H2O Assisted clustering

Assisted clustering is a special regime of H2O startup, invoked by setting the H2O_ASSISTED_CLUSTERING_REST environment variable to True. In this regime, AssistedClusteringEmbeddedConfigProvider is activated and spawns an HTTP endpoint at http://h2o-node-host:<port:default 8080>/clustering/flatfile, expecting a POST HTTP request with Content-type: text/plain and a standard H2O flatfile in the request body.

Example

curl --location --request POST 'localhost:8080/clustering/flatfile' \
--header 'Content-Type: text/plain' \
--data-raw '255.255.255.0:54321'

Flatfile format

This module itself does not make any assumptions about the flatfile formatting, besides the basic ones like an empty or missing request body. The flatfile is handed over to H2O and parsed by NetworkInit.java - which contains all the details. IPv4 and IPv6 formats are supported (not the shortened representation of IPv6). Both hostname/host ip and port in a format hostname:port must be present.

Example

A flatfile with an IPv6 and IPv4 address with ports defined in both cases (mandatory).

[1200:0000:AB00:1234:0000:2552:7777:1313]:54321
9.255.255.255:54321

Distribution

It is assumed this module is NOT part of H2O by default. It must be put onto the classpath manually. As it only relies on the contract of AbstractEmbeddedH2OConfig, which is considered to be fixed, it is possible to put this module on a classpath of even older H2Os which contain AbstractEmbeddedH2OConfig. This enables assisted clustering even to legacy H2O versions.

Testing

This module, besides its own test suite, is tested in Kubernetes environment inside the h2o-k8s module.

Cluster status endpoint

There is an endpoint for H2O Cluster status provided by this module, located via GET /cluster/status. If H2O has not clustered yet, the endpoint HTTP 204 - No content, according to RFC-2616. Once H2O cluster is formed, this endpoint returns HTTP 200 with a JSON containing two arrays - a list of healthy nodes and a list of unhealthy nodes.

Any node of the cluster can be queried and each will return the same response.

Example curl --location --request GET 'localhost:8080/cluster/status':

{
  "leader_node": "192.168.0.149:54321",
  "healthy_nodes": ["192.168.0.149:54321"],
  "unhealthy_nodes": []
}