWriteTransaction$5.onSuccess

This patch makes the methods package-private, which will eliminate
the accessor, improving the stack trace.

Change-Id: Idbd803c43d7ed7333fc392a17edaf61c9721d76f
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
6 years agoBug 8274: add missing configfile dependency 12/55812/2
Stephen Kitt [Fri, 21 Apr 2017 12:35:04 +0000 (14:35 +0200)]
Bug 8274: add missing configfile dependency

odl-jolokia's configfile was missing its corresponding dependency in
the POM; this patch adds it.

Change-Id: I4e5420978020b19de58b65d06c4b2482f55351d0
Signed-off-by: Stephen Kitt <skitt@redhat.com>
6 years agoRpcRegistrar unit test 85/55785/1
matus.kubica [Wed, 5 Apr 2017 15:16:46 +0000 (17:16 +0200)]
RpcRegistrar unit test

Change-Id: I90403cb3c5fb98854c9e7dcd80ba0ce6e5f944f4
Signed-off-by: matus.kubica <matus.kubica@pantheon.tech>
Signed-off-by: Ivan Hrasko <ivan.hrasko@pantheon.tech>
7 years agoBug 7747: Reply to the leader before applying previous state 35/55735/1
Tom Pantelis [Thu, 20 Apr 2017 13:15:43 +0000 (09:15 -0400)]
Bug 7747: Reply to the leader before applying previous state

Applying state to the data tree can be expensive so the follower
should reply to the leader before applying any previous state so
as not to hold up leader consensus.

Change-Id: Ic92ae2ac30d72d6a401bdc36fda900a0a7fb21d3
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
7 years agoBUG-5280: unwrap RuntimeRequestExceptions 33/55233/3
Robert Varga [Wed, 19 Apr 2017 13:59:22 +0000 (15:59 +0200)]
BUG-5280: unwrap RuntimeRequestExceptions

This patch adds the primitive to unwrap RuntimeRequestExceptions,
so the underlying cause is propagated.

Change-Id: I77771867a48eb5f63d35a6402aca6ad0bc5b12e3
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoFix intermittent testAddShardReplicaWithAddServerReplyFailure failure 99/55699/1
Tom Pantelis [Wed, 19 Apr 2017 20:34:54 +0000 (16:34 -0400)]
Fix intermittent testAddShardReplicaWithAddServerReplyFailure failure

ShardManagerTest#testAddShardReplicaWithAddServerReplyFailure failed:

java.lang.AssertionError: assertion failed: timeout (3 seconds) during expectMsgClass waiting for class org.opendaylight.controller.cluster.raft.messages.AddServer
20:14:24  at scala.Predef$.assert(Predef.scala:170)
20:14:24  at akka.testkit.TestKitBase$class.expectMsgClass_internal(TestKit.scala:472)
20:14:24  at akka.testkit.TestKitBase$class.expectMsgClass(TestKit.scala:459)
20:14:24  at akka.testkit.TestKit.expectMsgClass(TestKit.scala:814)
20:14:24  at akka.testkit.JavaTestKit.expectMsgClass(JavaTestKit.java:415)
20:14:24  at org.opendaylight.controller.cluster.datastore.shardmanager.ShardManagerTest$33.<init>(ShardManagerTest.java:1637)

The log shows:

08:14:06,302 PM [main] [INFO] ShardManagerTest - testAddShardReplicaWithAddServerReplyFailure starting
08:14:06,325 PM [main] [INFO] ShardManager - Starting ShardManager shard-manager-config22
08:14:06,329 PM [test-akka.actor.default-dispatcher-7] [INFO] ShardManager - Recovery complete : shard-manager-config22
08:14:09,339 PM [main] [INFO] TestActorFactory - Killing actor TestActor[akka://test/user/member-1-shard-astronauts-config]
08:14:09,340 PM [main] [INFO] TestActorFactory - Killing actor TestActor[akka://test/user/shardmanager-config22]
08:14:09,340 PM [main] [DEBUG] ShardManager - Got updated SchemaContext: # of modules 1
08:14:09,340 PM [main] [DEBUG] ShardManager - shard-manager-config22: onAddShardReplica: AddShardReplica[ShardName=astronauts]
08:14:09,340 PM [main] [INFO] ShardManager - Stopping ShardManager shard-manager-config22

So the ShardManager got the onAddShardReplica message but after the test timed out
after 3 seconds. The problem is that the test is using the default dispatcher for
TestActor which is the calling thread dispatcher which is problematic for persistent
actors. Either not use TestActor where we don't need access to the underlying actor
instance or use the system default dispatcher, which is async.

