Route {Journal,Snapshot}Protocol messages to DataPersistenceProvider

The way we wire our messages ends up being not exactly nice, as the
baseline RaftActor is not cognizant of all persistence operations. This
means it cannot correctly ignore responses to deletion of journal and
snapshot entries -- and hence we get dead letters logged.

Route JournalProtocol and SnapshotProtocol messages to
DataPersistenceProvider, which can then handle them and correctly
tell us whether to log them or not.

JIRA: CONTROLLER-2042
Change-Id: I47b8c3ae67d0a0ea0aad9f0a64e1bb8dc11400fc
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Improve AbstractClientConnection nullness guarantees

We have a few missing @NonNull annotations, make sure to add them.
Also remove a superfluous requireNonNull() in copy constructor.

Change-Id: I1bbd46d31856be3ce25483f05eae8e9f2a67370f
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Do not use NPE for consistency checks

We are using Preconditions.checkNotNull() when guarding against illegal
state. Improve this by using verifyNotNull(), which results in
VerifyException, making the intent clearer.

Also migrate a verify(foo != null) to a proper verifyNotNull(), using
the resoluting value.

Change-Id: I2689852d396315714f126743b70469ceef75ce45
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Clean up unhandled request exceptions

We have a few places which throw the same exception, make sure we
share the formatting and expose the proper control flow to Java.

Change-Id: I514faf987b780de813fe6a1374c5e634e22326a6
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Enable tell-based protocol by default in DTO

We have switched to tell-based protocol by default in datastore.cfg
configuration. Reflect that change in DTO DatastoreContext as well.

JIRA: CONTROLLER-2040
Change-Id: I2ed6c287fe7a39415afe49cb45fd485b64208e03
Signed-off-by: Ivan Hrasko <ivan.hrasko@pantheon.tech>

Improve LocalProxyTransaction.doExists()

We can use immediateBooleanFluentFuture() and skip allocating two
objects.

Change-Id: I11747ddef4fa8e3dbf950b5e1879d2e1ccc395ba
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Remove FindBugs workaround

SpotBugs no longer throws a fit on callback's nullness, remove the
workaround. Also do not execute read/exists if there is no callback
-- although there always should be.

Change-Id: I3609d3be076de488f7fa64c72ad3b875e77a633e
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Expose CursorAwareDataTreeSnapshot from local RW proxy

We can improve type safety a bit by exposing the fact the backing
snapshot supports cursors.

Change-Id: If65549d6a28851c0abd1c8b826e48503ef12860c
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump versions to 5.0.3-SNAPSHOT

This starts the next development iteration.

Change-Id: I3cf534a6c69bd38fa3a178d9becd61275fefae0f
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump to yangtools-8.0.3/mdsal-9.0.2

Pick up fixes from upstream.

Change-Id: I38ff023ba4828e36f5ebb89fed7d6f54db1fdc24
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Implement scatter/gather on module shards

ClientBackedDatastore does perform module shard root scatter/gather,
hence reads of root end up talking to only the default shard.

Refactor the scatter/gather logic from DistributedDatastore into a
common component and use it for both modes of operation.

JIRA: CONTROLLER-2038
Change-Id: Ib04efab46e36e512eb89606d026fd2082de1b693
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

NormalizedNodeAggregator should also report empty

DataTree root is always present as of now, which means that even if we
pretend there is no result, the result of aggregation is non-empty.

Fix NormalizedNodeAggregator to check whether there was any operation
applied before running actual transaction.

JIRA: CONTROLLER-2038
Change-Id: Ie7e4625ad2377483d0623d96234bcc77fbba033d
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Use ImmutableBiMap builder fluently

We can simply allocate the builder without intermediate local, saving
us an import. Also split the locked update part into a separate method
to aid inlining.

Change-Id: I34b3dffd49c5bd637792c743e9b71e31082201f4
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Cap shard-snapshot-chunk-size to 480KiB

We have a few places where we are using a default of 2MiB for chunk
size. Bring them down to 480KiB to match datastore.cfg.

