The implementation of KIP-258 broke the state store methods in KStreamTestDriver.
These methods were unused in this project, so the breakage was not detected.
Since this is an internal testing utility, and it was deprecated and partially removed in
favor of TopologyTestDriver, I opted to just complete the removal of the class.
Reviewers: A. Sophie Blee-Goldman <ableegoldman@gmail.com>, Boyang Chen <boyang@confluent.io>, Bill Bejeck <bill@confluent.io>, Matthias J. Sax <matthias@confluent.io>, Guozhang Wang <wangguoz@gmail.com>
This changes the field "generationid" to "generationId" to be consistent with other uses.
Reviewers: Shaobo Liu <lambda.tencent@gmail.com>, Jason Gustafson <jason@confluent.io>
Some shell scripts don't have corresponding batch files in bin\windows.
For improving Windows platform support, This PR adds the following batch files:
- bin\windows\kafka-delete-records.bat
- bin\windows\kafka-log-dirs.bat
Reviewers: Jason Gustafson <jason@confluent.io>
For static members join/rejoin, we encode the current timestamp in the new member.id. The format looks like group.instance.id-timestamp.
During consumer/broker interaction logic (Join, Sync, Heartbeat, Commit), we shall check the whether group.instance.id is known on group. If yes, we shall match the member.id stored on static membership map with the request member.id. If mismatching, this indicates a conflict consumer has used same group.instance.id, and it will receive a fatal exception to shut down.
Right now the only missing part is the system test. Will work on it offline while getting the major logic changes reviewed.
Reviewers: Ryanne Dolan <ryannedolan@gmail.com>, Jason Gustafson <jason@confluent.io>, Guozhang Wang <wangguoz@gmail.com>
This patch includes API changes for follower fetching per [KIP-392](https://cwiki.apache.org/confluence/display/KAFKA/KIP-392%3A+Allow+consumers+to+fetch+from+closest+replica) as well as the consumer implementation. After this patch, consumers will continue to fetch only from the leader, since the broker implementation to select an alternate read replica is not included here.
Adds new `client.rack` consumer configuration property is added which allows the consumer to indicate its rack. This is just an arbitrary string to indicate some relative location, it doesn't have to actually represent a physical rack. We are keeping the naming consistent with the broker property (`broker.rack`).
FetchRequest now includes `rack_id` which can optionally be specified by the consumer. FetchResponse includes an optional `preferred_read_replica` field for each partition in the response. OffsetForLeaderEpochRequest also adds new `replica_id` field which is similar to the same field in FetchRequest.
When the consumer sees a `preferred_read_replica` in a fetch response, it will use the Node with that ID for the next fetch.
Reviewers: Jason Gustafson <jason@confluent.io>
Deletion of a large number of topics can cause a ton of log spam. In a test case on 2.2, deletion of 50 topics with 100 partitions each caused about 158 Mb of data in the controller log. With the improvements to batch StopReplica and the patch here, we reduce that to about 1.5 Mb.
Kudos to gwenshap for spotting these spammy messages.
Author: Jason Gustafson <jason@confluent.io>
Reviewers: Gwen Shapira
Closes#6738 from hachikuji/remove-verbose-topic-deletion-log-message
Basic idea:
KTable-KTable join: set max(left-ts,right-ts) for result
#agg(...) (stream/table windowed/non-windowed): set max(ts1, ts2, ts3,...) of all input records that contribute to the aggregation result
for all stateless transformation: input-ts -> output-ts
Reviewers: Guozhang Wang <wangguoz@gmail.com>, John Roesler <john@confluent.io>, Andy Coates <andy@confluent.io>, Bill Bejeck <bbejeck@gmail.com
The replica fetcher thread is terminated in case a partition crashes which leads to under replication. This behavior can be improved by dropping the failed partition. The thread can continue monitoring the rest of the partitions. If all partitions of a thread have failed, the thread would be shut down. This is documented in KIP-461: https://cwiki.apache.org/confluence/display/KAFKA/KIP-461+-+Improve+Replica+Fetcher+behavior+at+handling+partition+failure.