Change-Id: Ib6521c345bd0db9502d0078928f8d0e5dcd7f747
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
7 years agoFix a typo 43/55243/2
Robert Varga [Wed, 19 Apr 2017 15:55:46 +0000 (17:55 +0200)]
Fix a typo

transacion -> transaction

Change-Id: I30b5b387dc9d21774798286984f67e46a2471e95
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoBUG-5280: fix snapshot accounting 38/55238/1
Robert Varga [Wed, 19 Apr 2017 15:13:58 +0000 (17:13 +0200)]
BUG-5280: fix snapshot accounting

The following warning is emitted under testing:

2017-04-19 08:49:34,707 | WARN  | ... | AbstractClientHistory            | ... | Could not find aborting transaction member-2-datastore-operational-fe-0-txn-19-0

Which is indicating that we cannot find the open transaction
inside AbstractClientHistory.

The problem is mis-routed invocation when we are taking a snapshot:
instead of going directy to subclass doCreateSnapshot() which only
allocates the transaction, invoke takeSnapshot(), which actually does
the appropriate book-keeping.

Change-Id: I07473f381d3147a7fc7d355afede254a781a3094
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoBug 8231: Fix testChangeListenerRegistration failure 80/55180/2
Tom Pantelis [Fri, 14 Apr 2017 13:03:51 +0000 (09:03 -0400)]
Bug 8231: Fix testChangeListenerRegistration failure

As described in Bug 8231, the sharing of the ListenerTree between the
ShardDataTree and the ShardDataTreeNotificationPublisherActor is
problematic. Therefore the ListenerTree (wrapped by the
DefaultShardDataTreeChangeListenerPublisher) is now owned by the
ShardDataTreeNotificationPublisherActor. On registration, a RegisterListener
messages is sent to the ShardDataTreeNotificationPublisherActor to perform
the on-boarding of the new listener, ie it atomically generates and sends
the initial notification and then adds the listener to the ListenerTree.

This change necessitated some refactoring of the DataChangeListenerSupport
class et al wrt to how the ListenerRegistration is handled. Prior the
ListenerRegistration was passed on creation of the registration actor. This
is now done indirectly by sending a SetRegistration message to the
registration actor via a Consumer callback passed in the RegisterListener
message. When the ListenerRegistration is obtained by the
ShardDataChangePublisherActor, it invokes the Consumer callback.

When a registration is initially delayed due to no leader, the
DelayedListenerRegistration is sent to the registration actor. When the
leader is elected later on, the actual ListenerRegistration is sent and
replaces the DelayedListenerRegistration.

The DOMDataTreeChangeListener registration classes were changed/refactored
similarly.

In addition, the 2 specific registration actor classes were replaced by a
generic reusable DataTreeNotificationListenerRegistrationActor that handles
both listener types. Also the 2 CloseData*ListenerRegistration and
CloseData*ListenerRegistrationReply messages were consolidated.

Change-Id: I79ac76b8044609351e5dd8367b691b589ea35075
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
7 years agoBUG-5280: update transaction statistics 49/55149/1
Robert Varga [Tue, 18 Apr 2017 10:50:20 +0000 (12:50 +0200)]
BUG-5280: update transaction statistics

This patch adds statistics-keeping to tell-based protocol code.

Change-Id: I377cd4d9075f96dc69dd74011458fdcf53a65add
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoBUG-5280: handle NotLeaderException 64/55064/4
Robert Varga [Fri, 14 Apr 2017 18:45:20 +0000 (20:45 +0200)]
BUG-5280: handle NotLeaderException

NotLeaderException is indicative of leader movement, in which
case we need to tear down the connection and resolve the new
leader.

