I have a use case where I need 100% reliability, idempotency (no duplicate messages) as well as order-preservation in my Kafka partitions. I'm trying to set up a proof of concept using the transactional API to achieve this. There is a setting called 'isolation.level' that I'm struggling to understand.
In this article, they talk about the difference between the two options
There are now two new isolation levels in Kafka consumer:
read_committed: Read both kind of messages that are not part of a
transaction and that are, after the transaction is committed.
Read_committed consumer uses end offset of a partition, instead of
client-side buffering. This offset is the first message in the
partition belonging to an open transaction. It is also known as “Last
Stable Offset” (LSO). A read_committed consumer will only read up till
the LSO and filter out any transactional messages which have been
aborted.
read_uncommitted: Read all messages in offset order without
waiting for transactions to be committed. This option is similar to
the current semantics of a Kafka consumer.
The performance implication here is obvious but I'm honestly struggling to read between the lines and understand the functional implications/risk of each choice. It seems like read_committed is 'safer' but I want to understand why.
First, the isolation.level setting only has an impact on the consumer if the topics it's consuming from contains records written using a transactional producer.
If so, if it's set to read_uncommitted, the consumer will simply read everything including aborted transactions. That is the default.
When set to read_committed, the consumer will only be able to read records from committed transactions (in addition to records not part of transactions). It also means that in order to keep ordering, if a transaction is in-flight the consumer will not be able to consume records that are part of that transation. Basically the broker will only allow the consumer to read up to the Last Stable Offset (LSO). When the transation is committed (or aborted), the broker will update the LSO and the consumer will receive the new records.
If you don't tolerate duplicates or records from aborted transactions, then you should use read_committed. As you hinted this creates a small delay in consuming as records are only visible once transactions are committed. The impact mostly depends on the sizes of your transactions, ie how often you commit.
If you are not using transactions in your producer, the isolation level does not matter. If you are, then you must use read_committed if you want the consumers to honor the transactional nature. Here are some additional references:
https://www.confluent.io/blog/transactions-apache-kafka/
https://docs.google.com/document/d/11Jqy_GjUGtdXJK94XGsEIK7CP1SnQGdp2eF0wSw9ra8/edit
if so, if it's set to read_uncommitted, the consumer will simply read everything including aborted transactions. That is the default.
To clarify things a bit for readers: this is the default only in the Java Kafka client. It was done to not change the semantics when transactions were introduced back in the day.
It's the opposite in librdkafka which sets the isolation.level configuration to read_committed by default. As a result, all libraries built on top of librdkafka will consume only committed messages by default: confluent-kafka-python, confluent-kafka-dotnet, rdkafka-ruby.
KafkaJS consumers also uses read committed by default (readUncommitted set to false).
Related
How can we make sure Kafka exactly-once semantics in read-process scenario. read means we are reading from Kafka topic and doing some processing and then we are trying to commit the offset.
Lets suppose, we processed the messages but could not able to commit and before commit the process crashed. after restart, again trying to consume the same message. so how to handle such scenarios? Can this be handled with Kafka Transaction APIs?
There is similar question but not able to understand it properly and left few comments there as well. Just wanted to confirm my understanding.
Confused about Kafka exactly-once semantics
Kafka Transaction offers EOS for consume-process-produce scenarios. This exactly once process works by committing the offsets by producers instead of consumer. i.e., the produce of result to kafka and committing the consumed messages all are done by kafka producer (instead of separate kafka consumer and producer) which brings the exactly once. The EOS in kafka transaction ensures that for each consumed message we have exactly one result (the result may contain multiple messages) on the kafka, but the message could be processed multiple times in failure scenarios.
So you cannot achieve exactly once in read-process. The only solution you can use is to make your messages idempotence and change your business logic somehow that duplicate messages do not have side effect. e.g.:
-Using deduplicate process if you use database and check the duplicate value before insert or process and drop the incoming message.
-In some scenarios that duplicates affect you database, we can commit the offsets to database and by that make the data insertions and offset commits in one transaction.
Kafka documentation states the following as the top scenario:
To process payments and financial transactions in real-time, such as
in stock exchanges, banks, and insurances
Also, regarding the main concepts, right at the very top:
Kafka provides various guarantees such as the ability to process
events exactly-once.
