Migration socketcluster 16.2.1 to 20.0.2

[stephwildcode]

Hello Jonathan,

We have been using SocketCluster Server (worker, broker, and state) version 16.2.1 and SocketCluster Client version 16.1.1 for four years.

Everything was working fine until now, but the worker instance is no longer working in cluster mode.

This is similar to the issue described here: SocketCluster/ag-simple-broker#2

We previously updated the ag-simple-broker version to 5.0.0 two years ago, and it was working.

However, for the past week, the worker instance has stopped working in cluster mode, even with ag-simple-broker version 5.0.0.

We are therefore considering migrating from version 16.2.1 to version 20.0.2.

To do this, we would like to know what significant changes need to be considered for this migration.

Indeed, we can't find any documentation on this.

Thanks for your help.

stephwildcode

[jondubois]

Hi @stephwildcode, nice to hear from you again.

Could the problem have been caused by a change in your underlying network infrastructure? Or an upgrade of one of the SC components or a sub-dependency? Or different traffic patterns?

In any case, it's a good idea to migrate all the different components to the latest version; the rule is to ensure that your scc-worker, scc-broker and scc-state instances have the same major version number; this guarantees compatibility.

In terms of breaking API changes. I don't believe there have been any since v16. Off the top of my head, the most significant change was in v18 when Gulp support was dropped (switched to Rollup.js) for the client and it was bundled as an ECMAScript module instead of building a script tag with global scope. So you should be able to use the import statement without a bundler now (though you are welcome to keep using a bundler too; should still work with Webpack and others).

Most of the major version increases reflect increased stability and major-version updates to dependencies (e.g. ws module and jwt modules).

You can browse different version of the documentation on the website to confirm: https://socketcluster.io/versions/
The current docs are valid from v17+ and apply to v20 too.

But please let me know if you run into any difficulties.

[stephwildcode]

Hi Jonathan,

Thanks for your quick reply.

Here are the actions taken so far:

• Migration from Debian 11 to 13 (Trixie) -> Which Linux version do you recommend? Debian Slim? Alpine?

• Migration from Node.js 16.16 to 18.20.8 -> Which Node.js version do you recommend? 18.15 as in your Docker images?

• Running Node.js in user mode, not root -> This doesn't seem to impact functionality and is a good security practice.

• No changes made to the Socket Cluster version (we stayed at 16.2.1 on the server side and 16.1.1 on the client side).

• New tests performed today with 3 worker pods, 1 state pod, and 3 brokers. All events were received on the channel. Our application behaved correctly.

-> I don't believe that migrating to Debian 13 and Node.js will solve the problem.

So, I contacted the infrastructure team, and a Traefik version upgrade on the Ingressroutes was performed to 2.11.2.
I had doubts about a possible side effect, but today everything seems to be working on our build environments.

Also, we use these Kubernetes versions in our datacenter, but I don't think they're the cause since they were installed 55 days ago, and the problems only appeared a week ago.

kubectl version
Client Version: v1.33.0
Kustomize Version: v5.6.0
Server Version: v1.31.12+vmware.1

In short, we'll continue testing on our build environments. And if everything is OK, we'll deploy it to production.

Best regards,

stephwildcode