[
https://issues.apache.org/jira/browse/MARVIN-78?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17222353#comment-17222353
]
Lucas Cardoso Silva commented on MARVIN-78:
-------------------------------------------
We can use docker-compose to setup a persistence instead of copying the engine
on image build, right? I can do it if it solves the problem.
> Include steps ID into metadata persisted to external storages
> -------------------------------------------------------------
>
> Key: MARVIN-78
> URL: https://issues.apache.org/jira/browse/MARVIN-78
> Project: Apache Marvin
> Issue Type: Bug
> Reporter: Lucas Bonatto Miguel
> Priority: Major
>
> Currently, Marvin already supports loading the most recent model when running
> the engine local, and having local persistent configured.
> This tend to be an issue on container-based platforms, given containers are
> usually stateless.
> Currently, when configured to use external persistence, like S3, Marvin is
> not able to load the most recent model from S3 if it's being started from a
> container, like Docker.
>
> Steps to reproduce:
> - Create a new engine and configure s3 as persistence
> - Start your engine in a docker
> - Run a request to the /pipeline endpoint
> - Kill your container, and start it again
> - You will see that marvin doesn't load the most recent model from s3
> - You can try just calling `/predictor/reload?protocol=`, in order to have
> it reloading the most recent model (as specified), but that will not work as
> well, because the metadata file is not present at the local MARVIN_DATA folder
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