JIRA: CONTROLLER-2037
Change-Id: I4dfc4cf3f41acd8c4b782e6450d17f5d5e8bf281
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Add supervisor to EOS singleton actor

The actor may fail, for example due to AskTimeoutException. This still
needs to be investigated, but mitigate the failure by adding an
OwnerSupervisor to restart it on failure.

JIRA: CONTROLLER-2035
Change-Id: I2dafaa35804357415d9aaa5400372cfd8e54b85b
Signed-off-by: Shibu Vijayakumar <shibu.vijayakumar@infosys.com>

Bump versions to 5.0.2-SNAPSHOT

This starts the next development iteration.

Change-Id: Icae7ce95024658ee30b8672300fa11c40f50f47b
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Update upstreams

Adopt yangtools-8.0.2 and mdsal-9.0.1.

Change-Id: I7d010a073386cd5e9144f92c83d4351058f4b5d2
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump akka to 2.6.19

https://akka.io/blog/news/2022/03/21/akka-2.6.19-released

Change-Id: Ib53536c127fa0b4b2e6e0d80f5c78828bec1e750
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump versions to 5.0.1-SNAPSHOT

This starts the next development iteration.

Change-Id: I082d9652feda9aa875f5a5d23d577a0d04c62553
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump mdsal to 9.0.0

Use released version instead of snapshots.

Change-Id: I3563eae608f91b7b8cc94181fd7c1aee2d2c68a7
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump yangtools to 8.0.1

yangtools-8.0.0 has a small bug, make sure we have the fix for it.

Change-Id: If437c6e12855bdde39d4d118f02c8b2362d1ae33
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump yangtools to 8.0.0

There is a proper release out there, use it instead of snapshots.

Change-Id: I6897c2b69f064ce6af95a98492d4bc7979e2199c
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Adjust Blueprint to ensure the container is a ChildOf

InstanceIdentifier requires data tree steps, make sure we enforce
that invariant.

JIRA: MDSAL-370
Change-Id: Icb3c08eb7052eed28ef6638facb884857baec3e9
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Introduce remaining cli commands for clustering-test-app

There are some RPCs that were not reworked to Karaf CLI commands
during first round of introducing these commands, so we need to
introduce them as well and have complete set of commands.

JIRA: CONTROLLER-1995
Change-Id: Ifbb67e26fb4ede4a589dd07999b52e3639e5c794
Signed-off-by: Dominik Vrbovsky <dominik.vrbovsky@pantheon.tech>
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Remove an unneeded @SuppressFBWarnings

The warning is no longer emitted here with our current SpotBugs, hence
we can remove the suppression.

Change-Id: Id1bb3c87a05f14cba76dc8a1af73740a461a1fc1
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Use CommitInfo in AsyncNotifyingSettableFuture

We are performing a future transformation which boils down to converting
a null (Void) into an CommitInfo.empty().

Rather than doing that, let's make sure AsyncNotifyingSettableFuture
results directly in CommitInfo, removing the need for transformation,
saving us an object allocation and a few CPU cycles.

Change-Id: I7e34ad5e24dc6470d16243dad5ce7ba7395c5397
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Use SettableFuture<Empty> for readinessFuture

java.lang.Void is promoting use of nulls, whereas Empty.value() is
a non-null object. Migrate internal users.

Change-Id: I6a8a85917d9c3a8c3d5a9aadf691c9d1a63a767b
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Use Empty instead of Void in cohorts

Void is inherently null-happy, switch to Empty, which has a proper
value.

Change-Id: I6ce60a0933e15e7b779fcbef149481c59aa93970
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Fixup comparison formatting

Make sure the '&&' is on a single line for readability.