Change-Id: I068e97f9a7feb75cc30afb5f5449f0adf00aa217
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoBUG-5280: activate testTransactionRetryWithInitialAskTimeoutExOnCreateTx 63/55063/2
Robert Varga [Fri, 14 Apr 2017 18:16:41 +0000 (20:16 +0200)]
BUG-5280: activate testTransactionRetryWithInitialAskTimeoutExOnCreateTx

This test should work reliably, re-enable it.

Change-Id: I401983ea3579b95a3b37d2144a7085f132eba640
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoBUG-5280: fix invalid local transaction replay 43/54843/4
Robert Varga [Wed, 5 Apr 2017 16:41:14 +0000 (18:41 +0200)]
BUG-5280: fix invalid local transaction replay

When we transition from a connecting to connected local connection,
we may encounter operations which are invalid and these violations
are detected during transaction replay.

If such replay fails, we need to suppress reporting the error until
the user initiates canCommit or directCommit, at which point we need
to report the delayed failure.

For reasons of consistency, we perform this suppression even under
normal connected circumstances.

Change-Id: I2018498afff0e463dbdceaec5c50e8ebf088001b
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoUnit test for RemoteRpcRegistryMXBeanImpl class 24/54824/6
Ivan Hrasko [Wed, 5 Apr 2017 12:54:06 +0000 (14:54 +0200)]
Unit test for RemoteRpcRegistryMXBeanImpl class

Change-Id: Ic00c607f3f66b327336b49f92afe6eb29c144a92
Signed-off-by: Ivan Hrasko <ivan.hrasko@pantheon.tech>
7 years agoFix intermittent failure in ClusterAdminRpcServiceTest.testModuleShardLeaderMovement 82/55082/1
Tom Pantelis [Sat, 15 Apr 2017 02:00:57 +0000 (22:00 -0400)]
Fix intermittent failure in ClusterAdminRpcServiceTest.testModuleShardLeaderMovement

java.lang.AssertionError: Rpc failed with error: RpcError [message=leadership transfer failed, severity=ERROR, errorType=APPLICATION, tag=operation-failed, applicationTag=null, info=null, cause=org.opendaylight.controller.cluster.raft.LeadershipTransferFailedException: Failed to transfer leadership to member-2-shard-cars-config_testModuleShardLeaderMovement. Follower is not ready to become leader]
  at org.opendaylight.controller.cluster.datastore.admin.ClusterAdminRpcServiceTest.verifySuccessfulRpcResult(ClusterAdminRpcServiceTest.java:461)
  at org.opendaylight.controller.cluster.datastore.admin.ClusterAdminRpcServiceTest.doMakeShardLeaderLocal(ClusterAdminRpcServiceTest.java:450)
  at org.opendaylight.controller.cluster.datastore.admin.ClusterAdminRpcServiceTest.testModuleShardLeaderMovement(ClusterAdminRpcServiceTest.java:263)

It failed when trying to make member-2 the leader for a couple reasons. One is that
member-2 hadn't yet received the MemberUp event for member-3 from akka clustering and
thus didn't have its address when it started the election and tried to send
RequestVote.

The second problem is a result of the first - since member-2 couldn't get a vote
from member-3, it needed the vote from member-1, which was in the process of stepping
down as leader. When member-1 received the RequestVote with the higher term, it
switched to Follower. Therefore member-2 didn't receive any votes for that election
term. The request to transfer leadership, which was issued on member-1, then timed out
and failed.

The wait period for the new leader to be elected is 2 sec. This was chosen b/c
originally leadership transfer was only used on shutdown and we don't want to
block shutdown for too long. However, when requesting leadership outside of shutdown,
we should wait at least one election timeout period (plus some cushion to take into
account the variance).

This alleviates the time out but it still failed sometimes if member-1 timed out
in the Follower state and started a new election before member-2 timed out in
Candidate state. member-1 would then win the election and grab leadership back.
To alleiviate this, it would be ideal if member-1 replied to the RequestVote from
member-2 prior to switching to Follower. Normally when it receives a RaftRPC with
a higher term, the Leader is supposed to immediately switch to Follower and not
process and reply to the RaftRPC, as per raft. However if it's in the process of
transferring leadership it makes sense to process the RequestVote and make every
effort to get the requesting node elected.

I also fixed a couple issues in the test code, mainly adding waitForMembersUp.

