some drop tests statistics were not cleaned on clearstats command

Change-Id: I58f7831f7d6f90dcc5a04d948f80bae152d7e2a1
Signed-off-by: Martin Bobak <mbobak@cisco.com>

DeviceDataException erradicated

OFJ's DeviceRequestFailedException is used instead of ofp own exception
that just wrapped exception from OFJ.

Change-Id: I53f118782ed677a82958a6ad42dedf37b6aee109
Signed-off-by: Martin Bobak <mbobak@cisco.com>

registration of outbound queue handler simplified

Change-Id: Idfd809354c9f9cd6455662b6ea4de7b5ea449439
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Limit drop-test queue limit

 - limited to 10k
 - blocking behavior
 - internal statistics cleaned

Change-Id: I02590b194f5e9dafc9b8a5503e369b33ffd86ac2
Signed-off-by: Michal Rehak <mirehak@cisco.com>
Signed-off-by: Martin Bobak <mbobak@cisco.com>

added missing information in droptest stats output

Change-Id: I68340ea8144bb9c0f5fea28dbc3f258e64e39e28
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Merge "openflow plugin internal statistics available trough karaf commands"

Merge "more logging in drop test added"

Merge "Returning of future from request context - SalMeterService."

Merge "Returning of future from request context - SalGroupService."

openflow plugin internal statistics available trough karaf commands

Change-Id: I8d5b6ce4dd897a1f1af9d223a977641e0353eecd
Signed-off-by: Martin Bobak <mbobak@cisco.com>

more logging in drop test added

Change-Id: Idce5769c4440e1192ebbd355d0ec154aa36e9a5a
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Returning of future from request context - SalGroupService.

Change-Id: I510ee97435596ee6940afad630362727e6311a47
Signed-off-by: Jozef Gloncak <jgloncak@cisco.com>

Returning of future from request context - SalMeterService.

Change-Id: I7815d6eed2979833380e1d7fd41b1819d0a2cc73
Signed-off-by: Jozef Gloncak <jgloncak@cisco.com>

Returning of future from request context - SalPortService.

Change-Id: I8e0060d74ef49984a2798efd9eae15a639c99e96
Signed-off-by: Jozef Gloncak <jgloncak@cisco.com>

Fixed result processing in several services.

- statistics gathering service doesn't set result to req. context
  on future's success that's multi message collector's job
- SalTableService doesn't use settable future instead of request
  context's future
- added missing statistics counter to salFlowService

Change-Id: If2107099602abb935f91c5f65bd84e079316798a
Signed-off-by: Martin Bobak <mbobak@cisco.com>

added statistics counters for drop test future's result

Change-Id: I471d4d6c6cae53995c84af2ebafb0726a39b66d2
Signed-off-by: Martin Bobak <mbobak@cisco.com>

PacketProcessingService uses future from requestContext

Change-Id: I0a04f8f19042073ab60d3cb4bce4f16f0f0d673a
Signed-off-by: Martin Bobak <mbobak@cisco.com>

SGService uses future from requestContext

Change-Id: I6afa88210e0bb40c96ac6e2bf764eba1ce83cdf4
Signed-off-by: Martin Bobak <mbobak@cisco.com>

device context holds OQHandler registration

Handler registration is closed when device context is closed.

Change-Id: Ib4a58c943789c5505542858767e5b6311bea3f3b
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Merge "Do not register OpenflowProtocolListenerFullImpl in constructor"

Merge "Returning of future from request context - FlowCapableTransactionService."

Merge "Returning of future from request context - SalEchoService."

Merge "Returning of future from request context - NodeConfigService."

Merge "Returning of future from request context - PacketProcessingService."

Merge "Rework of handleServiceCall method"

Do not register OpenflowProtocolListenerFullImpl in constructor

This is hard to read, as the caller of the constructor does not see the
leaked reference. Move registration out.

Change-Id: I01cb6419ec9f42af5cb4959411fc52281a0d9fdb
Signed-off-by: Robert Varga <rovarga@cisco.com>

Remove a potential NPE

If the packet is null, it will throw NPE, fix that.

Change-Id: I513161e0ebe996c1dfa7ce405c303b1d0f7e3bf4
Signed-off-by: Robert Varga <rovarga@cisco.com>

Fix generic warnings

RequesstContext is generic, hence eclipse gives warnings.

Change-Id: I5e831565e0861d8444025072a82fd1f4828c03e9
Signed-off-by: Robert Varga <rovarga@cisco.com>

Returning of future from request context - FlowCapableTransactionService.