Change-Id: I7c1f07e2c967c3c1bff1826afdde94a59a0c7cb9
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump upstream SNAPSHOTS

Adopt:
- odlparent-10.0.0
- yangtools-8.0.0-SNAPSHOT
- mdsal-9.0.0-SNAPSHOT

Change-Id: I60991718a6d1ef6f108c2b1f7b2c09b0c5f973c4
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump mdsal to 8.0.12

Pick up the single fix from upstream.

Change-Id: I27e3c14ae8dc86110932466519ba4b824ded9c21
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
(cherry picked from commit 6fe3278a3a5421758cc49f7e65097887cd4a5c48)

Fix yangtools mis-alignment

We have a mis-aligned yangtools versions, fix that up.

Change-Id: I5153b518d2146eef8e3d7cb239a05686515755cc
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
(cherry picked from commit fcf0f01fbb4ea92258699ac40bdd5143a06d51c9)

Bump upstream versions

Adopt:
- odlparent-9.0.13
- yangtools-7.0.13
- mdsal-8.0.11

Change-Id: I4987c100930e883b51f8b519d2a1f9696196a032
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
(cherry picked from commit 40929a81c7a1e38b2cba76da0e684a1368d0e2ee)

Remove odl-controller-blueprint from odl-mdsal-broker

Only a few services need ODL blueprint extensions, minimize their
exposure by not installing them for odl-mdsal-broker.

JIRA: CONTROLLER-2006
Change-Id: I63f4b8b977d993b989f45351b7b22ae3bd92dbec
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Migrate to Empty.value()

Empty.getInstance() is deprecated for removal, use its replacement.

Change-Id: I07d6796e8211fa1ef9a08bc8bc8826d66c1b7ec9
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Fix clearing of candidates from previous iterations

During initialization the CandidateRegistry wasnt clearing
candidates from the previous instance in a reliable way.
Event with ReadMajority if the cluster members havent been updated
yet, it acted as if there were no candidates present so it did
not clear them.

Introduce a mechansim that goes through the akka singleton that
picks owners, so it only executes the clearing of candidates once
the cluster is in a healthy state.

JIRA: CONTROLLER-2025
Change-Id: I5a43908e45273adfea325621189f633ce493ca87
Signed-off-by: Tomas Cere <tomas.cere@pantheon.tech>
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump upstream versions

Adopt:
- odlparent-9.0.12
- yangtools-7.0.12
- mdsal-8.0.10

Change-Id: Ia0b658d8c92e5ae41e50e13f01c848fb4ffb6b81
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
(cherry picked from commit 99bd02d911c4c284b13dea70f4692c378636fa9d)

Restart downed nodes.

Nodes can be downed by sbr(for example when Isolated, sbr makes the
decision to down the minority), so we need to make sure they can
come up automatically unless we have another mechanism in place
for bringing up downed nodes.

JIRA: CONTROLLER-2025
Change-Id: I23d3ca2cee471c51d0eadc6c426461aa6eef193d
Signed-off-by: Tomas Cere <tomas.cere@pantheon.tech>

Enable Split-Brain Resolver

In order to make EOS service work, we need to have a provider handling
dead nodes. Enable SBR to fill that role, so that we properly recover
from nodes going away.

JIRA: CONTROLLER-2025
Change-Id: Idf817455bfe2a90d6e02011eee4ed407e1254fd2
Signed-off-by: Tomas Cere <tomas.cere@pantheon.tech>
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Switch to using tell-based protocol

Using Artery with ask-based protocol in our current settings leads
to easily running out of native memory. Switch to tell-based protocol,
which can slice its messages to fit in smaller buffers.

JIRA: CONTROLLER-1983
Change-Id: I0a296dbb3ba6e4e659c94761d78cfb985633061c
Signed-off-by: Ivan Hrasko <ivan.hrasko@pantheon.tech>
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump versions to 5.0.0-SNAPSHOT

This starts the next major version development.