Change-Id: Ibb1b00f03065680fe1fd338c3d26161ec6336d5a
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
7 years agoFix incorrect last history update 78/55078/1
Robert Varga [Sat, 15 Apr 2017 01:37:16 +0000 (03:37 +0200)]
Fix incorrect last history update

This is a thinko -- the codepath will never trigger, eventhough
it should normally trigger all the time.

Change-Id: I29b24a3823c08c64c8c8a74e7be3b96e07672313
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoChange DistributedShardedDOMDataTree's ctor signature 78/54878/3
Jakub Morvay [Wed, 12 Apr 2017 14:12:29 +0000 (16:12 +0200)]
Change DistributedShardedDOMDataTree's ctor signature

We should inject DistributedShardedDOMDataTree with AbstractDataStore
instead of DistributedDataStore, so we can allow different
implementations of distributed DOM store

Change-Id: I11d1b49e1413dcc233350a3c853b283df176bffa
Signed-off-by: Jakub Morvay <jmorvay@cisco.com>
7 years agoBUG-8159: add payload debugs 84/54884/5
Robert Varga [Wed, 12 Apr 2017 15:59:33 +0000 (17:59 +0200)]
BUG-8159: add payload debugs

This patch adds debugging of metadata snapshot application
and recovery operations.

Change-Id: I9498f53af6ddc8fecf42eb239c7da7da08d3f0c6
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoRemoteRpcProviderFactory and RpcErrorsException unit tests 77/54877/3
matus.kubica [Tue, 4 Apr 2017 12:04:04 +0000 (14:04 +0200)]
RemoteRpcProviderFactory and RpcErrorsException unit tests

Change-Id: Ife8c638d43810baede654cccac22fa8efccae1d0
Signed-off-by: matus.kubica <matus.kubica@pantheon.tech>
Signed-off-by: Ivan Hrasko <ivan.hrasko@pantheon.tech>
7 years agoRemove akka-distributed-data-experimental 76/54976/1
Robert Varga [Thu, 13 Apr 2017 13:28:56 +0000 (15:28 +0200)]
Remove akka-distributed-data-experimental

This module was used during development, but we stopped using it,
remove it from dependencies.

Change-Id: I415347f4e8a264a0daf604375815728f3a77837a
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoBug 8206: Fix IOException from initiateCaptureSnapshot 98/54898/3
Tom Pantelis [Wed, 12 Apr 2017 19:49:23 +0000 (15:49 -0400)]
Bug 8206: Fix IOException from initiateCaptureSnapshot

Modified the install snapshot chunking to be idempotent to avoid attempts
to send the same chunk twice. This fixes the error:

java.io.IOException: The # of bytes read from the imput stream, -1, does not match the expected # 3075

Change-Id: I5336c88125f226d0976f0d7fe17d03c0d181e12d
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
7 years agoBUG-8205: use updated DatastoreContext 48/54848/6
Robert Varga [Wed, 12 Apr 2017 10:15:30 +0000 (12:15 +0200)]
BUG-8205: use updated DatastoreContext

DatastoreContext is updated by the config admin overlay, which means
we cannot refer to the initial one passed in when we are deciding
which data store to instantiate.

This fixes up the protocol propagation and adds and initial info about
which protocol is in use.

Change-Id: I3c2f1a5eec1c7346fff3aca2d85609f47990723a
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoImproved unit tests for AveragingProgressTracker class 23/54823/2
Ivan Hrasko [Fri, 10 Mar 2017 14:45:13 +0000 (15:45 +0100)]
Improved unit tests for AveragingProgressTracker class

Change-Id: I079b45304d82bfc9022321a1648fbdba13409c90
Signed-off-by: Ivan Hrasko <ivan.hrasko@pantheon.tech>
7 years agoBUG-7783: increase precision of execution times 54/54854/2
Robert Varga [Wed, 12 Apr 2017 11:12:34 +0000 (13:12 +0200)]
BUG-7783: increase precision of execution times

Document the time units we are using for measuring execution
and make sure they can hold any long.

