I am reading how pulsar client discover brokers and it looks like client will consult zookeeper for the broker discovery in the background periodically: https://www.bookstack.cn/read/pulsar-2.5.2-zh/a3656d4de1a283ca.md#9hybm8. could this be a bottleneck considering there are thousands of pulsar clients? I am not quite familiar with zookeeper but the zookeeper cluster does not scale well and only contains 2 or 3 hosts usually
Pulsar 2.5.2 is quite old.
The Pulsar Client asks to any of the Pulsar Brokers about the current owner for a topic.
The Pulsar Broker has a cache over ZooKeeper metadata, so this is usually not a problem.
Usually you can go with a 3 nodes ZooKeeper cluster for most of the Pulsar usecase. Keep an eye on Metrics in order to see if the ZooKeeper nodes are working well or are satu
Related
I have a five node kafka cluster(confluent 5.5 community edition) with 3 zookeeper nodeseach on different aws instances.
While doing failover testing , noticed that the kafka cluster works fine even if all zookeeper nodes are down.
I was able to produce , consume and also create new consumers.
why does the kafka cluster not stop if it cannot connect to any zookeeper nodes ?
What would be the possible issues if we are unaware of such a failure scenario in production and kafka cluster continues to run without zookeeper connectivity ?
how do we handle such a scenario ?
Broker leader election, topic creation, simple ACLs (if you use them) still depend on Zookeeper. For other basic functions relying on the Kafka bootstrap protocols, they might still work, sure. There should definitely be broker logs indicating connection was lost
Ideally you'd have basic process healthchecking and incident management software that you shouldn't miss critical services going down in prod
How to handle? Restart Zookeeper...
While I am creating cluster setup for kafka I came to know zookeeper quorum set up is needed for coordination between kafka brokers.
Are there any other scenarios where we use zookeeper other than only for kafka setup in real time?
This link lists many applications and organisations using ZooKeeper
https://zookeeper.apache.org/doc/r3.6.2/zookeeperUseCases.html
ZooKeeper is used with many Apache projects and is a distributed coordination service used to manage a large set of hosts. In simple terms, Zookeeper allows workers to get on with their jobs and handles all the other complexities i.e. if a leader goes down, alerting the workers, electing a new leader etc.
We have Kafka cluster with the following details
3 kafka machines
3 zookeeper servers
We also have Hadoop cluster that includes datanode machines
And all application are using the zookeeper servers, including the kafka machines
Now
We want to do the following changes
We want to add additional 3 zookeeper servers that will be in a separate cluster
And only kafka machine will use this additional zookeeper servers
Is it possible ?
Editing the ha.zookeeper.quorum in Hadoop configurations to be separate from zookeeper.connect in Kafka configurations, such that you have two individual Zookeeper clusters, can be achieved, yes.
However, I don't think Ambari or Cloudera Manager, for example, allow you to view or configure more than one Zookeeper cluster at a time.
Yes, that's possible. Kafka uses Zookeeper to perform various distributed coordination tasks, such as deciding which Kafka broker is responsible for allocating partition leaders, and storing metadata on topics in the broker.
After closing kafka, the original zookeeper cluster data will be copied to the new cluster using tools, this is a zookeeper cluster data transfer util zkcopy
But if your Kafka cluster didn't stop work, you should think about Zookeeper data transfer to additional zookeeper servers.
I am working on scaling the kafka cluster in Prod. Confluent provides easy way to add kafka brokers. However, how do I know how to scale zookeeper along with Kafka. What should be the ratio? Right now we have 5 zookeeper nodes for 5 kafka brokers. If I have 10 kafka brokers how many zookeeper nodes should be there?
Zookeeper works as a coordination service for Apache Kafka which stores metadata of kafka cluster. Zookeeper cluster is called ensemble.
Number of servers in a zookeeper ensemble are an odd number(3,5 etc).These numbers represents, how much your cluster is fault tolerant.A three node ensemble ,you can run with one node missing.
With five node ensemble,you can run with two nodes missing and your cluster will be available.
You can add as many zookeeper servers based on how much you want system to be functional inspire of failures, however a ZooKeeper cluter of more than 7 nodes is not recommended for issues with overhead of latency and over-communication between those nodes.
If I have, say, 3 partitions with replication factor 3. Now what I understood is that they have all to connect to the same zookeeper. Ok what if they can't due to network issues ? Will the replication continue when the network is avaialble again?
If ZK is down, your Kafka cluster will have limited functionality. For details, see How does Kafka depend on Zookeeper?
Kafka requires Zookeeper (ZK). If ZK is down, then the entire Kafka cluster will be "down" (meaning: will be almost unusable). ZK is used for a bunch of things like managing internal topics etc.
If ZK becomes available to the Kafka cluster, the cluster will be operational.