It’s funny the document says:
Many systems claim to provide "exactly once" delivery semantics, but
it is important to read the fine print, most of these claims are
misleading…
It seems obvious that payments/financial transactions must be processed „exactly-once“, but the rest of Kafka documentation doesn't make it obvious how this should be accomplished.
Let’s focus on the producer/publisher side:
If a producer attempts to publish a message and experiences a network
error it cannot be sure if this error happened before or after the
message was committed. This is similar to the semantics of inserting
into a database table with an autogenerated key. … Since 0.11.0.0, the
Kafka producer also supports an idempotent delivery option which
guarantees that resending will not result in duplicate entries in the
log.
KafkaProducer only ensures that it doesn’t incorrectly resubmit messages (resulting in duplicates) itself. Kafka cannot cover the case where client app code crashes (along with KafkaProducer) and it is not sure if it previously invoked send (or commitTransaction in case of transactional producer) which means that application-level retry will result in duplicate processing.
Exactly-once delivery for other destination systems generally
requires cooperation with such systems, but Kafka provides the offset
which makes implementing this feasible (see also Kafka Connect).
The above statement is only partially correct, meaning that while it exposes offsets on the Consumer side, it doesn’t make exactly-once feasible at all on the producer side.
Kafka consume-process-produce loop enables exactly-once processing leveraging sendOffsetsToTransaction, but again cannot cover the case of the possibility of duplicates on the first producer in the chain.
The provided official demo for EOS (Exactly once semantics) only provides an example for consume-process-produce EOS.
Solutions involving DB transaction log readers which read already committed transactions, also cannot be sure if they will produce duplicate messages in case they crash.
There is no support for a distributed transaction (XA) involving a database and the Kafka producer.
Does all of this mean that in order to ensure exactly once processing for payments and financial transactions (Kafka top use case!), we absolutely must perform business-level message deduplication on the consumer side, inspite of the Kafka transport-level “guarantees”/claims?
Note: I’m aware of:
Kafka Idempotent producer
but I would like a clear answer if deduplication is inevitable on the consumer side.
You must deduplicate on consumer side since rebalance on consumer side can really cause processing of events more than once in a consumer group based on fetch size and commit interval parameters.
If a consumer exits without acknowledging back to broker, Kafka will assign those events to another consumer in the group. Example if you are pulling a batch size of 5 events, if consumer dies or goes for a restart after processing first 3(If the external api/db fails OR the worse case your server runs out of memory and crashes), the current consumer dies abruptly without making a commit back/ack to broker. Hence the same batch gets assigned to another consumer from group(rebalance) where it starts supplies the same event batch again which will result in re-processing of same set of records resulting in duplication. A good read here : https://quarkus.io/blog/kafka-commit-strategies/
You can make use of internal state store of Kafka for deduplication. Here there is no offset/partition tracking, its kind of cache(persistent time bound on cluster).
In my case we push correlationId(a unique business identifier in incoming event) into it on successful processing of events, and all new events are checked against this before processing to make sure its not a duplicate event. Enabling state store will create more internal topics in Kafka cluster, just an FYI.
https://kafka.apache.org/10/documentation/streams/developer-guide/processor-api.html#state-stores
I have a use case where I need to send 3 messages to 3 different topics within a single transaction.
The issue is that one of the topics is compacted, and I'm pretty new to Kafka transactions so I'm not really sure how transaction cancelling works.
My question is: what actually happens if the transaction fails or aborts (application crash, exception etc)? Will the records of the aborted transaction eventually be removed from the compacted topic's log (like null records)? Is it same with non-compacted topics?
Thanks.
It's a bit late but according to my experience Kafka keeps messages from aborted transactions in the log but marks them with a special marker so that consumer using read_committed isolation level would filter them out but consumer that uses read_uncommitted would still see them.
This could lead to interesting issues like this.
Based on the fact that aborted messages are filtered at the consumer level I think that they should fall under general compaction logic and would be compacted if there's a message with the same key further down the topic but aborted message shouldn't override last committed message with the same key because that would violate guarantees Kafka provides for transactions.
