redpanda & btrfs: disk space issue

[andvary]

Greetings, everyone

I have an instance of Redpana running atop btrfs (which was a big mistake on my part, as I've come to understand), but at the moment I cannot change the FS I'm using. The problem is, Redpanda files occupy over 60Gb of space (instead of 2-3Gb intended by my retention policies) seemingly because of this defrag issue

My question: is it somehow possible to fine-tune Redpanda to mitigate this problem for the time being, before the btrfs developers fix it or I can migrate to a different FS?

Maybe some of you have experience running Redpanda with btrfs and can share some advice on how to set up both?

Thanks.

[dotnwat]

Is there a summary of how the defrag issue is addressed in other contexts, or direct link to a summary? That might help us map it on to the case of Redpanda. That that linux-btrfs thread is pretty long to read through.

[andvary]

Thanks for your response.

Here Qu states that in his opinion this is a defrag only problem.
As far as I am aware, there are a few patches underway to address it:
btrfs: defrag: add under utilized extent to defrag target list
btrfs: defrag: better lone extent handling
btrfs: defrag: allow fine-tuning on lone extent defrag behavior

As of now, I'm trying the only thing I can: deleted all files in /var/lib/redpanda/data and disabled CoW for this folder. In a few days will report if it had any positive effect.

[andvary]

Apparently, disabling CoW does not solve the issue. At least on my server the "Unreachable" space keeps growing, albeit not too fast (about 2Gb per week).

Would be most grateful for any advice.