R/Python and H2O Version Mismatches

[hasithjp]

PROBLEM

People often encounter version mismatches between the H2O R/Python client and the H2O back-end.
This prevents you from successfully connecting to the H2O instance.

The error message looks like this from R:

> h2o.init(ip = "xx.xx.xx.xx", port = 54321)
Successfully connected to http://xx.xx.xx.xx:54321/ 

R is connected to the H2O cluster (in client mode): 
    H2O cluster uptime:         2 minutes 35 seconds 
    H2O cluster version:        3.6.0.3 
    H2O cluster name:           mycluster
    H2O cluster total nodes:    1 
    H2O cluster total memory:   25.60 GB 
    H2O cluster total cores:    4 
    H2O cluster allowed cores:  4 
    H2O cluster healthy:        TRUE 

Error in h2o.init(ip = "xx.xx.xx.xx", port = 54321) : 
  Version mismatch! H2O is running version 3.6.0.3 but R package is version 3.6.0.8

In this case, the issue is that the R package for H2O is using 3.6.0.8 but the H2O back-end instance is running software version 3.6.0.3. They must have identical versions.

SOLUTION

First, shut down your H2O cluster using the h2o.shutdown() command (in Python or R), or kill the Java process directly. You will need to kill the H2O instance so that you can install a new version.

You have three options:

1. Upgrade both to the latest version. (Recommended)

2. Upgrade the client package for H2O to the same version the server is running.

3. Upgrade the server version of H2O to the same version the client is running.

To install a new version of H2O, go here to download the latest stable version of H2O. Select the appropriate tab for your use case ("Install in R" vs "Install in Python", etc) and follow the commands to install the latest stable version of H2O.

NOTES

The current stable release of H2O is always available here.

Please note that we cannot move models from version to version so you will have to keep using the version of H2O that you built your model in, or rebuild your model in the newer release.

Also, if you see the version mismatch error (in both R and Python upon running h2o.init()), and you already have an H2O cluster running, you should kill that cluster. You can do this either by using h2o.shutdown(), or by killing the background Java process from your Activity Monitor on Mac/the top command on Linux.

JIRA Issue Migration Info

Jira Issue: TN-7
Assignee: Avni Wadhwa
Reporter: Avni Wadhwa
State: Open