Does H2O in clustered formation support a configurable number of concurrent HTTP connections?

merler1

I have followed the H2O documentation to create an H2O cluster in Kubernetes using the Docker h2oai/h2o-open-source-k8s:3.38.0.2 image. When I load test the cluster it can support about 80 concurrent HTTP connections that are opened with this Python code:

import h2o
h2o.connect(url='http://' + "clusterhostname" +':0')

My use case needs an H2O cluster to support 1000+ concurrent connections. When additional connections are opened I start seeing the run time error below which hints at an HTTP connection pool possibly being exhausted. I would be sincerely grateful if H2O experts can help me understand the following:
1. Is an H2O cluster intended to support 1000+ concurrent connections?
2. Is there a default limit on the number of concurrent connections (and what is that limit)?
3. If the limit is configurable then how can the limit be changed?

Here is the error information:

Could not establish link to the H2O cloud http://clusterhostname:0 after 5 retries

[21:41.45] H2OConnectionError: Unexpected HTTP error: HTTPConnectionPool(host='clusterhostname', port=80):
Max retries exceeded with url: /3/Metadata/schemas/CloudV3 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPConnection object at 0x7f94e621c1c0>, 'Connection to clusterhostname timed out. (connect timeout=3.0)'))

parulpandey

Hi @merler1 , have you tried MOJO for scoring as an alternative? https://docs.h2o.ai/driverless-ai/1-10-lts/docs/userguide/python-mojo-pipelines.html#:~:text=The MOJO (Model Objects, Optimized,and R wrappers are provided.