Reviewers: Jun Rao <junrao@gmail.com>, Jason Gustafson <jason@confluent.io>
When users provide a name for operation via the Streams DSL, we need to increment the counter used for auto-generated names to make sure any operators downstream of a named operator still produce a compatible name.
This PR is a subset of #6411 by @fhussonnois. We need to merge this PR now because it covers cases when users name repartition topics or state stores.
Updated tests to reflect the counter produces expected number even when the user provides a name.
Matthias J. Sax <mjsax@apache.org>, John Roesler <john@confluent.io>
This upgrade exposes a number of new options, including the WriteBufferManager which -- along with existing TableConfig options -- allows users to limit the total memory used by RocksDB across instances. This can alleviate some cascading OOM potential when, for example, a large number of stateful tasks are suddenly migrated to the same host.
The RocksDB docs guarantee backwards format compatibility across versions
Reviewers: Matthias J. Sax <mjsax@apache.org>, Bill Bejeck <bbejeck@gmail.com>,
Initiate `unreachable server` scenario before starting admin client to avoid timing issues if node is disconnected from the test thread while admin client network thread is processing a metadata request.
Reviewers: Ismael Juma <ismael@juma.me.uk>
Implementation to enable policy for Connector Client config overrides. This is
implemented per the KIP-458.
Reviewers: Randall Hauch <rhauch@gmail.com>
Added the incremental cooperative rebalancing in Connect to avoid global rebalances on all connectors and tasks with each new/changed/removed connector. This new protocol is backward compatible and will work with heterogeneous clusters that exist during a rolling upgrade, but once the clusters consist of new workers only some affected connectors and tasks will be rebalanced: connectors and tasks on existing nodes still in the cluster and not added/changed/removed will continue running while the affected connectors and tasks are rebalanced.
This commit attempted to minimize the changes to the existing V0 protocol logic, though that was not entirely possible.
This commit adds extensive unit and integration tests for both the old V0 protocol and the new v1 protocol. Soak testing has been performed multiple times to verify behavior while connectors and added, changed, and removed and while workers are added and removed from the cluster.
Author: Konstantine Karantasis <konstantine@confluent.io>
Reviewers: Randall Hauch <rhauch@gmail.com>, Ewen Cheslack-Postava <me@ewencp.org>, Robert Yokota <rayokota@gmail.com>, David Arthur <mumrah@gmail.com>, Ryanne Dolan <ryannedolan@gmail.com>
Per KIP-465, kept existing behavior of `/connectors` resource in the Connect's REST API, but added the ability to specify `?expand` query parameter to get list of connectors with status details on each connector. Added unit tests, and verified passing existing system tests (which use the older list form).
See https://cwiki.apache.org/confluence/display/KAFKA/KIP-465%3A+Add+Consolidated+Connector+Endpoint+to+Connect+REST+API.
Author: Dan Norwood <norwood@confluent.io>
Reviewer: Randall Hauch <rhauch@gmail.com>
This fixes a regression caused by KAFKA-8275. The least loaded node selection
should take into account nodes which are currently being connect to. This
includes both the CONNECTING and CHECKING_API_VERSIONS states since
`canSendRequest` would return false in either case.
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>
When choosing the next record to process, we should look at the head record's timestamp of each partition and choose the lowest rather than choosing the lowest of the partition's streamtime.
This change effectively makes RecordQueue return the timestamp of the head record rather than its streamtime. Streamtime is removed (replaced) from RecordQueue as it was only being tracked in order to choose the next partition to poll from.
Reviewers: Matthias J. Sax <mjsax@apache.org>, Bill Bejeck <bbejeck@gmail.com>
To make static consumer group members more persistent, we want to avoid kicking out unjoined members through rebalance timeout. Essentially we allow static members to participate in a rebalance using their old subscription without sending a JoinGroup. The only catch is that an unjoined static member might be the current group leader, and we may need to elect a different leader.
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Jason Gustafson <jason@confluent.io>
See https://cwiki.apache.org/confluence/display/KAFKA/KIP-449%3A+Add+connector+contexts+to+Connect+worker+logs
Added LoggingContext as a simple mechanism to set and unset Mapped Diagnostic Contexts (MDC) in the loggers to provide for each thread useful parameters that can be used within the logging configuration. MDC avoids having to modify lots of log statements, since the parameters are available to all log statements issued by the thread, no matter what class makes those calls.