Change-Id: I0bc9fed64efeee1dcb30b1a464321e9654c14eec
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump akka to 2.6.18

https://akka.io/blog/news/2021/12/20/akka-2.6.18-released

JIRA: CONTROLLER-2027
Change-Id: I7ec7581f48610d8d8479e971186a16d6b870ed9c
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump scala-java8-compat_2.13 to 1.0.2

Akka 2.6.16+ is using version 1.0.0, and they are binary compatible,
let's make sure we align to a newer version as well.

https://github.com/scala/scala-java8-compat/releases/tag/v1.0.0
https://github.com/scala/scala-java8-compat/releases/tag/v1.0.1
https://github.com/scala/scala-java8-compat/releases/tag/v1.0.2

Change-Id: I6c5f21c296dae9ca7236487e460755f0ef3e7622
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump scala to 2.13.8

https://github.com/scala/scala/releases/tag/v2.13.7
https://github.com/scala/scala/releases/tag/v2.13.8

Change-Id: Ifc5ac268e531b10a9920c7a16bfbdbb3de12d611
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Rework karaf CLI commands to print results on console

Some of these karaf CLI commands may produce return values and some CSIT components need to
read them from console.

Change-Id: Iae54c9481515c1e8fc42e9b1cf7b1530d523a142
Signed-off-by: Dominik Vrbovsky <dominik.vrbovsky@pantheon.tech>

Bump upstream versions

Adopt:
- odlparent-9.0.10
- yangtools-7.0.11
- mdsal-8.0.9

Change-Id: I24be0e4be695c02720f139b25e0be9f7854410e7
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump upstream versions

Adopt:
- odlparent-9.0.9
- yangtools-7.0.10
- mdsal-8.0.8

Change-Id: Ifbfb9c8b335bbf49df4a2965c5a4744ce558089b
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Cleanup asserts a bit

Rather than using String operations, use assertThat() and matchers.

Change-Id: I0e76854950965fbfef739c38b061ccda1a9f1e37
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

akka.actor.provider set to 'cluster'

A number of sal-akka-raft test logs are polluted by ConfigurationException.
ActorSystem needs to have 'akka.actor.provider' set to 'cluster' in the
configuration.

JIRA: CONTROLLER-2023
Change-Id: If6e873864ffe0ea8261c80c233b6d40459d05164
Signed-off-by: Dominik Vrbovsky <dominik.vrbovsky@pantheon.tech>

Bump versions to 4.0.8-SNAPSHOT

This starts the next development iteration.

Change-Id: I96de621e9636e25b8855e38c0d0cfc9c19fc063e
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Add an actor for entity rpc execution.

Instead of entity-owners rpcs being handled by owner supervisor
directly add an actor that handles the execution, preventing failures
that happen after the akka singleton has moved to a different node.

This new actor first attempts to retrieve current data from the supervisor
and if that fails, falls back to using distributed-data.

JIRA: CONTROLLER-2010
Change-Id: Idf71d66d380402c10e276c726edef2f5034f7363
Signed-off-by: Tomas Cere <tomas.cere@pantheon.tech>
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Cleanup test format

We have wrong spaces, fix them up.

Change-Id: I166314fe85d350292fa25f36cfdbb7cd5424be76
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Use DateTimeFormatter in ShardStats

This conversion allows us to drop a global lock. Also clean up the
associated test.

Change-Id: Ie91e99dd0d1ba4726b802c1c8953cfe1226714e1
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Handle ModifyTransactionRequest in forward path

When we have remote -> local handoff, we will use a
ModifyTransactionRequest, which we need to apply before moving on.

JIRA: CONTROLLER-2022
Change-Id: I04bdfda162d474966b153223c7480fd5b3af8099
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Modernize TransactionRateLimitingCallbackTest

Use assertThrows() and mockito runner.

Change-Id: I8cda7b267ebc923304e5b1b60c42d78a70d17d19
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Modernize ThreePhaseCommitCohortProxyTest

Use List.of(), assertEquals() and mockito runner.

Change-Id: Ide20b467ddc7128680f3f29668f4e978a978696e
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Cleanup  AbstractDOMBrokerWriteTransactionTest

Use mockito runner instead of initMocks().

