We have a Kafka streams application that reads a record from one source Kafka topic does some validation, integration with some other systems via some API calls, based on integration results it builds up some other record and publishes it in a destination Kafka topic.
Both source and destination Kafka topics are created with CreateTime timestamp type. I am not sure if this timestamp type was set this way because of some business needs or just because this is the default timestamp type if you don't set it yourself.
Now we have a performance measuring tool that is capable of injecting a big load of messages into Kafka source topic as well as consuming the processing results from Kafka destination topic. When doing this it records the timestamp of each record into an embedded database for each key then evaluates the various percentiles we measure the performance for.
Because of topics timestamp type are set to CreateTime this is not working because both start time and end time have exactly the same value. Changing timestamp type to LogAppendTime for destination Kafka topic will solve our problem. However even if I am not aware about any business requirements to have CreateTime as topics timestamp types it does not mean there could not be such a requirement. Modifying your infrastructure design to satisfy your testing needs sounds to me a bad approach.
Wondering if there is another more elegant way of achieving this.
Thank you in advance for your inputs/suggestions.
Related
We have a micro-services architecture, with Kafka used as the communication mechanism between the services. Some of the services have their own databases. Say the user makes a call to Service A, which should result in a record (or set of records) being created in that service’s database. Additionally, this event should be reported to other services, as an item on a Kafka topic. What is the best way of ensuring that the database record(s) are only written if the Kafka topic is successfully updated (essentially creating a distributed transaction around the database update and the Kafka update)?
We are thinking of using spring-kafka (in a Spring Boot WebFlux service), and I can see that it has a KafkaTransactionManager, but from what I understand this is more about Kafka transactions themselves (ensuring consistency across the Kafka producers and consumers), rather than synchronising transactions across two systems (see here: “Kafka doesn't support XA and you have to deal with the possibility that the DB tx might commit while the Kafka tx rolls back.”). Additionally, I think this class relies on Spring’s transaction framework which, at least as far as I currently understand, is thread-bound, and won’t work if using a reactive approach (e.g. WebFlux) where different parts of an operation may execute on different threads. (We are using reactive-pg-client, so are manually handling transactions, rather than using Spring’s framework.)
Some options I can think of:
Don’t write the data to the database: only write it to Kafka. Then use a consumer (in Service A) to update the database. This seems like it might not be the most efficient, and will have problems in that the service which the user called cannot immediately see the database changes it should have just created.
Don’t write directly to Kafka: write to the database only, and use something like Debezium to report the change to Kafka. The problem here is that the changes are based on individual database records, whereas the business significant event to store in Kafka might involve a combination of data from multiple tables.
Write to the database first (if that fails, do nothing and just throw the exception). Then, when writing to Kafka, assume that the write might fail. Use the built-in auto-retry functionality to get it to keep trying for a while. If that eventually completely fails, try to write to a dead letter queue and create some sort of manual mechanism for admins to sort it out. And if writing to the DLQ fails (i.e. Kafka is completely down), just log it some other way (e.g. to the database), and again create some sort of manual mechanism for admins to sort it out.
Anyone got any thoughts or advice on the above, or able to correct any mistakes in my assumptions above?
Thanks in advance!
I'd suggest to use a slightly altered variant of approach 2.
Write into your database only, but in addition to the actual table writes, also write "events" into a special table within that same database; these event records would contain the aggregations you need. In the easiest way, you'd simply insert another entity e.g. mapped by JPA, which contains a JSON property with the aggregate payload. Of course this could be automated by some means of transaction listener / framework component.
Then use Debezium to capture the changes just from that table and stream them into Kafka. That way you have both: eventually consistent state in Kafka (the events in Kafka may trail behind or you might see a few events a second time after a restart, but eventually they'll reflect the database state) without the need for distributed transactions, and the business level event semantics you're after.
(Disclaimer: I'm the lead of Debezium; funnily enough I'm just in the process of writing a blog post discussing this approach in more detail)
Here are the posts
https://debezium.io/blog/2018/09/20/materializing-aggregate-views-with-hibernate-and-debezium/
https://debezium.io/blog/2019/02/19/reliable-microservices-data-exchange-with-the-outbox-pattern/
first of all, I have to say that I’m no Kafka, nor a Spring expert but I think that it’s more a conceptual challenge when writing to independent resources and the solution should be adaptable to your technology stack. Furthermore, I should say that this solution tries to solve the problem without an external component like Debezium, because in my opinion each additional component brings challenges in testing, maintaining and running an application which is often underestimated when choosing such an option. Also not every database can be used as a Debezium-source.
To make sure that we are talking about the same goals, let’s clarify the situation in an simplified airline example, where customers can buy tickets. After a successful order the customer will receive a message (mail, push-notification, …) that is sent by an external messaging system (the system we have to talk with).