Change-Id: I859349e27604c75d426ad7c4eec9d6870b081291
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoBug 8206: Prevent decr follower next index beyong -1 83/54783/6
Tom Pantelis [Tue, 11 Apr 2017 13:39:49 +0000 (09:39 -0400)]
Bug 8206: Prevent decr follower next index beyong -1

If a follower's next index is already -1, we shouldn't decrement it
further, ie -1 is the lowest allowed value. This can result in AbstractLeader
continuously decrementing and logging an info message while in the
process of sending an install snapshot.

member-3-shard-default-config (Leader): follower member-1-shard-default-config last log term 2 conflicts with the leader's 3 - dec next index to -2

Modified decrNextIndex to return a boolean if next index was decremented
which is checked  by AbstractLeader.

Change-Id: I29454d4e71a7f9128b3b47f6a4e3403615c2c8d2
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoFix intermittent failure in testWriteTransactionWithSingleShard 22/54822/3
Tom Pantelis [Wed, 12 Apr 2017 06:20:18 +0000 (02:20 -0400)]
Fix intermittent failure in testWriteTransactionWithSingleShard

DistributedDataStoreRemotingIntegrationTest.testWriteTransactionWithSingleShard
fails intermittently with tell-based protocol in verifyCars after it has
reinstated the follower where it's expecting just car1 but the
data tree contains car 1 and car2. This is b/c the delete transaction for car2
prior to reinstatement wasn't applied on recovery due to the corresponding
ApplyJournalEntries message missing from the persisted journal. The test
expects 2 ApplyJournalEntries messages to be persisted corresponding to the
2 transactions but tell-based persists other payloads as well so there may
be 3 ApplyJournalEntries messages. I changed the code to handle this case.

Also the assertion failure in verifyCars caused it to bypass shutting down
the ActorSystem which resulted in several other failures in tests that try to
use the same port configuration due to the port already in use. So I made
changes to ensure ActorSystems are shutdown properly.

Change-Id: Id6316d71fcd9eb3e768c6b1f676fa0e9be1287a2
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
7 years agoFix intermittent failures in FollowerTest 15/54815/2 15/54815/3
Tom Pantelis [Tue, 11 Apr 2017 21:39:38 +0000 (17:39 -0400)]
Fix intermittent failures in FollowerTest

FollowerTest.testCaptureSnapshotOnLastEntryInAppendEntries:1152 Persisted journal entries size: [] expected:<1> but was:<0>

The test waits on the deletion of journal entries after the snapshot is saved
to occur and then checks the persistent journal for the remaining
ApplyJournalEntries. But occasionally the persisting of the ApplyJournalEntries
message occurs after the deletion so the assertion fails b/c the
ApplyJournalEntries wasn;t persisted yet. This is a little odd b/c the
sequencing in the raft code is that the ApplyJournalEntries write is done
before the delete so it should also be observed the same way in the
InMemoryJournal, even though it doesn't really matter either way.

To alleviate the problem I added a wait for the ApplyJournalEntries
message in the journal in the 3 similar tests.

I also made a couple other minor changes that I observed while running the
tests.

Change-Id: I67cbb8fd79c91cd1cc23c363b78e7f5e9b9f2bbe
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
7 years agoBug 5280: Enable tests for ClientBackedDatastore 09/54809/3
Andrej Mak [Tue, 4 Apr 2017 07:49:54 +0000 (09:49 +0200)]
Bug 5280: Enable tests for ClientBackedDatastore

Change-Id: I33d6312c9b18493e519b8607307c21c1b3a9bc75
Signed-off-by: Andrej Mak <andrej.mak@pantheon.tech>
7 years agoMake AbstractIdentifiablePayloadTest public 06/54806/1
Tom Pantelis [Tue, 11 Apr 2017 17:24:01 +0000 (13:24 -0400)]
Make AbstractIdentifiablePayloadTest public

For some reason, tests derived from AbstractIdentifiablePayloadTest
fail b/c AbstractIdentifiablePayloadTest isn't public when running from
eclipse - runs fine from command line.

Change-Id: Ie6ed1d6e0e130a1ffc5ad04db93e037ea6a79549
Signed-off-by: Tom Pantelis <tompantelis@gmail.com>
7 years agoBUG-5280: Correct reconnect retry logic 76/54776/1
Robert Varga [Thu, 30 Mar 2017 13:14:04 +0000 (15:14 +0200)]
BUG-5280: Correct reconnect retry logic

Our reconnect logic failed to account for various timers
during resolution. This patch makes the BackendInfoResolver
explicit about the type of failures it can report and fixes
AbstractShardBackendResolver to conform to them.

