Workspace best practices to keep low memory consumption
🗒️area-templatesother
Hey everyone, not sure if this help section is the right place to ask this, but couldn't find anything in the search.
Our Coder setup is basically the architecture image example from the docs with a VM for Coder and a VM for postgres, with the Coder being run as a Docker compose container. We are using 2 custom templates built upon the Docker Container and Docker DinD, and also using custom images from our container registry.
We are still using it as a POC and it's going very well in terms of speed and governability, the feedbacks from the users are great. However, as more users join the POC, more memory is used and from time to time, the Coder VM freezes because of the increased system load caused by the swap memory being still on.
I read that swap in container/kubernetes context is bad and I'm thinking of removing it from the VM, that may solve the freeze but at the cost of probably killing workspaces mid execution. I also thought on adding docker memory limit to containers but the scripts people run in the workspaces vary a lot and I want to avoid the experience of being disconnected from the workspace because of OOM caused by bad limit configuration.
Turning off swap and adding limit is still a valid option to avoid everyone being disconnected from their workspaces because the whole VM froze, but I'm trying to find a better "sweet spot" if its possible.
I noticed a bunch of containers that stays with an average of 3Gb of memory in hours that people probably aren't running anything on it, so I started reading about the workspace schedule. For what I understood, if someone runs their code and leave the VS Code open, it will keep bumping the activity hours, so the best way would be to instruct people to close it after finish running their codes and hope they actually do.
So my main questions in this help post are:
- Is there a way to configure activity bump by running process or something that would indicate the workspace is actually being used?
- Is there any other best practice we could setup in the template or the VM to improve this case of memory handling? I may point that I'm no expert in Linux, so really did no extra configuration on the VM.
Our Coder setup is basically the architecture image example from the docs with a VM for Coder and a VM for postgres, with the Coder being run as a Docker compose container. We are using 2 custom templates built upon the Docker Container and Docker DinD, and also using custom images from our container registry.
We are still using it as a POC and it's going very well in terms of speed and governability, the feedbacks from the users are great. However, as more users join the POC, more memory is used and from time to time, the Coder VM freezes because of the increased system load caused by the swap memory being still on.
I read that swap in container/kubernetes context is bad and I'm thinking of removing it from the VM, that may solve the freeze but at the cost of probably killing workspaces mid execution. I also thought on adding docker memory limit to containers but the scripts people run in the workspaces vary a lot and I want to avoid the experience of being disconnected from the workspace because of OOM caused by bad limit configuration.
Turning off swap and adding limit is still a valid option to avoid everyone being disconnected from their workspaces because the whole VM froze, but I'm trying to find a better "sweet spot" if its possible.
I noticed a bunch of containers that stays with an average of 3Gb of memory in hours that people probably aren't running anything on it, so I started reading about the workspace schedule. For what I understood, if someone runs their code and leave the VS Code open, it will keep bumping the activity hours, so the best way would be to instruct people to close it after finish running their codes and hope they actually do.
So my main questions in this help post are:
- Is there a way to configure activity bump by running process or something that would indicate the workspace is actually being used?
- Is there any other best practice we could setup in the template or the VM to improve this case of memory handling? I may point that I'm no expert in Linux, so really did no extra configuration on the VM.