Change-Id: Ie0d56eea35035378e96cc61c620de46f2e14fc4f
Signed-off-by: Jozef Gloncak <jgloncak@cisco.com>

Returning of future from request context - NodeConfigService.

Change-Id: I8f9fd184428378e3039ac6d3ab5345f2456ccc9f
Signed-off-by: Jozef Gloncak <jgloncak@cisco.com>

Returning of future from request context - SalEchoService.

Change-Id: I1096e5e605deddb0e8c1caf2f8407239a9c5cc34
Signed-off-by: Jozef Gloncak <jgloncak@cisco.com>

Returning of future from request context - PacketProcessingService.

Change-Id: I3404fb48b937c948f6ae1bd152f50a9e69cba721
Signed-off-by: Jozef Gloncak <jgloncak@cisco.com>

Remove unused variable

This variable is not used anywhere, remove it.

Change-Id: Iec929c0b8f0029f79974f07f7720d95ff57d8ea8
Signed-off-by: Robert Varga <rovarga@cisco.com>

Make methods static

The two methods can be made static, so do that to increase clarity.

Change-Id: I2063a28db722d11ab83c7bf4361c61d7a43360f5
Signed-off-by: Robert Varga <rovarga@cisco.com>

Move invariants out of the loop

Instead of performing lookup inside the loop, cache the results.

Change-Id: I88f735ff4589a853a12250692f48090379311f85
Signed-off-by: Robert Varga <rovarga@cisco.com>

Rework of handleServiceCall method

Second type parameter of method was removed.

Change-Id: I447fae85a8e2142a0c755d67f838f0c7326a1bd9
Signed-off-by: Jozef Gloncak <jgloncak@cisco.com>
Signed-off-by: Martin Bobak <mbobak@cisco.com>

future from request context should be always returned

Change-Id: Ic6fd71a1372c431cbffa13242e7fba64f527fdb0
Signed-off-by: Martin Bobak <mbobak@cisco.com>

added spying on overall OFJ backpressure on/off operations

Change-Id: Icadae7c2a2842dbaade28259e8e823c838561b91
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Revert "binding thread pools in drop test to target node"

This change is suspect of significant decrease of throughput performance by He-codebase ofPlugin.

This reverts commit 673a667dc331d5877f321e45396ad143e0c8055e.

Change-Id: I34c23c280c0c1e6d2e31816af4da0ec2e9ae2725
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Fix unnecessary timeout submit

Note: TransactionChainManager submits transactionChain after timeout
(submitScheduledTransaction method) without check nr. of actual
transaction. So we'd like to submit tx if we are some tx inside
only. Otherwise we see bad exception in logs.

Change-Id: Ieff610a8bb337d799aefe131a8c3b787fa4081f9
Signed-off-by: Vaclav Demcak <vdemcak@cisco.com>

throttling fix

 - fixed stuck condition

Change-Id: I6824fe66e1125f8f842a4e8ebd61ce31ac6c41e9
Signed-off-by: Martin Bobak <mbobak@cisco.com>

defensively create empty list for datastore nodes for datastore operations to complete

Change-Id: I31e4bef62d90f2e58475e8fabf6481f14a0ce406
Signed-off-by: Martin Bobak <mbobak@cisco.com>

multimessage collector provides way how to explicitly cancel processing on selected req. ctx.

Change-Id: Ib6781ebee15c01df7ffdf4dfe185b321633c1fcc
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Throttling via packetIn filter proposal

 - throttling is triggered by rejection of packetIn by md-sal
 - outstanding notifications are counted
 - when throttling is active then it is switched off upon
   outstanding notifications amound falls to 0

Change-Id: Ic5e630a8924a5995883e4b59a6b84d8591664805
Signed-off-by: Martin Bobak <mbobak@cisco.com>

initial condition - api change in unit tests

Change-Id: I94306db05a69394c10600a912699e092c01828d2
Signed-off-by: Martin Bobak <mbobak@cisco.com>

unregister request ctx from rpc stack when transmitPacket operation fails

Change-Id: If38c097ba19a2c2062f7ebac110f29a2a8e1a538
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Merge "flow descriptors should be always stored when addFlow is called"

Merge "call get on transaction submit operation when writing Nodes node into operational DS on OFP starts"

flow descriptors should be always stored when addFlow is called

Change-Id: Idef70e2e2f28e231dacbe2ccda8a8a9d484570e1
Signed-off-by: Martin Bobak <mbobak@cisco.com>

call get on transaction submit operation when writing Nodes node into operational DS on OFP starts

Change-Id: I1c7caf563ca35b890f02c1c9aa3cc4ac74db2fd2
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Do not use calendar time

Instead of using System.currentTimeMillis(), as it travels with wall
time. Use nanoTime(), which is faster and monotonic.