I am implementing Exactly-once semantics for a simple data pipeline, with Kafka as message broker. I can force Kafka producer to write each produced record exactly once by setting set enable.idempotence=true.
However, on the consumption front I need to guarantee that the consumer reads each record exactly once (I am not interested in storing the consumed records to external system or to another Kafka topic just processing). To achieve this, I have to ensure that polled records are processed and their offsets are committed to __consumer_offsets topic atomically/transactionally (both succeed/fail together).
In such case do I need to resort to Kafka transaction APIs to create a transactional producer in the consumer polling loop, where inside the transaction I perform: (1) processing of the consumed records and (2) committing their offsets, before closing the transaction. Would the normal commitSync/commitAsync serve in such case?
"on the consumption front I need to guarantee that the consumer reads each record exactly once"
The answer from Gopinath explains well how you can achieve exactly-once between a KafkaProducer and KafkaConsumer. These configurations (together with the application of Transaction API in the KafkaProducer) guarantees that all data send by the producer will be stored in Kafka exactly once. However, it does not guarantee that the Consumer is reading the data exactly once. This, of course, depends on your offset management.
Anyway, I understand your question that you want to know how the Consumer itself is processing a consumed message exactly once.
For this you need to manage your offsets on your own in a atomic way. That means, you need build your own "transaction" around
fetching data from Kafka,
processing data, and
storing processed offsets externally.
The methods commitSync and commitAsync will not get you far here as they can only ensure at-most-once or at-least-once processing within the Consumer. In addition, it is beneficial that your processing is idempotent.
There is a nice blog that explains such an implementation making use of the ConsumerRebalanceListener and storing the offsets in your local file system. A full code example is also provided.
"do I need to resort to Kafka transaction APIs to create a transactional producer in the consumer polling loop"
The Transaction API is only available for KafkaProducers and as far as I am aware cannot be used for your offset management.
'Exactly-once' functionality in Kafka can be achieved by a combination of these 3 settings:
isolation.level = read_committed
transactional.id = <unique_id>
processing.guarantee = exactly_once
More information on enabling the exactly-once functionality:
https://www.confluent.io/blog/simplified-robust-exactly-one-semantics-in-kafka-2-5/
https://www.confluent.io/blog/exactly-once-semantics-are-possible-heres-how-apache-kafka-does-it/
Issue we were facing:
In our system we were logging a ticket in database with status NEW and also putting it in the kafka queue for further processing. The processors pick those tickets from kafka queue, do processing and update the status accordingly. We found that some tickets are left in NEW state forever. So we were guessing whether tickets are failing to get produced in the queue or are no getting consumed.
Message loss / duplication scenarios (and some other related points):
So I started to dig exhaustively to know in what all ways we can face message loss and duplication in Kafka. Below I have listed all possible message loss and duplication scenarios that I can find in this post:
How data loss can occur in different approaches to handle all replicas down
Handle by waiting for leader to come online
Messages sent between all replica down and leader comes online are lost.
Handle by electing new broker as a leader once it comes online
If new broker is out of sync from previous leader, all data written between the
time where this broker went down and when it was elected the new leader will be
lost. As additional brokers come back up, they will see that they have committed
messages that do not exist on the new leader and drop those messages.
How data loss can occur when leader goes down, while other replicas may be up
In this case, the Kafka controller will detect the loss of the leader and elect a new leader from the pool of in sync replicas. This may take a few seconds and result in LeaderNotAvailable errors from the client. However, no data loss will occur as long as producers and consumers handle this possibility and retry appropriately.
When a consumer may miss to consume a message
If Kafka is configured to keep messages for a day and a consumer is down for a period of longer than a day, the consumer will lose messages.
Evaluating different approaches to consumer consistency
Message might not be processed when consumer is configured to receive each message at most once
Message might be duplicated / processed twice when consumer is configured to receive each message at least once
No message is processed multiple times or left unprocessed if consumer is configured to receive each message exactly once.
Kafka provides below guarantees as long as you are producing to one partition and consuming from one partition. All guarantees are off if you are reading from the same partition using two consumers or writing to the same partition using two producers.
