We’ve been trying to set up Linux UBA agents with mixed success, and I think we’re starting to figure out why after a lot of digging and trial and error, but throwing out the question to see if there is some better guidance or a better approach. We followed the limited documentation, have the machines set up, reporting in, they can take compute work and succeed. The issue we’re running into is that they do not seem to write work to the file system, instead using a tmpfs RAM based “file system”. This causes write errors and OOMs when Wine+HordeAgent hit the maximum allowed RAM for that tmpfs mount.
By increasing the allowed size of it we are able to reduce the failures and eventually get to “stability” but this forces us to use machine specs with more RAM than should be necessary when we have machines with plenty of disk space available. I first saw this behavior when trying to bring online some Linux EC2 nodes, and wrote it off as an AWS issue I was struggling to solve, but we recently spun up some Linux hardware in our DC that has 64GB of phys RAM and 14 physical cores. It is running Ubuntu 24.04, and by default that tmpfs mounts with a max allocation of 10% of the physical RAM, so 6GB. At this point we’ve increased that maximum, so it looks like this in df -h
tmpfs 32G 27G 4.8G 85% /run/user/1005
Additionally we had been targeting the Wine binary directly but have since switched to use the bootstrap script that is in the docs, hoping that one of those env var changes was some kind of secret black magic to fix this, but alas, no change.
How do y’all have your Linux UBA nodes configured? Are you using tmpfs? Is there a setting you have configured to force Wine and/or HordeAgent to write to the physical disk? Any guidance appreciated.