Change-Id: I4b1089a23dbeba69acdceb4e0e55a0ec0457612f
Signed-off-by: Robert Varga <rovarga@cisco.com>

Do not keep per-switch state in DeviceManagerImpl

All per-device state is encapsulated in a DeviceContext, keeping a
provider as a field breaks thread-safety, as it is changed on each
request.

Change-Id: I60e818cee662f2f952e5435ff2e9ba9d1c98d053
Signed-off-by: Robert Varga <rovarga@cisco.com>

Optimize HashUtil

Do not use String.replace(), as it compiles a regex on each invocation,
but rather compile the pattern upfront and use that for multi-char
replacement. For single-character replacement, use Guava's CharMatcher,
which is much more efficient.

Use String.indexOf(char) instead of String.contains(String).
Use String.isEmpty() instead of comparison to an empty string.

Change-Id: I9d51151295dd66167c5750ed4e961ae8a621c8a2
Signed-off-by: Robert Varga <rovarga@cisco.com>

Introduce a shared successful RpcResult

Also fix message counters in PacketProcessing.

Change-Id: I7a270efa50a88dc430691002031d4079e3e369bf
Signed-off-by: Robert Varga <rovarga@cisco.com>

Mark DeviceDataException as deprecated

Change-Id: I9f8dca65426b60ca9d0919cb5f1fe833bf65a5ad
Signed-off-by: Robert Varga <rovarga@cisco.com>

Use StringBuilder instead of StringBuffer

StringBuffer is needlessly synchronized, so do not use it.

Change-Id: Iba425633aa626495e1bfadf918273309a824f3b1
Signed-off-by: Robert Varga <rovarga@cisco.com>

Node  created when device connects is created with empty Node list

Change-Id: I1f44dffb9406e7d31783cac75baa4d5f6cec007f
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Merge "RequestContext's are not held by DeviceContext anymore"

Merge "fix callback handling"

fix callback handling

Change-Id: I911d8614d8ad477b5f856e67caf94e05bfde064d
Signed-off-by: Martin Bobak <mbobak@cisco.com>

RequestContext's are not held by DeviceContext anymore

All necessary work is done in callbacks registered when commiting entry in  outbound queue.
That makes registration of req. context inside device context worthless. Collection of
multiparts is done in message collector on request context registered when entry is commited.
Request contexts not held by DeviceManager and barrier reuqests issued by OFJ makes
OutsandingMessageExtractor useless as well.

Change-Id: Ib259020ed4e3cca56463696827bb24bf7bc97069
Signed-off-by: Martin Bobak <mbobak@cisco.com>

shift statistics services to new OFJ codepath

Change-Id: Id3fa22878605029989baabbc5722394fa3e92728
Signed-off-by: Martin Bobak <mbobak@cisco.com>

removed unwanted git file created by merge

Change-Id: If225d432498f9767428757c8cb9be33e412121ab
Signed-off-by: Martin Bobak <mbobak@cisco.com>

ThrottledNotificationsOfferer should not be generic

The only use of the type argument is to bind method arguments -- which
we can spell out. This way we can reuse an offerer across multiple types
without incurring warnings.

Change-Id: Ie003f6c9ebe00f1abae4f8171086a7240843a14c
Signed-off-by: Robert Varga <rovarga@cisco.com>

Fix RpcResultBuilder/RpcContext raw references

This adds proper type arguments to eliminate eclipse warnings.

Change-Id: I63373fadd17c1f99f8e8e465190adc65a2597a0a
Signed-off-by: Robert Varga <rovarga@cisco.com>

Do not use InstanceIdentifier.builder()

These instance identifiers are expected to be short-lived, so use the
appropriate utility child() method instead.

Change-Id: I23a6e6d33fec7b904794ac38ff153b4d46063858
Signed-off-by: Robert Varga <rovarga@cisco.com>

Fix various warnings

Fixes raw RequestContext references, as well as unclosed test resources.

Change-Id: Ia2aa821e0d520fa26e2de66e30f09de7cf300ba5
Signed-off-by: Robert Varga <rovarga@cisco.com>

Mark methods as static

Some methods can be marked as static -- do that to improve clarity.

