Support Questions

Find answers, ask questions, and share your expertise
Announcements
Celebrating as our community reaches 100,000 members! Thank you!

Best practices for zookeeper placement?

avatar

What other services are best to colocate on a host with Zookeeper, and how does this change as number of hosts increases?

Does it make sense not to run it on a host with HA services, since those are what it protects? If running on a NodeManager, what adjustments should be made to memory available for YARN containers?

1 ACCEPTED SOLUTION

avatar
Master Mentor

@Alex Miller

Generally, Its ok to deploy ZK with other components (dedicated server not required). As you know , odd number of zk is the best practice and I see no issues in deploying in HA node but I would deploy in non HA node too to keep the balance.

Link

Here are some common problems you can avoid by configuring ZooKeeper correctly:

Gotchas: Common Problems and Troubleshooting

  1. If you are using watches, you must look for the connected watch event. When a ZooKeeper client disconnects from a server, you will not receive notification of changes until reconnected. If you are watching for a znode to come into existance, you will miss the event if the znode is created and deleted while you are disconnected.
  2. You must test ZooKeeper server failures. The ZooKeeper service can survive failures as long as a majority of servers are active. The question to ask is: can your application handle it? In the real world a client's connection to ZooKeeper can break. (ZooKeeper server failures and network partitions are common reasons for connection loss.) The ZooKeeper client library takes care of recovering your connection and letting you know what happened, but you must make sure that you recover your state and any outstanding requests that failed. Find out if you got it right in the test lab, not in production - test with a ZooKeeper service made up of a several of servers and subject them to reboots.
  3. The list of ZooKeeper servers used by the client must match the list of ZooKeeper servers that each ZooKeeper server has. Things can work, although not optimally, if the client list is a subset of the real list of ZooKeeper servers, but not if the client lists ZooKeeper servers not in the ZooKeeper cluster.
  4. Be careful where you put that transaction log. The most performance-critical part of ZooKeeper is the transaction log. ZooKeeper must sync transactions to media before it returns a response. A dedicated transaction log device is key to consistent good performance. Putting the log on a busy device will adversely effect performance. If you only have one storage device, put trace files on NFS and increase the snapshotCount; it doesn't eliminate the problem, but it can mitigate it.
  5. Set your Java max heap size correctly. It is very important to avoid swapping. Going to disk unnecessarily will almost certainly degrade your performance unacceptably. Remember, in ZooKeeper, everything is ordered, so if one request hits the disk, all other queued requests hit the disk.

    To avoid swapping, try to set the heapsize to the amount of physical memory you have, minus the amount needed by the OS and cache. The best way to determine an optimal heap size for your configurations is to run load tests. If for some reason you can't, be conservative in your estimates and choose a number well below the limit that would cause your machine to swap. For example, on a 4G machine, a 3G heap is a conservative estimate to start with.

View solution in original post

2 REPLIES 2

avatar
Master Mentor

@Alex Miller

Generally, Its ok to deploy ZK with other components (dedicated server not required). As you know , odd number of zk is the best practice and I see no issues in deploying in HA node but I would deploy in non HA node too to keep the balance.

Link

Here are some common problems you can avoid by configuring ZooKeeper correctly:

Gotchas: Common Problems and Troubleshooting

  1. If you are using watches, you must look for the connected watch event. When a ZooKeeper client disconnects from a server, you will not receive notification of changes until reconnected. If you are watching for a znode to come into existance, you will miss the event if the znode is created and deleted while you are disconnected.
  2. You must test ZooKeeper server failures. The ZooKeeper service can survive failures as long as a majority of servers are active. The question to ask is: can your application handle it? In the real world a client's connection to ZooKeeper can break. (ZooKeeper server failures and network partitions are common reasons for connection loss.) The ZooKeeper client library takes care of recovering your connection and letting you know what happened, but you must make sure that you recover your state and any outstanding requests that failed. Find out if you got it right in the test lab, not in production - test with a ZooKeeper service made up of a several of servers and subject them to reboots.
  3. The list of ZooKeeper servers used by the client must match the list of ZooKeeper servers that each ZooKeeper server has. Things can work, although not optimally, if the client list is a subset of the real list of ZooKeeper servers, but not if the client lists ZooKeeper servers not in the ZooKeeper cluster.
  4. Be careful where you put that transaction log. The most performance-critical part of ZooKeeper is the transaction log. ZooKeeper must sync transactions to media before it returns a response. A dedicated transaction log device is key to consistent good performance. Putting the log on a busy device will adversely effect performance. If you only have one storage device, put trace files on NFS and increase the snapshotCount; it doesn't eliminate the problem, but it can mitigate it.
  5. Set your Java max heap size correctly. It is very important to avoid swapping. Going to disk unnecessarily will almost certainly degrade your performance unacceptably. Remember, in ZooKeeper, everything is ordered, so if one request hits the disk, all other queued requests hit the disk.

    To avoid swapping, try to set the heapsize to the amount of physical memory you have, minus the amount needed by the OS and cache. The best way to determine an optimal heap size for your configurations is to run load tests. If for some reason you can't, be conservative in your estimates and choose a number well below the limit that would cause your machine to swap. For example, on a 4G machine, a 3G heap is a conservative estimate to start with.

avatar

Excellent tips, thank you.

Is there a guideline for when to add another pair of ZK servers? Cluster size, number of services that use ZK, any services that are particularly demanding, etc?