Submitted 1 year ago by root@lemmy.world to selfhosted@lemmy.world
I’ve noticed that sometimes when a particular VM/ service is having issues, they all seem to hang. For example, I have a VM hosting my DNS (pihole) and another hosting my media server (jellyfin). If Jellyfin crashes for some reason, my internet in the entire house also goes down because it seems DNS is unable to be reached for a minute or so while the Jellyfin VM recovers.
Is this expected, and is there a way to prevent it?
I’d check high I/O wait, specially if your vms are on HDDs.
one of the solution I had for this issue was to have multiple DNS servers. solved it by buying a raspberry pi zero w and running a 2nd small instance of pihole there. I made sure that the piZeroW is plugged on a separate circuit in my home.
Good point. I just checked and streaming something to my TV causes IO delay to spike to like 70%. I’m also wondering if maybe me routing my Jellyfin (and some other things) through NGINX (also hosted on Proxmox) has something to do with it… Maybe I need to allocate more resources to NGINX(?)
The system running Proxmox has a couple Samsung Evo 980s in it, so I don’t think they would be the issue.
lemme know if you need some tshooting remotely, if schedules permit, we can do screenshares
I had this issue when I used kubernetes, sata SSDs cant keep up, not sure what Evo 980 is and what it is rated for but I would suggest shutting down all container IO and do a benchmark using fio.
my current setup is using proxmox, rusts configured in raid5 on a NAS, jellyfin container.
all jf container transcoding and cache is dumped on a wd750 nvme, while all media are store on the NAS (max. BW is 150MBps)
you can monitor the IO using IOstat once you’ve done a benchmark.
Why are you running full VMs for something that can be put in a container? Sounds to me (without having any evidence or proof) that you’re running out of memory and you’re swapping and it’s taking forever. That’s what causes the VMs to slow/stop.
I typically prefer VM’s just because I can change the kernel as I please (containers such as LXC will use the host kernel). I know it’s overkill, but I have the storage/ memory to spare. Typically I’m at about 80% (memory) utilization under full load.
If the VM crashing is because of high CPU usage on all cores, high IO delay on the storage, or an out of memory situation on the host, that would cause all of the other VMs to struggle as well.
How many cores do you have configured for jellyfin?
4 currently with 8GB RAM and no pass through for transcoding (only direct play)
This happens to me when there is an app keeping a file opened on NFS storage mapping
Acronyms, initialisms, abbreviations, contractions, and other phrases which expand to something larger, that I’ve seen in this thread:
| Fewer Letters | More Letters |
|---|---|
| DNS | Domain Name Service/System |
| LXC | Linux Containers |
| NAS | Network-Attached Storage |
| SSD | Solid State Drive mass storage |
[Thread #205 for this sub, first seen 10th Oct 2023, 06:25] [FAQ] [Full list] [Contact] [Source code]
Good bot.
NeoNachtwaechter@lemmy.world 1 year ago
Your services may run in separate VMs, but there are still some dependencies between them. You need to know, and think about, all the dependencies between your VMs.
For example, they share a common network interface (the one of the host machine). That is a dependency. If one VM is able to clog the network interface (and maybe your crashing one is doing exactly that), then it is clogged for all the other VMs too.
To resolve that dependency, you can either put another network interface card in your host machine and let only the pihole VM use it, or run the pihole on a real physical Pi.
You could also resolve the jellyfin’s own problem. But resolving the dependency might give you a more reliable system.
root@lemmy.world 1 year ago
That’s a good point; My Virtualization server is running on a (fairly beefy) Intel NUC, and it has 2 eth ports on it. One is for management, and the other I plug my VLAN trunk into, which is where all the traffic is going through. I will limit the connection speed of the client that is pulling large video files in hopes the line does not saturate, and long term I’ll try to get a different box where I can separate the VLAN’s onto their own ports instead of gloming them all into one port.