In a traditional JMS world with an XA transaction between our database (where we store orders) and the JMS provider it would look like the following: The client sets the order to our app where we start a transaction. The app stores the order in its database. Then the message is sent to JMS and you can commit the transaction. Both operations participate at the transaction even when they’re talking to their own resources. As the XA transaction guarantees ACID we’re fine.
Let’s bring Kafka (or any other resource that is not able to participate at the XA transaction) in the game. As there is no coordinator that syncs both transactions anymore the main idea of the following is to split processing in two parts with a persistent state.
When you store the order in your database you can also store the message (with aggregated data) in the same database (e.g. as JSON in a CLOB-column) that you want to send to Kafka afterwards. Same resource – ACID guaranteed, everything fine so far. Now you need a mechanism that polls your “KafkaTasks”-Table for new tasks that should be send to a Kafka-Topic (e.g. with a timer service, maybe #Scheduled annotation can be used in Spring). After the message has been successfully sent to Kafka you can delete the task entry. This ensures that the message to Kafka is only sent when the order is also successfully stored in application database. Did we achieve the same guarantees as we have when using a XA transaction? Unfortunately, no, as there is still the chance that writing to Kafka works but the deletion of the task fails. In this case the retry-mechanism (you would need one as mentioned in your question) would reprocess the task an sends the message twice. If your business case is happy with this “at-least-once”-guarantee you’re done here with a imho semi-complex solution that could be easily implemented as framework functionality so not everyone has to bother with the details.
If you need “exactly-once” then you cannot store your state in the application database (in this case “deletion of a task” is the “state”) but instead you must store it in Kafka (assuming that you have ACID guarantees between two Kafka topics). An example: Let’s say you have 100 tasks in the table (IDs 1 to 100) and the task job processes the first 10. You write your Kafka messages to their topic and another message with the ID 10 to “your topic”. All in the same Kafka-transaction. In the next cycle you consume your topic (value is 10) and take this value to get the next 10 tasks (and delete the already processed tasks).
If there are easier (in-application) solutions with the same guarantees I’m looking forward to hear from you!
Sorry for the long answer but I hope it helps.
All the approach described above are the best way to approach the problem and are well defined pattern. You can explore these in the links provided below.
Pattern: Transactional outbox
Publish an event or message as part of a database transaction by saving it in an OUTBOX in the database.
http://microservices.io/patterns/data/transactional-outbox.html
Pattern: Polling publisher
Publish messages by polling the outbox in the database.
http://microservices.io/patterns/data/polling-publisher.html
Pattern: Transaction log tailing
Publish changes made to the database by tailing the transaction log.
http://microservices.io/patterns/data/transaction-log-tailing.html
Debezium is a valid answer but (as I've experienced) it can require some extra overhead of running an extra pod and making sure that pod doesn't fall over. This could just be me griping about a few back to back instances where pods OOM errored and didn't come back up, networking rule rollouts dropped some messages, WAL access to an aws aurora db started behaving oddly... It seems that everything that could have gone wrong, did. Not saying Debezium is bad, it's fantastically stable, but often for devs running it becomes a networking skill rather than a coding skill.
As a KISS solution using normal coding solutions that will work 99.99% of the time (and inform you of the .01%) would be:
Start Transaction
Sync save to DB
-> If fail, then bail out.
Async send message to kafka.
Block until the topic reports that it has received the
message.
-> if it times out or fails Abort Transaction.
-> if it succeeds Commit Transaction.
I'd suggest to use a new approach 2-phase message. In this new approach, much less codes are needed, and you don't need Debeziums any more.
https://betterprogramming.pub/an-alternative-to-outbox-pattern-7564562843ae
For this new approach, what you need to do is:
When writing your database, write an event record to an auxiliary table.
Submit a 2-phase message to DTM
Write a service to query whether an event is saved in the auxiliary table.
With the help of DTM SDK, you can accomplish the above 3 steps with 8 lines in Go, much less codes than other solutions.
msg := dtmcli.NewMsg(DtmServer, gid).
Add(busi.Busi+"/TransIn", &TransReq{Amount: 30})
err := msg.DoAndSubmitDB(busi.Busi+"/QueryPrepared", db, func(tx *sql.Tx) error {
return AdjustBalance(tx, busi.TransOutUID, -req.Amount)
})
app.GET(BusiAPI+"/QueryPrepared", dtmutil.WrapHandler2(func(c *gin.Context) interface{} {
return MustBarrierFromGin(c).QueryPrepared(db)
}))
Each of your origin options has its disadvantage:
The user cannot immediately see the database changes it have just created.
Debezium will capture the log of the database, which may be much larger than the events you wanted. Also deployment and maintenance of Debezium is not an easy job.
"built-in auto-retry functionality" is not cheap, it may require much codes or maintenance efforts.