Kafka makes the following guarantees about data consistency and availability:
Messages sent to a topic partition will be appended to the commit log in the order they are sent,
a single consumer instance will see messages in the order they appear in the log,
a message is ‘committed’ when all in sync replicas have applied it to their log, and
any committed message will not be lost, as long as at least one in sync replica is alive.
Approach I came up with:
After reading several articles, I felt I should do following:
If message is not enqueued, producer should resend
For this producer should listen for acknowledgement for each message sent. If no ackowledement is received, it can retry sending message
Producer should be async with callback:
As explained in last example here
How to avoid duplicates in case of producer retries sending
To avoid duplicates in queue, set enable.idempotence=true in producer configs. This will make producer ensure that exactly one copy of each message is sent. This requires following properties set on producer:
max.in.flight.requests.per.connection<=5
retries>0
acks=all (Obtain ack when all brokers has committed message)
Producer should be transactional
As explained here.
Set transactional id to unique id:
producerProps.put("transactional.id", "prod-1");
Because we've enabled idempotence, Kafka will use this transaction id as part of its algorithm to deduplicate any message this producer sends, ensuring idempotency.
Use transactions semantics: init, begin, commit, close
As explained here:
producer.initTransactions();
try {
producer.beginTransaction();
producer.send(record1);
producer.send(record2);
producer.commitTransaction();
} catch(ProducerFencedException e) {
producer.close();
} catch(KafkaException e) {
producer.abortTransaction();
}
Consumer should be transactional
consumerProps.put("isolation.level", "read_committed");
This ensures that consumer don't read any transactional messages before the transaction completes.
Manually commit offset in consumer
As explained here
Process record and save offsets atomically
Say by atomically saving both record processing output and offsets to any database. For this we need to set auto commit of database connection to false and manually commit after persisting both processing output and offset. This also requires setting enable.auto.commit to false.
Read initial offset (say to read after recovery from cache) from database
Seek consumer to this offset and then read from that position.
Doubts I have:
(Some doubts might be primary and can be resolved by implementing code. But I want words from experienced kafka developer.)
Does the consumer need to read the offset from database only for initial (/ first after consumer recovery) read or for all reads? I feel it needs to read offset from database only on restarts, as explained here
Do we have to opt for manual partitioning? Does this approach works only with auto partitioning off? I have this doubt because this example explains storing offset in MySQL by specifying partitions explicitly.
Do we need both: Producer side kafka transactions and consumer side database transactions (for storing offset and processing records atomically)? I feel for producer idempotence, we need producer to have unique transaction id and for that we need to use kafka transactional api (init, begin, commit). And as a counterpart, consumer also need to set isolation.level to read_committed. However can we ensure no message loss and duplicate processing without using kafka transactions? Or they are absolutely necessary?
Should we persist offset to external db as explained above and here
or send offset to transaction as explained here (also I didnt get what does it exactly mean by sending offset to transaction)
or follow sync async commit combo explained here.
I feel message loss / duplication scenarios 1 and 2 are handled by points 1 to 4 of approach I explained above.
I feel message loss / duplication scenario 3 is handled by point 6 of approach I explained above.
How do we implement different consumer consistency approaches as stated in message loss / duplication scenario 4? Is their any configuration or it needs to be implemented inside custom logic inside consumer?
Message loss / duplication scenario 5 says: "Kafka provides below guarantees as long as you are producing to one partition and consuming from one partition."? Is it something to concern about while building correct system?
Is any consideration unnecessary/redundant in the approach I came up with above? Also did I miss any necessary consideration? Did I miss any message loss / duplication scenarios?
Is their any other standard / recommended / preferable approach to ensure no message loss and duplicate processing than what I have thought above?
Do I have to actually code above approach using kafka APIs? or is there any high level API built atop kafka API which allows to easily ensure no message loss and duplicate processing?
Looking at issue we were facing (as stated at very beginning), we were thinking if we can recover any lost/unprocessed messages from files in which kafka stores messages. However that isnt correct, right?
(Extremely sorry for such an exhaustive post but wanted to write question which will ask all related question at one place allowing to build big picture of how to build system around kafka.)