Change-Id: I610ddb6e062e223557d46e2950a552de6e7d3843
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoUpdate .gitreview to stable/carbon 39/54639/1
Anil Belur [Tue, 11 Apr 2017 01:26:29 +0000 (11:26 +1000)]
Update .gitreview to stable/carbon

Change-Id: I47a0426c3aad7f26c21bc954f098251e0e75deb0
Signed-off-by: Anil Belur <abelur@linuxfoundation.org>
7 years agoHandle odl-mdsal-common with Karaf 4.0.9 07/54607/1
Stephen Kitt [Mon, 10 Apr 2017 15:00:17 +0000 (17:00 +0200)]
Handle odl-mdsal-common with Karaf 4.0.9

Karaf 4.0.9 simplifies feature dependencies (so that dependencies
specified in feature.xml can be completed from the POM), but that
causes issues with odl-mdsal-common since features in Karaf are
identified by their name only:

* if both the controller odl-mdsal-common and the mdsal
  odl-mdsal-common are encountered in the dependency tree, whichever
  one came first (as dependencies are resolved) is the one that ends
  up being kept;
* in some circumstances, the mdsal repository replaces controller's
  even when the controller dependency is retained (this is a Karaf bug
  which I'll submit a patch for, but we can work around it).

Change-Id: I5400a829560ae96cb2f264e103020cccd1d225c3
Signed-off-by: Stephen Kitt <skitt@redhat.com>
7 years agoBUG-5280: add the concept of a recorded failure 71/54371/4
Robert Varga [Wed, 5 Apr 2017 16:36:48 +0000 (18:36 +0200)]
BUG-5280: add the concept of a recorded failure

This patch reworks LocalReadWriteProxyTransaction to be defensive
of its internal modification and introduces the concept of a delayed
recorded error (currently unused).

The defensiveness checks allow us to get rid of FailedDataTreeModification,
as we do not give out our modification at all in the codepaths which
would leak this implementation.

Change-Id: I5f91218ac308f7450a3b59252d44f953be54626c
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoBug 7805: Add make-leader-local rpc for module based shard. 00/54100/13
Tomas Cere [Thu, 30 Mar 2017 12:44:32 +0000 (14:44 +0200)]
Bug 7805: Add make-leader-local rpc for module based shard.

csit testing scenarios require movement of the shard leader for module
based shards aswell so add this into ClusterAdminRpcService.

Change-Id: Ib8a310cdba728c0a42d8850703740bf4698adbe0
Signed-off-by: Tomas Cere <tcere@cisco.com>
7 years agoBug 7806 - Implement agent RPCs for shard replica manipulation testing 38/54038/9
Tomas Cere [Wed, 29 Mar 2017 10:14:33 +0000 (12:14 +0200)]
Bug 7806 - Implement agent RPCs for shard replica manipulation testing

These can be implemented as a part of ClusterAdminRpcService instead
of creating new rpcs that would be part of the lowlevel suite.

Change-Id: I891f9d3703a9357e829159691cbf18f95523d529
Signed-off-by: Tomas Cere <tcere@cisco.com>
7 years agoBUG-5280: log a message when tell-based protocol is active 20/54320/2
Robert Varga [Tue, 4 Apr 2017 16:27:34 +0000 (18:27 +0200)]
BUG-5280: log a message when tell-based protocol is active

Discerning the two access modes is critical to understanding
when failures occur. Add an explicit note when the tell-based
protocol is enabled on a data store.