I have a very common problem, yet I am unable to find the "right" or "correct" way to test this
I have a simple Spark job that gets events from Kafka (events are in protobuf format), applies some transformations on them and then stores the result in ES. I am saving single events
I need to know how to test this properly. I am using BulkProcessor, therefore, I am manually committing the offsets when I think they should be. Therefore, it makes sense to test this workflow properly because I don't want to lose events
My understanding is that I need to have a mock Kafka instance, need to call the appropriate function that handles all the transformations and then store the result in ES. However, I don't how to do all this. Also, I don't know how to write test events in protobuf format into Kafka topics
P.S. I am NOT using Spring framework
I am trying to integrate Spark/Kafka to build a streaming app.
Kakfa version: 0.9
spark: 1.6.2
how do i handle offsets after processing data in RDD batch.
Can you give me more insight on handling offsets?
Does spark had inbuilt to store and read offsets automatically? or do i need to guide spark to read offsets from some store like mongo or oracle?
JavaInputDStream<String> directKafkaStream = KafkaUtils.createDirectStream(jsc, String.class, String.class,
StringDecoder.class, StringDecoder.class, String.class, kafkaParams, topicMap,
(Function<MessageAndMetadata<String, String>, String>) MessageAndMetadata::message);
directKafkaStream.foreachRDD(rdd -> {
The answer to your question depends on your desired message delivery semantics:
at most once: each message will be processed at most once
at least once: each message will be processed at most once
exactly once: at most once and at least once at the same time
First of all, I would recommend reading those slides as well as this blog post.
I am assuming that you are pursuing exactly-once, since the remaining ones are pretty easy to figure out. Anyway, a couple of approaches to consider:
Checkpointing
Spark Streaming allows you to checkpoint your DStreams. If you use direct Stream from KafkaUtils, the offsets will be checkpointed as well. The streaming job might fail anywhere between checkpoints, so some messages might get replayed. To achieve exactly once semantics with this approach, one would have to use idempotent output operation (in other words - the downstream system is able to distinguish/ignore replayed messages).
Pros: easy to achieve; comes out-of-the-box
Cons: at least once semantics; checkpoints become invalidated after code change; offsets are stored in Spark, not in Zookeeper
Transactional data storage
You might want to store the offsets yourself in a custom data store that supports transactions, i.e a relational database like MySQL. In this case you need to make sure that processing stream and saving offsets are contained in a single transaction.
Pros: exactly once semantics
Cons: harder to set up, requires a transactional data store
WAL-based Receiver
You can use the older Kafka connector based on WAL.
Pros: works with other data sources as well; stores offsets in Zookeeper
Cons: it depends on HDFS; you cannot access offsets directly; it makes parallelism harder to achieve
To sum up, it all depends on your requirements - perhaps you can lift some restrictions to simplify this problem.
When you want to consume data from Kafka topic using Spark Streaming, there are two ways you can do that.
1.Receiver based approach
In this approach , offsets are managed in Zookeeper and it is automatically update the offsets in zookeeper. For more information.
http://spark.apache.org/docs/latest/streaming-kafka-integration.html#approach-1-receiver-based-approach
2. Direct Approach(No Receiver)
This approach is that it does not update offsets in Zookeeper, hence Zookeeper-based Kafka monitoring tools will not show progress. However, you can access the offsets processed by this approach in each batch and update Zookeeper yourself.
http://spark.apache.org/docs/latest/streaming-kafka-integration.html#approach-2-direct-approach-no-receivers
I wonder if it is possible to add a listener to Cassandra getting the table and the primary key for changed entries? It would be great to have such a mechanism.
Checking Cassandra documentation I only find adding StateListener(s) to the Cluster instance.
Does anyone know how to do this without hacking Cassandras data store or encapsulate the driver and do something on my own?
Check out this future jira --
https://issues.apache.org/jira/browse/CASSANDRA-8844
If you like it vote for it : )
CDC
"In databases, change data capture (CDC) is a set of software design
patterns used to determine (and track) the data that has changed so
that action can be taken using the changed data. Also, Change data
capture (CDC) is an approach to data integration that is based on the
identification, capture and delivery of the changes made to enterprise
data sources."
-Wikipedia
As Cassandra is increasingly being used as the Source of Record (SoR)
for mission critical data in large enterprises, it is increasingly
being called upon to act as the central hub of traffic and data flow
to other systems. In order to try to address the general need, we,
propose implementing a simple data logging mechanism to enable
per-table CDC patterns.
If clients need to know about changes, the world has mostly gone to the message broker model-- a middleman which connects producers and consumers of arbitrary data. You can read about Kafka, RabbitMQ, and NATS here. There is an older DZone article here. In your case, the client writing to the database would also send out a change message. What's nice about this model is you can then pull whatever you need from the database.