Change-Id: Ib73b989ae0bc81617ffa8a07d715922d9198da2b
Signed-off-by: Robert Varga <rovarga@cisco.com>

Remove unused imports

Eclipse-reported unused imports -- remove them.

Change-Id: I0ecfda5cd3730ee685740f300d7268898f5d748c
Signed-off-by: Robert Varga <rovarga@cisco.com>

Do not override versions

Fix a few maven warnings about overriding managed versions.

Change-Id: Icd980761fb33b35213ffc8b43fd85e25d725ef3f
Signed-off-by: Robert Varga <rovarga@cisco.com>

MessageSpy should not be generic

This reduces the amount of warnings and confusion around when the actual
argument should be.

Change-Id: I614f2ba44f8432e3be3f3fd79c37f5d95f87fd76
Signed-off-by: Robert Varga <rovarga@cisco.com>

Use pre-constructed NODES_IDENTIFIER

We have a constant for the first-order identifier. Make sure it is a
proper constant and use it as a base for all child identifiers.

Change-Id: I53d6ddfc782424aae0e4ac0997afb735c9ca7688
Signed-off-by: Robert Varga <rovarga@cisco.com>

check for manadatory attribute

Change-Id: Ibe1ba13e63b65a8cb4a2799a2c8b902d7feaf8bb
Signed-off-by: Martin Bobak <mbobak@cisco.com>

corrected way of handlig XID

Change-Id: Iad1386ca6a1b47a1849ceabeabd234ba012d1a4f
Signed-off-by: Martin Bobak <mbobak@cisco.com>

corrected message spy counters per service

Change-Id: I62bf785d96f8dd1e4226f4d46278de243e3214e5
Signed-off-by: Martin Bobak <mbobak@cisco.com>

fixed npes in service processing and message spying

Change-Id: I607200b0ac2a9324c44c2460a6c633000f0029f1
Signed-off-by: Martin Bobak <mbobak@cisco.com>

removed worthless callbacks

Change-Id: I8e408f2dbd0d7c7a52eb8731cd3ae79049d03ea0
Signed-off-by: Martin Bobak <mbobak@cisco.com>

FlowCapableTransactionServiceImpl  and SalEchoServiceImpl.java shifted to new OFJ codepath

Change-Id: I3787c1c8ed16a2ebabdc26425ea8f4e823c72cc8
Signed-off-by: Martin Bobak <mbobak@cisco.com>

NodeConfigService shifted to new OFJ codepath

Change-Id: I5b5e6cb1906e1fc9a9a99a446606dc4b106c0399
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Make sure RequestContext has a constant XID

All callers end up talking to the underlying device to acquire an XID.
Since all RequestContextStack implementations already have (potential)
access to a DeviceContext, there is no need to do this.

Thus make sure RequestContext has an XID allocated by the underlying
device, which unifies code and makes callers simpler. If we fail to
allocate one, the caller will see that XID as null.

We also take this opportunity to make Xid take a Long instead of a
primitive type, as that's what we need for DataObjects and that's what
we get from the Device anyway -- thus preventing autoboxing operations.

Change-Id: I7f553a9a1283d1c82ba7e287d0f21cf4147779e2
Signed-off-by: Robert Varga <rovarga@cisco.com>

DataCrate removed and its usage replaced by RequestContext

DataCrate created unnecessary code and in light of newest OFJ changes became obsolete.
All we need to have when handling rpc call is request context.

Change-Id: Iff565c48af6634c9000a2b09057997b81cbb062b
Signed-off-by: Martin Bobak <mbobak@cisco.com>

removed redundat xid reservation

Change-Id: Ic6bfadf9b163b53a6930219df10da7d3865dd213
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Cleanup RequestContextStack

Instead of exposing the limits as a tweakable API, we should specify the
limit at instantiation as an implementation-specific thing.

We document that RequestContextStack.createRequestContext() can return
null and let callers deal with that.

This change allows us to implement RpcContext throttling using a simple
counting Semaphore, which is more scalable and efficient than an
explicit queue. To do that, we turn RequestContextImpl into AbstractRequestContext,
which is subclassed as needed by the various RequestContextStack
implementations.

Change-Id: Ibbeafad6cd7f740284264cca22412e11696cede8
Signed-off-by: Robert Varga <rovarga@cisco.com>

Remove DeviceContextImpl.XidGenerator

The generator is not used anymore, remove it.