Change-Id: I205513eb6600deb9380db9b0380e2f70ac556f89
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Clean up ClientBackedReadTransactionTest

Use mockito junit runner instead of initMocks().

Change-Id: I4d51748007191df9d490bf5b25cf03acdad8cc75
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Cleanup TransactionRateLimiterTest

Use guava stopwatch and mockito runner.

Change-Id: I87fd7f6b32379307875a2d3f353e6ad2d5bc2d38
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Migrate AbstractClientHandleTest

Mockito.initMocks() is deprecated, migrate to using mockito runner.

Change-Id: I6fbf7dd3502ed054701a1f6f41cd26e537b163e7
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Fold AbstractShardManagerTest

We have only a single subclass, ShardManagerTest, hence we can reduce
complexity by merging the two classes together. Also use
MockitoJUnitRunner to initialize mocks.

Change-Id: Idadafff3a3c96b38a5695ba903998382dc4054eb
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Migrate AbstractProxyTransactionTest to MockitoJUnitRunner

Do not use Mockito.openMocks(), but rather a proper runner.

Change-Id: I26d9d15cd10f8717c7bc36ccb91591a70305bda2
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Enable testTransactionWithIsolatedLeader() for tell-based protocol

Behaviour differs between ask-based (which fails) and tell-based (which
continues retrying). Modify the test suite to account for this
difference.

JIRA: CONTROLLER-2018
Change-Id: I208fd8aa8088558dbe2160168ee35182e02c97fa
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Cleanup IntegrationTestKit

Use equality on optionals, assertThrows(), and static imports of test
DSL.

Change-Id: I1e66e5b1a3e10dd2d866f355de0d24da604d2c3a
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Fix error message concatenation

We are missing a space here, fix that up.

Change-Id: I2eba64ddcb7db14a56eb65a84d25851e4869ba5a
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Clean up modification tests

Make sure we close the read transaction and improve assertions with
Optional.

Change-Id: Ia465f64c5845e3d5e215759cc32048e06478104b
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Enable testLeadershipTransferOnShutdown()

We are looking at cohort caches here -- and those are not used by
ClientBackedDatastore.

Change-Id: I3abfa3299d8222df6edfa98cc4a3d108a05454e4
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Track skipped transactions

We are allocating transaction identifiers which are shared across
shards. As a transaction can touch only some shards, non-participating
shards will never see the transaction ID, leading to holes in their
UnsignedLongSets -- and those holes gradually eat up more and more
memory.

Track when we have such a hole and lazily forward a new request, which
purges one or more such identifiers -- plugging the memory leak.

The crux of the machinery is exchanging state between ProxyHistory
and FrontendHistoryMetadataBuilder.

JIRA: CONTROLLER-1991
Change-Id: I3817fa2841e5f9c405bb20ff1a104537ad459ce3
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Handle SnapshotComplete message

This is an internally-generated message, make sure we handle it in
support to indicate when we have finished completely propagating
state.

JIRA: CONTROLLER-2019
Change-Id: I4f5c5231070140d54531c2e5379dffcaf1c5c4ee
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Disable state tracking on ReadyLocalTransaction

We are failing to recognize ReadyLocalTransaction as an ask-based
message, leading to us populating state.

JIRA: CONTROLLER-2021
Change-Id: I48b3c57e11430a5ca15e0559f2a8e13627f4bab1
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Drop a test re-enable FIXME

Add an explicit JIRA tag to the FIXME.

Change-Id: If567cf66476a16ce3193d1a64e18538d7bdcab53
JIRA: CONTROLLER-2020
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Add update FIXME to enable test

Add an explicit reference to the issue tracking enablement of this
test.

JIRA: CONTROLLER-2018
Change-Id: I62fb6377e138cfb00deb0a3fd9093ec13f76dc46
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Move assumption

Let's assume things only just before we are hitting and error and tag
the FIXME with an issue.