Kafka is interesting because it can also store data. In some cases, you might be able to dispose of the database altogether.
Are you looking for something like triggers?
https://github.com/apache/cassandra/tree/trunk/examples/triggers
A database trigger is procedural code that is automatically executed
in response to certain events on a particular table or view in a
database. The trigger is mostly used for maintaining the integrity of
the information on the database. For example, when a new record
(representing a new worker) is added to the employees table, new
records should also be created in the tables of the taxes, vacations
and salaries.
I am trying to resolve a design difference of opinion where neither of us has experience with JMS.
We want to use JMS to communicate between a j2ee application and the stand-alone application when a new event occurs. We would be using a single point-to-point queue. Both sides are Java-based. The question is whether to send the event data itself in the JMS message body or to send a pointer to the data so that the stand-alone program can retrieve it. Details below.
I have a j2ee application that supports data entry of new and updated persons and related events. The person records and associated events are written to an Oracle database. There are also stand-alone, separate programs that contribute new person and event records to the database. When a new event occurs through any of 5-10 different application functions, I need to notify remote systems through an outbound interface using an industry-specific standard messaging protocol. The outbound interface has been designed as a stand-alone application to support scalability through asynchronous operation and by moving it to a separate server.
The j2ee application currently has most of the data in memory at the time the event is entered. The data would consist of approximately 6 different objects; a person object and some with multiple instances for an average size in the range of 3000 to 20,000 bytes. Some special cases could be many times this amount.
From a performance and reliability perspective, should I model the JMS message to pass all the data needed to create the interface message, or model the JMS message to contain record keys for the data and have the stand-alone Java application retrieve the data to create the interface message?
I wouldn't just focus on performance for the decision, but also on other non-functional considerations.
I've been working on a system where we decided to not send the data in the message, but rather the PK of the data in database. Our approach was closer to the command message pattern. Our choice was motivated by the following reasons:
Data size: we would store the data in BLOB because it could bu hughe. In your case, the size of the data probably fit in a message anayway.
Message loss: we planned for the worse. If the messages were lost, we could recover the data and we had a recovery procedure to resubmit the messages. Looks maybe paranoid, but here are two scenario that could lead to some message being lost: (1) queue is purged by mistake (2) an error occurs and messages can't be delivered for a long time. They go to the dead message queue (DMQ) which eventually reaches its limit and start discarding messages, if not configured correctly.
Monitoring: different messages/command could update the same row in database. That was easy to monitor and troubleshoot.
Using a JMS + database did however complicates a bit the design:
distributed transactions: this adds some complexity, and sometimes some problems. Distributed transactions have subtle differences with "regular" transactions, such as distributed timeout.
persitency: the code is less intuitive. Data must first be persisted to have the PK, which leads to some complexity in the code if an ORM is used.
I guess both approaches can work. I've described what led us to not send the data in the message, but your system and requirements might be different, so it might still be easier to send the data in the message in your case. I can not provide a definitive answer, but I hope it helps you make your decision.
Send the data, not the pointer. I wouldn't consider your messages to be an extraordinary size that can't be handled.
It will be no problem for the queue to handle the data, the messages in the queue are persisted anyway (memory, file or database persistence whatever fits better for the size of your queue).
If you just put a handle to the data in the queue the application that process the queue will make unnecessary work to get the data that the sender already has.
Depending on your question I cannot say what's the best in your case. Sure there are performance implications because of the message size and stuff, but first you need to know which information needs to be sent to the remote system by your message consumer, especially in a system which may have concurring updates on the same data.
It is relevant whether you need to keep the information stored in the remote system in sync with the version of the record just stored in your database, and whether you want to propagate a complete history along to the remote system which is updated by the message reciever. As a lot of time may pass in between the message send and the processing on the other end of the queue.
Assume (for some reason) there are a whole lot of messages in the queue, and within a few seconds or minutes three or four update notifications on the same object hit the queue. Assume the first message is processed after the fourth update to the record was finished, and its update notification is put in the queue. When you only pass along the ID of the record, all four messages would perform exactly the same operation on the remote system, which for one is absolutely superfluous. In addition, the remote system sees four updates, all the same,but has no information of the three intermediating states of the object, thus, the history, if relevant, is lost for this system.
Beside these semantic implications, technical reasons for passing the id or the whole data are whether it's cheaper to unwrap the updated information from the message body or to load them from the database. This depends on how you want to serialize/deserialize the contents. The message sizes you provided should be no problem for decent JMS implementation when you want to send the data along.
When serializing java objects into messages you need to hold the class format in sync between sender and consumer, and you have to empty the queue before you can update to a newer version of the class on the consuming site. Of course the same counts for database updates when you just pass along the id.
When you just send the ID to the consumer you will have additional database connections, this might also be relevant depending on the load on the database and how complex the queries are you need to execute to get the objects.