Change-Id: Ia4a50a0b2b02b34f3854a4f09c4b307ccebb24c2
Signed-off-by: Robert Varga <rovarga@cisco.com>

Fix various small warnings

All of these are reported by eclipse: needlessly raw types, missing
serialversionUUID, methods which can be made static.

Change-Id: I5b71c35ee37167ead77f7baa2e5d6e5c23f99834
Signed-off-by: Robert Varga <rovarga@cisco.com>

Merge "Speed FlowUtil up a bit"

Merge "Explicitly deal with rejected offer operation"

Speed FlowUtil up a bit

Instead of taking a Short, it can take the primitive short, increasing
performance just a tiny bit.

Change-Id: I7a2c4b0f20108c943110b83ba1237a9ad21d57e0
Signed-off-by: Robert Varga <rovarga@cisco.com>

Explicitly deal with rejected offer operation

Performance traces show a whopping 30% of the method's time being spent
in ImmediateFailedFuture.get(). Make sure this does not occur by
checking for the well-known constant first, shaving 4% of global CPU
usage.

Change-Id: Id7fdfbb32a071634e7e5467441beb242181c68bb
Signed-off-by: Robert Varga <rovarga@cisco.com>

removed misplaced internal statistics counters

Change-Id: I8ca8fde20d01982c7564a47dc308d2eaf6193021
Signed-off-by: Martin Bobak <mbobak@cisco.com>

fixe NPE when logging in add flow's future callback

Change-Id: Iaf53511e8b1f09676ddb7284b3c5b239d778c1d4
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Merge "Do not use InstanceIdentifier.builder()"

Make sure we wait for the reserve XID

 Base openflowjava OutboundQueue has non-blocking semantics, so can fail
easily. Convert our handler to be a blocking proxy working on top of it.

- no throttling on packetIn
- fix npe when logging response from queue
- unregister request context in queue callback when request fails
  - positive callbacks are gathered by multipart message collector
    and unregistered when all multiparts are collected by device
    context

- OutboundQueueProvider encapsulation in deviceContext

Change-Id: I52cc145d2b59bf51594c7da08796570b35b8250b
Signed-off-by: Martin Bobak <mbobak@cisco.com>

common service doesn't wrap future into BaseCallback
removed unwanted callbacks in SalFlowService

Change-Id: I83a55752544826aaeeb5d1246e5a622fe1021532
Signed-off-by: Martin Bobak <mbobak@cisco.com>

statistics groups renamed

- fixed statistics counting in device context's notification forwarding
- added new statistics group for spying on unregistering items from rpc stack
- changed way how barrier nanos value is expressed
- spying on rpc stack operations

Change-Id: I82df21076543458f9e34f677007bfe31aa697634
Signed-off-by: Martin Bobak <mbobak@cisco.com>

added new statistics groups for measuring refused notifications and XID requests

- spying for new statistics
- initial OFJ queue length set to 25600
- when first XID reservation on outbound queue fails, we attempt for second one
- PacketProcessingService uses outbound queue for sending requests to OFJ
- StatisticsGatheringService doesn't use uoutbound queue in synchronized block

Change-Id: I9f40be8c65c2ff4d75d81b9b814fa28cc1cae8f2
Signed-off-by: Martin Bobak <mbobak@cisco.com>

DeviceContext provides reserved values for XID

Change-Id: I1db70922f8a4bbbb4d0b44c66888975dc8b949b4
Signed-off-by: Martin Bobak <mbobak@cisco.com>

MultipartMessageCollector is used one per device context

- Multipart messages are collected inside OutboundQueue callback.
- StatisticsGatheringService collects its multipart responses in queue callback

Change-Id: I7e4cf52d2e396f08b94d4d7f08af8b591061dfc3
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Introduced OutboundQueue from OFJ for outbound messages processing

BarrierProcessor and related classes are removed
Device static info gathered trough OutboundQueue
Outbound queue used in CommonService and SalFlowService

Change-Id: Ieeff15184aac44cdfd0a902b2f3ae89bfa149461
Signed-off-by: Martin Bobak <mbobak@cisco.com>

Do not use InstanceIdentifier.builder()

Adding a single node to a InstanceIdentifier is much faster than
instantiating a builder.

Change-Id: I9a41aa2c46b9a7977bdc579577852b23e6804a0e
Signed-off-by: Robert Varga <rovarga@cisco.com>
Signed-off-by: Michal Rehak <mirehak@cisco.com>