The design intentionally minimizes the number of changes to any existing classes, and does not use Java 8 features so it can be easily backported if desired, although per this KIP it will be applied initially only in AK 2.3 and later and must be enabled via the Log4J configuration.
Reviewers: Jason Gustafson <jason@conflent.io>, Guozhang Wang <wangguoz@gmail.com>
WorkerSourceTask is catching the exception from wrong package org.apache.kafka.common.errors. It is not clear from the API standpoint as to which package the connect framework supports - the one from common or connect. The safest thing would be to support both the packages even though it's less desirable.
Author: Magesh Nandakumar <magesh.n.kumar@gmail.com>
Reviewers: Arjun Satish <arjun@confluent.io>, Randall Hauch <rhauch@gmail.com>
SslFactory: split the part of SslFactory that creates SSLEngine instances into SslEngineBuilder. When (re)configuring, we simply create a new SslEngineBuilder. This allows us to make all the builder fields immutable. It also simplifies the logic for reconfiguring. Because we sometimes need to test old SslEngine instances against new ones, being able to use both the old and the new builder at once is useful.
Create an enum named SslClientAuth which encodes the possible values for ssl.client.auth. This will simplify the handling of this configuration.
SslTransportLayer#maybeProcessHandshakeFailure should treat an SSLHandshakeException with a "Received fatal alert" message as a handshake error (and therefore an authentication error.)
SslFactoryTest: add some line breaks for very long lines.
ConfigCommand#main: when terminating the command due to an uncaught exception, log the exception using debug level in slf4j, in addition to printing it to stderr. This makes it easier to debug failing junit tests, where stderr may not be kept, or may be reordered with respect to other slf4j messages. The use of debug level is consistent with how we handle other types of exceptions in ConfigCommand#main.
StateChangeLogMerger#main: spell out the full name of scala.io.Source rather than abbreviating it as io.Source. This makes it clearer that it is part of the Scala standard library. It also avoids compiler errors when other libraries whose groupId starts with "io" are used in the broker.
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>
Enable reconfiguration of SSL keystores and truststores in client-side channel builders used by brokers for controller, transaction coordinator and replica fetchers. This enables brokers using TLS mutual authentication for inter-broker listener to use short-lived certs that may be updated before expiry without restarting brokers.
Reviewers: Manikumar Reddy <manikumar.reddy@gmail.com>
Updated AbstractConfig to be able to resolve variables in config values when the configuration includes config provider properties.
Author: Tejal Adsul <tejal@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@gmail.com>, Randall Hauch <rhauch@gmail.com>
`ConsumerTopicCreationTest` relied on `KafkaConsumer#poll` to send a `MetadataRequest` within 100ms to verify if a topic is auto created or not. This is brittle and does not guarantee if the request made it to the broker or was processed successfully. This PR fixes the flaky test by adding another topic; we wait until we consume a previously produced record to this topic. This ensures MetadataRequest was processed and we could then check if the topic we're interested in was created or not.
Reviewers: Boyang Chen <bchen11@outlook.com>, Jason Gustafson <jason@confluent.io>
MINOR: update documentation for the log cleaner max compaction lag feature (KIP-354) implemented in KAFKA-7321
Author: Xiongqi Wu <xiowu@linkedin.com>
Reviewer: Joel Koshy <jjkoshy@gmail.com>
The main problem we are trying to solve here is the batching of StopReplica requests and the lack of test coverage for `ControllerChannelManager`. Addressing the first problem was straightforward, but the second problem required quite a bit of work because of the dependence on `KafkaController` for all of the events. It seemed to make sense to separate the events from the processing of events so that we could remove this dependence and improve testability. With the refactoring, I was able to add test cases covering most of the logic in `ControllerChannelManager` including the generation of requests and the expected response handling logic. Note that I have not actually changed any of the event handling logic in `KafkaController`.