JIRA: CONTROLLER-2017
Change-Id: I114ff0f3ef2e72fdc1be5e8eb61adac8ed00ff42
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Add a FIXME for enabling a test

Record the problem observed and assign an issue to track it.

JIRA: CONTROLLER-2016
Change-Id: Idbf52a2598a93f55797954223de98cd974abac3a
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Rename UnsignedLongSet.size()

Since we have 'Set' in our name, size() invokes Set.size(), which is
not we provide w.r.t. the number of distinct elements that are contained
in the set.

We are communicating how many ranges are stored, hence the method's name
is 'rangeSize()'.

Change-Id: Iaf9999db86fe03a8e753107638fc2e189b680677
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Make UnsignedLongSet.Entry immutable

Having entries mutable is just a drag when transferring them to
mutable, as we need to perform deep copies. Let's turn them into
immutables and be done with it.

JIRA: CONTROLLER-2015
Change-Id: I3be807cbdf71a51e6b9506e0932857a230924986
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Fix UnsignedLongSet entry lifecycle

When we are copying things around, we have to ensure we copy Entries,
because they cannot be shared.

JIRA: CONTROLLER-2015
Change-Id: I8a484a3bf52ffeaf4f4d0e770bc77dba31dbb221
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Fix Eclipse compilation error

JDT has trouble with Scala classes here, make sure we specify
more general scala.collections.Set instead of its immutable counterpart.

Change-Id: I60b56e6b7ec5705409aee8d1e30a8ae9e5579bfb
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Fixup assertions a bit

The tests are using different transaction mix, pass down explicit
expected transaction count.

Change-Id: I5007fe0ad90009fcafb727bac8d37bd07dfaae94
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Improve tell-based protocol assumptions

Make failures a bit more visible by comparing human-readable things.

Change-Id: I682c020be348c61fcdf4c3f31949ba0f666af020
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Partially enable testSingleTransactionsWritesInQuickSuccession()

Split the asserts into two methods, asserting the two datastore
options. The ask-based path remains disabled because it currently
fails tests.

Also improve tell-based protocol assertions, which need to wait
for the purge process to settle.

Change-Id: I8b2f3d84b2c7cd01dec4f7994eda716d022a98aa
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Cleanup DistributedDataStoreRemotingIntegrationTest

We have a few @Ignored tests and assumptions scattered a bit, perform
a general cleanup on them.

JIRA: CONTROLLER-1991
Change-Id: I70054276b5ffa179337a302e5d2498e925b65ce8
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Cleanup JsonExportActor

Remove unneeded 'this.' qualifiers and using an
EffectiveStatementInference instead of SchemaPath when instantiating
JSON codec.

Change-Id: Iac6b980771c4ad395341a9c88ee0d16c34d8a02a
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Cleanup ShardInformation

Drop unneeded 'this.' qualifiers and use a simple return expression
in getSerializedLeaderActor().

Change-Id: I9a11410f8bc6d8246bacf99fa86c42de85c328ae
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Cleanup ShardWriteTransaction

A few arguments can be made final and an if/else block can be
simplified.

Change-Id: Ib5464482b4e4cc1d73f755b4b467470fc3b2af25
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Clean up AbstractTransactionContextFactory

Add a few @NonNull annotations and eliminate two else blocks.

Change-Id: Ide21c363f349af7b24fb4835b859a224357586c8
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Expose ImmutableList from FrontendClientMetadata

ImmutableList is an API contract, exposed it to callers.

Change-Id: I7fb2a09c2b225ce48ece7400128c8215d70292fe
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Cleanup cds-access-api dependencies

We are using Guava, declare that as a dependency.

Change-Id: Ibe905f0d2d87c5e7dfb4fad598acb27b4323c7b8
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Add MutableUnsignedLongSet.addAll()

Add the ability to merge two UnsignedLongSets. This is useful when
we are using ImmutableUnsignedLongSet as a data interchange format
to communicate changes to a MutableUnsignedLongSet.

