:wave: Hi, I have a couple of questions about upgr...
# troubleshooting
d
👋 Hi, I have a couple of questions about upgrading Pinot to the latest version. We've been running Pinot 1.0 in production for over a year, on a small EKS cluster with a single node (t3.xlarge) and EBS storage. Segments are also written to DeepStorage in S3. We kept most of the default Helm chart values, apart from a few tweaks like replica count, volume sizes, Xmx etc. In terms of problems, we only had a single incident with the Zookeeper instance hitting the node limit (we generate a lot of small segments), which we solved by increasing the
jute.maxbuffer
. So it's about time we reviewed and upgraded our setup, including setting up a merge rollup task to solve the small segments situation. Main plan: • Upgrade Pinot to the latest version ◦ Is it best to start a fresh cluster/helm installation and restore segments from S3, or can we just upgrade Helm? Are segments backwards-compatible? • Use 2+ replicas for each component, and at least 2 K8S nodes for HA. ◦ Any recommendations on instance types? I hear that M/T are generally good. ◦ I read somewhere that it's best to run Zookeeper on its own, and not use the "embedded" Helm chart. Is this correct? Any other suggestions/pitfalls? Really appreciate the help.
x
if you already using helm then just check the latest helm version, technically you just need to update the pinot docker image tag
for server, we prefer larger ram instances, so R-series is better but M/T is also ok
note that for zookeeper we suggest to have 3 replicas but make sure you read how to scale zookeeper from 1 -> 3 it’s not as trivial as changing zookeeper statefulset from 1 ->3
d
Thanks for the info 🙌