While refactoring this logic, I found that the event queue time metric was not being correctly computed. The problem is that many of the controller events were singleton objects which inherited the `enqueueTimeMs` field from the `ControllerEvent` trait. This would never get updated, so queue time would be skewed.
Reviewers: Jun Rao <junrao@gmail.com>
Currently in maybeFailWithError, we always wrap the lastError as a KafkaException. For ProducerFencedException however, we should just throw the exception itself; however we throw a new instance since the previous book-kept one's call trace is not from this call, and hence could be confusing.
Reviewers: Boyang Chen <boyang@confluent.io>, Jason Gustafson <jason@conflent.io>
KAFKA-7321: Add a Maximum Log Compaction Lag (KIP-354)
Records become eligible for compaction after the specified time interval.
Author: Xiongqi Wu <xiowu@linkedin.com>
Reviewer: Joel Koshy <jjkoshy@gmail.com>
The log cleaner attempts to preserve the last entry for each producerId in order to ensure that sequence/epoch state is not lost. The current validation checks only the last sequence number for each producerId in order to decide whether a batch should be retained. There are two problems with this:
1. Sequence numbers are not unique alone. It is the tuple of sequence number and epoch which is uniquely defined.
2. The group coordinator always writes batches beginning with sequence number 0, which means there could be many batches which have the same sequence number.
The complete fix for the second issue would probably add proper sequence number bookkeeping in the coordinator. For now, we have left the coordinator implementation unchanged and changed the cleaner logic to use the last offset written by a producer instead of the last sequence number.
Reviewers: Guozhang Wang <wangguoz@gmail.com>
For session-windows, the result record should have the window-end timestamp as record timestamp.
Rebased to resolve merge conflicts. Removed unused classes TupleForwarder and ForwardingCacheFlushListener (replace with TimestampedTupleForwarder, SessionTupleForwarder, TimestampedCacheFlushListerner, and SessionCacheFlushListener)
Reviewers: John Roesler <john@confluent.io>, Bruno Cadonna <bruno@confluent.io>, Boyang Chen <boyang@confluent.io>, Bill Bejeck <bill@confluent.io>, Guozhang Wang <wangguoz@gmail.com>
Consolidated the unit tests by having {RocksDB/InMemory}{Window/Session}StoreTest extend {Window/Session}BytesStoreTest. Besides some implementation-specific tests (eg involving segment maintenance) all tests were moved to the abstract XXXBytesStoreTest class. The test coverage now is a superset of the original test coverage for each store type.
The only difference made to existing tests (besides moving them) was to switch from list-based equality comparison to set based, in order to reflect that the stores make no guarantees regarding the ordering of records returned from a range fetch.
There are some implementation-specific tests that were left in the corresponding test class. The RocksDBWindowStoreTest, for example, had several tests pertaining to segments and/or the underlying filesystem. Another key difference is that the in-memory versions should delete expired records aggressively, while the RocksDB versions should only remove entirely expired segments.
Reviewers: John Roesler <john@confluent.io>, Guozhang Wang <wangguoz@gmail.com>
When Kafka Connect does not have cluster ACLs to create topics,
it fails to even access its internal topics which already exist.
This was originally fixed in KAFKA-6250 by ignoring the cluster
authorization error, but now Kafka 2.0 returns a different response
code that corresponds to a different error. Add a patch to ignore this
new error as well.
Reviewers: Jason Gustafson <jason@confluent.io>
Due to the lack of a blank line, a code section in TROGDOR.md
is not properly rendered. This PR fixes it.
Reviewers: Jason Gustafson <jason@confluent.io>
John Roesler
Jason Gustafson
Boyang Chen
Kengo Seki
commandini
David Arthur
Matthias J. Sax
Aishwarya Gune
Bill Bejeck
A. Sophie Blee-Goldman
Shaobo Liu
Rajini Sivaram
Magesh Nandakumar
Konstantine Karantasis
dan norwood
Mickael Maison
sandmannn
Randall Hauch
Paul Davidson
Colin Patrick McCabe
Ismael Juma
tadsul
Dhruvil Shah
Chris Egerton
Xiongqi Wu
Guozhang Wang
Xiongqi Wu
Vahid Hashemian
Arabelle Hou