Change-Id: I3e2b1d2f84a73ce1a3759d419176c47a6dd0ad12
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
7 years agoPass no op callback instead of null during replay 22/54022/5
Andrej Mak [Wed, 29 Mar 2017 09:32:25 +0000 (11:32 +0200)]
Pass no op callback instead of null during replay

Change-Id: Ife964481dc225bbc1d5b312035384f8bd597d740
Signed-off-by: Andrej Mak <andrej.mak@pantheon.tech>
7 years agoAdd AbstractIdentifiablePayload unit tests 52/54252/3
Andrej Mak [Mon, 3 Apr 2017 09:01:04 +0000 (11:01 +0200)]
Add AbstractIdentifiablePayload unit tests

Change-Id: I884f3c35d1767ed02accabc3b9a775ef9c667716
Signed-off-by: Andrej Mak <andrej.mak@pantheon.tech>
7 years agoUnit test for ClientBackedTransaction derived classes 88/53988/21
Ivan Hrasko [Tue, 28 Mar 2017 15:21:21 +0000 (17:21 +0200)]
Unit test for ClientBackedTransaction derived classes

Change-Id: I2967a0e224fc783ffac73a994def666e86a423a6
Signed-off-by: Ivan Hrasko <ivan.hrasko@pantheon.tech>
7 years agoUnit tests for ClientBackedTransactionChain class 18/54018/10
Ivan Hrasko [Wed, 29 Mar 2017 08:53:23 +0000 (10:53 +0200)]
Unit tests for ClientBackedTransactionChain class

Change-Id: I97953cfdc32619c31295cfed2584b7466d48aa5d
Signed-off-by: Ivan Hrasko <ivan.hrasko@pantheon.tech>
7 years agoAbstractClientHistory derived classes tests 43/54043/17
matus.kubica [Wed, 29 Mar 2017 14:13:57 +0000 (16:13 +0200)]
AbstractClientHistory derived classes tests

Change-Id: I1261eb764c730bbce6eb833644db99e4bbf0605c
Signed-off-by: matus.kubica <matus.kubica@pantheon.tech>
Signed-off-by: Ivan Hrasko <ivan.hrasko@pantheon.tech>
7 years agoAbstractDOMDataBroker fix annotations 02/54202/2
Jie Han [Sat, 1 Apr 2017 02:54:49 +0000 (10:54 +0800)]
AbstractDOMDataBroker fix annotations

Change-Id: I7938965d805ba3e4228e9fc5a36c75c52f6ec881
Signed-off-by: Jie Han <han.jie@zte.com.cn>
7 years agoBug 7805 - Implement agent RPCs for shard leader movement testing 01/53901/12
Tomas Cere [Mon, 27 Mar 2017 14:30:27 +0000 (16:30 +0200)]
Bug 7805 - Implement agent RPCs for shard leader movement testing

Change-Id: Ic19d1867f3c54ec22d600e9b80c6490d5a4b99bb
Signed-off-by: Tomas Cere <tcere@cisco.com>
7 years agoBUG-5280: Close client history after all histories are closed 37/54037/9
Ivan Hrasko [Wed, 29 Mar 2017 13:51:33 +0000 (15:51 +0200)]
BUG-5280: Close client history after all histories are closed

Make sure record history state as closed once we are done with
it.

Change-Id: Icbdf947ad166b082e06df896741e618e801ecf2e
Signed-off-by: Ivan Hrasko <ivan.hrasko@pantheon.tech>
7 years agoBUG-5280: switch tests to ClientBackedDataStore 18/48718/102
Robert Varga [Mon, 12 Dec 2016 18:34:38 +0000 (19:34 +0100)]
BUG-5280: switch tests to ClientBackedDataStore

Enable integration tests to run
on the new frontend code with parametrized JUNIT.

Not working tests for new code are ignored.
For old code all tests run and pass.

Change-Id: Ib5656ecd2333a56d5c466e633fbdd477accc4095
Signed-off-by: Robert Varga <rovarga@cisco.com>
Signed-off-by: Ivan Hrasko <ivan.hrasko@pantheon.tech>
7 years agoBUG 7801: prevent OptimisticLockFailedExceptions in write-transactions. 84/53784/7
Tomas Cere [Fri, 24 Mar 2017 10:29:16 +0000 (11:29 +0100)]
BUG 7801: prevent OptimisticLockFailedExceptions in write-transactions.

When multiple instances of this rpc are running concurrently in paralel
we would run into an optimistic lock since every instance tries to write
the topmost parent list first.
When these happen handle these failures as expected and resume with the
next stage of the rpc.

Change-Id: I43efaea3315b04272113eb86733e68609e434984
Signed-off-by: Tomas Cere <tcere@cisco.com>