I have installed Nextcloud Hub 26 Winter as an AIO with Nextcloud Office on a new server. The underlying system is Proxmox with an Ubuntu 24.04 VM running Docker. On this setup, Office is really slow — it can take up to 20 seconds to open an empty file. Chrome and Firefox sometimes even report that the tab has crashed; when I choose “Wait,” the file eventually opens after a while.
Collabora is running in a separate container and not as CODE.
The VM has 4 vCPU cores and 8 GB of RAM, and these resources are never fully utilized, even when opening files. The system disk is 250 GB, of which only about 30 GB are used. The server is connected internally with 10 Gbit and then to a 2.5 Gbit network.
My old server, which also runs an Ubuntu 24.04 VM with Docker, has Nextcloud Hub 26 Autumn installed, and everything runs without any problems there. The settings are largely identical.
There are no errors in the logs, and I also cannot find anything else that could be slowing it down.
So my question to you: does anyone have an idea what might be causing this?
I think below are possible reasons of the slowness
This is a pretty common issue when running Collabora on Proxmox, and the fact that your resources aren’t being maxed out actually narrows it down. Here’s what I’d check:
Most likely: Proxmox CPU type
This is the #1 cause of sluggish Collabora on Proxmox. If your VM’s CPU type is set to something generic like kvm64 or x86-64-v2-AES, Collabora won’t have access to modern instruction sets (SSE4.2, AVX/AVX2) and falls back to extremely slow software paths. Go to your VM’s Hardware settings in Proxmox and change the CPU type to host. This passes through your physical CPU’s full feature set to the VM.
Since your old server works fine with an otherwise identical setup, I’d bet that one already has CPU type set to host.
Other things worth checking:
Shared memory (/dev/shm): Docker defaults this to 64 MB, which isn’t enough for Collabora. Make sure your Collabora container has shm_size: "512m" (or higher) set. You can verify inside the container with df -h /dev/shm.
Container resource limits: Even though the VM has headroom, Docker or AIO might be capping the Collabora container itself. Run docker inspect <collabora_container> | grep -i -A5 "resources" and check for CPU/memory limits.
DNS between containers: If the Nextcloud container resolves the Collabora hostname slowly (e.g., IPv6 fallback), every WOPI request adds latency. Try exec-ing into the Nextcloud container and timing a curl to the Collabora URL.
seccomp/AppArmor: Ubuntu 24.04 has fairly strict default profiles. Collabora uses some syscalls (clone3, io_uring) that can get blocked, forcing slow fallbacks. You can test by temporarily running the container with --security-opt seccomp=unconfined — if that fixes it, you can create a tailored profile afterwards.
Start with the CPU type — it’s a one-line change, just needs a VM reboot, and resolves this in the vast majority of cases. Let us know how it goes!
Thank you for your replies. After a lot of back and forth and much tinkering, I have finally found the solution. The issue was with the Reverse Proxy. I had previously been using NPMPlus and decided to test the regular NPM as a trial. It worked immediately, so I then spent some more time adjusting the NPMPlus settings to try to get it working. Unfortunately, I was unable to get it to work with NPMPlus, so I am now simply using NPM instead.
However, I am now facing a new problem, and I am not sure whether it is related to NPM again. Please see the screenshot below — if this turns out to be a new, unrelated issue, I would open a new thread, unless you happen to have a solution on hand.
It appears that the Nextcloud interface is now overlapping the menu bar of Nextcloud Office (Collabora).
