Page MenuHomePhabricator

Increases the memory available to the corenlp tool container
Closed, ResolvedPublic

Description

The CoreNLP tool [1] is running on the Kubernetes webservice platform and provides a natural language processing service to be used by other tools like platypus-qa and may be used to power research projects.

Because the ML models it uses are large (they take more than 1 GB each), that does not allow to have models for more than one language to be loaded in memory at the same time and so leads to loading times of ~5s when a request in an other language come in.

Is it possible to increase the memory limit of the corenlp container to something like 6GB? It would allow models for English, German, Spanish and French to be hopefully all in memory at the same time.

[1] http://tools.wmflabs.org/corenlp

Event Timeline

bd808 added a subscriber: bd808.

The current webservice backend for Kubernetes does not have a method for modifying the per-container memory cap. I thought we had a bug about this somewhere, but I can't find it. So the first thing we will need to do is fix webservice so that this is possible to resolve.

aborrero assigned this task to Bstorm.
aborrero added a subscriber: aborrero.

We are working on a new kubernetes cluster for Toolforge. This cluster has better user quota management, see T234702: Review and establish configurable quotas for users in the new Kubernetes cluster for reference.
Closing this task now.