Some directories are separate from --user-data-dir

[code-asher]

Steps to Reproduce

Set --user-data-dir.

Expected

Everything should move to that directory.

Actual

Some things move, but not:

• heartbeat
• coder-logs
• Generated certificates

Workaround

You can set xdg data home environment variable instead of using the flag.

Notes

Changing this could be considered be a breaking change? If someone was setting --user-data-dir but then relying on these files being in the other location. Not 100% sure how best to proceed. Maybe check if they exist, and if not use the new location, but even that could be breaking.

Maybe introduce a new flag and deprecate --user-data-dir.

[bdkuhman]

Might be related, or a separate issue.

I'm not setting --user-data-dir but seeing ~/date-time-01-/home/coder/.local/share/code-server/coder-logs/ directories fill up in ~.

the logs look like it's correct though, and that file/dir actually exists.
[CHANGED] /home/coder/.local/share/code-server/coder-logs/code-server-stdout.log

My thought is that these directories should exist as /home/coder/.local/share/code-server/coder-logs/date-time-01-/code-server-stdout.log

[code-asher]

Oh weird yeah I got this issue once but then I could never reproduce it. It is especially weird because we are not adding any date time to the file path so I have no idea where it is coming from. Are you on the latest version of code-server?

Do you have any XDG* environment variables set? And just for sanity's sake, is the HOME env var the right value?

[bdkuhman]

Yep, latest.
Running in kubernetes w/ a deployment yaml.
Only env var I set myself is PASSWORD.
no XDG vars, HOME is /home/coder

I am running with args --log trace, along with the tls args.
I suppose trace logs might be it, but I haven't tried it. this was just left over from debugging other things.

[code-asher]

I was going to experiment with podman run -p 8080:8080 codercom/code-server --log trace and see if I could get it to reproduce, but seems Docker Hub is down right now. 🔥 I will try later!

[bdkuhman]

I noticed that all the logs that are left over are 10M,
I think it might be hitting the limit and getting rotated out by rotating-file-stream

taking a quick look at the docs, I think the first param for createStream is a filename, not a path.
Additionally there's a path option that:

If present, it is prepended to generated file names as well as for history file.

So I think with trace on, logs have just been getting big enough to be rotated and when they're rotated rfs is prepending the time/date to the whole path, rather than the filename since that's what was passed in.

[code-asher]

Ohhh! Really good find!!! I reproduced by lowering the limit.

Pushed up a fix here: d431c9c

Thank you for figuring this out. It will make it into the 4.90.3 release today.