JIRA: CONTROLLER-2015
Change-Id: Ia84474b685872586722914f20db8f96a3c172f97
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Add UnsignedLongBitmap

Rather than using a ImmutableMap<UnsignedLong, Boolean>, we can do
have a much denser representation with a specialized class. Here we
are introducing an explicit UnsignedLongBitmap, which stores an array
of longs and an array of booleans, ditching the intermediate objects.

Also clean up error reporting, throwing an IOException instead of a
VerifyException when things should go south unexpectedly.

JIRA: CONTROLLER-2013
Change-Id: Ie64da0af68ea2898dc77368afd4fce8abd2cccea
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Sprinkle @NonNull annotations

We have a few return methods which could use @NonNull annotations,
sprinkle them around.

Change-Id: I48c0f8879127673eb122321dd6c92d4572098f68
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Inline Entry.contains()

MutableUnsignedLongSet.add() has half of the check done in contains()
as an invariant. Inline the second part there, improving performance a
bit.

Thisleaves only UnsignedLongSet.contains() as the sole caller of
Entry.contains(). Inline it there, reducing the footprint a bit.

JIRA: CONTROLLER-2014
Change-Id: Ie38283a1fc46aff8fbc717a23aa9a068805636db
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Remove UnsignedLongSet.addImpl()

We now do not share utilities in UnsignedLongSet, hence relocate
implementation code to MutableUnsignedLongSet.add().

JIRA: CONTROLLER-2014
Change-Id: Id1578a8ea639ed512d55b2e6dca26de5abdf3ab0
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Improve UnsignedLongSet efficiency

We can use NavigableMap's higher() and floor() methods to elide object
object allocations.

JIRA: CONTROLLER-2014
Change-Id: I75747c58027caee2678e32187dbd628d60ce6604
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Bump versions to 4.0.7-SNAPSHOT

This starts the next development iteration.

Change-Id: I837a7d5dcc6df7deb8360d3a662d0628fb1a199f
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Minor sal-distributed-datastore cleanups

Corrent a @GuardedBy annotations, add @NonNull annotations, fix javadocs
and make a few methods final.

Change-Id: Icb93aae229fb0aece5ece223a759ce9f36ee0297
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Use ImmutableList.Builder in FrontendClientMetadata

We are performing a defensive copy in constructor, hence it is
better to allocate a builder, skipping an untrusted copy operation.

Change-Id: I95d7aa90f6da325aaae6a340caf49a6e17fe9f64
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Use ImmutableSortedSet for small ImmutableUnsignedLongSets

This adds a bit of indirection, but for a few entries this ends up
using arrays instead of an RB tree, offering better density.

JIRA: CONTROLLER-2012
Change-Id: I09714c6bf3272946f6cee4360f80a106453d0e61
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Use UnsignedLongSet instead of RangeSet in metadata

Split UnsignedLongSet into two implementations, mutable and immutable.
Use the mutable implementation in runtime tracking of identifiers and
use the immutable implementation for tracking identifiers in metadata.

The serialization format is kept compatible with RangeSets, although
that implies a minor penalty in serdes.

This switch ends up potentially using more objects for small sets, but
that is offset by not having Cut indirections and most notably being
resistent to allocation of huge arrays.

JIRA: CONTROLLER-2011
Change-Id: I0c84ffaaa4ce39299cef9006784b8aff78dd0f83
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>

Reimplement UnsignedLongRangeSet

We are tracking simple discrete values, for which a RangeSet is an
overkill due to it being generalized utility working with all possible
ranges.

Reimplement UnsignedLongRangeSet as UnsignedLongSet, tracking discrete
closed ranges of longs, which improves both memory footprint and speed
of contains()/add() operations.

This also spots a failure to disconnect purged histories during
FrontendHistoryMetadataBuilder.toLeaderState().

JIRA: CONTROLLER-1720
Change-Id: I24fae4174201fd133a282f27589d6c274c06c8dc
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>