Zookeeper

Last updated: Aug 09, 2022

ZooKeeper is a centralized service for maintaining configuration information, naming, providing distributed synchronization, and providing group services.

At PostHog we use it to store metadata information for ClickHouse and Kafka.

Failure modes

Disk space usage increases rapidly

It has been observed that Zookeepers can suddenly increase it's disk usage, after being in a stable state for some time. This can sometimes be resolved by ensuring that old Zookeeper snapshots are cleared. If you experience this issue you can validate this solution by running zkCleanup.sh:

Terminal

kubectl exec -it -n posthog posthog-posthog-zookeeper -- df -h /bitnami/zookeeper
kubectl exec -it -n posthog posthog-posthog-zookeeper -- /opt/bitnami/zookeeper/bin/zkCleanup.sh -n 3
kubectl exec -it -n posthog posthog-posthog-zookeeper -- df -h /bitnami/zookeeper

This will remove all snapshots aside from the last three, printing out the disk usage before and after.

In newer versions of our Helm chart we run snapshot cleanups periodically every hour. If you experience Zookeeper space issues and are on chart 18.2.0 or below, you can update to a later version to enable this. Alternatively you can specify the Helm value zookeeper.autopurge.purgeInterval=1 which will cause the clean up job to run every hour.

If you wish to further debug what is being added to your cluster, you can inspect a snapshot diff by running zhSnapshotComparer.sh e.g.:

Terminal

kubectl exec -it -n posthog posthog-posthog-zookeeper -- /opt/bitnami/zookeeper/bin/zkSnapshotComparer.sh -l /bitnami/zookeeper/data/version-2/snapshot.fe376 -r /bitnami/zookeeper/data/version-2/snapshot.ff8c0 -b 2 -n 1

This will give you a breakdown of the number of nodes in each snapshot, as well as the exact node difference between the two. For example:

Terminal

Deserialized snapshot in snapshot.fe376 in 0.045252 seconds
Processed data tree in 0.038782 seconds
Deserialized snapshot in snapshot.ff8c0 in 0.018605 seconds
Processed data tree in 0.006101 seconds
Node count: 1312
Total size: 115110
Max depth: 10
Count of nodes at depth 0: 1
Count of nodes at depth 1: 2
Count of nodes at depth 2: 5
Count of nodes at depth 3: 4
Count of nodes at depth 4: 12
Count of nodes at depth 5: 262
Count of nodes at depth 6: 546
Count of nodes at depth 7: 317
Count of nodes at depth 8: 162
Count of nodes at depth 9: 1

Node count: 1312
Total size: 115112
Max depth: 10
Count of nodes at depth 0: 1
Count of nodes at depth 1: 2
Count of nodes at depth 2: 5
Count of nodes at depth 3: 4
Count of nodes at depth 4: 12
Count of nodes at depth 5: 262
Count of nodes at depth 6: 546
Count of nodes at depth 7: 317
Count of nodes at depth 8: 162
Count of nodes at depth 9: 1

Printing analysis for nodes difference larger than 2 bytes or node count difference larger than 1.
Analysis for depth 0
Analysis for depth 1
Analysis for depth 2
Analysis for depth 3
Analysis for depth 4
Analysis for depth 5
Analysis for depth 6
Node /clickhouse/tables/0/posthog.events/blocks/202203_10072597193275699042_5516746108958885708 found only in right tree. Descendant size: 20. Descendant count: 0
...

Useful links

Official documentation

Questions?

Was this page useful?

Disaster recovery considerations

It's important to make sure you understand how to recovery from scenarios before they happen. Data considerations Data is stored in PostgreSQL, ClickHouse and Kafka. At a minimum you should have backups of PostgreSQL and ClickHouse, these are crucial to PostHog. Kafka stores the incoming analytics events which are then persisted into ClickHouse. Note: in order to work, ClickHouse and Kafka use Zookeeper as distributed key value store and locking service. While Zookeeper doesn't host any data…