caplam

rebuild is finished and as of now, it seem to be running ok. when i mentionned slow speed i was thinking of parity rebuid. until now i made data rebuild and parity rebuild on several occasions. Usually data rebuild is fast : nearly the speed of the disk (around 250MB/S) but what isaw is that parity rebuild is always slower. that said it was the first time i had data+parity rebuild at the same time. i will now try to have zabbix monitor the disks but without spinning them up.

rebuild is on going but it's pretty slow. i noticed the last time i did this that parity rebuild is very slow unlike data rebuild which is usually done at full disk speed. right now it started 5 min ago and running at 8MB/S edit: speed up to 120MB/S after 15 min still slow (usually around 250MB/S) but acceptable.

here are the diag just after reboot. parity2 has still 8 reallocated sectors. dionysos-diagnostics-20261101-1615.zip

i asked my ai agent about the problem. it suggests to disable disklocation, dynamix system temp. modify plugin i need to use smartd instead of smartctl (smartd is part of smartmontools and has an option to scan disks without spinning up spun down disks) i will explore this and report here. My goal is to remove the hba reset problem but keep smart alerts. spin down delay is already at 1 hour for array disks. (set globally at 1 hour and default in individual settings) for unassigned i can't find a setting for spin down.

Thanks for your answer. I may have another thing running smart : i have zabbix stack installed. could it also introduce some troubles? Could i deactivate spin down ? i've always thought that a disk would last longer without spinning down and up continuously.

Since a few weeks i have trouble with my array: it started with a parity2 disk disabled. i showed 8 reallocated sectors. I ran a preclear (1 full cycle) with no problem so i made a parity rebuild. then i had a disk3 disabled. I rebuild it with a new disk. now in few hours i had parity 2 and disk2 disabled. What would be your advice ? disk2 doesn't show any reallocated sectors : do i rebuild it on the same disk ? i have a spare disk in the server that is about the same age as the others of array. For now i haven't stopped the array and can't run smart on disabled parity2. so for now i plan to rebuild disk2 with 1 of the 2 unassigned drives. The disks or in a 8 bay sas shelf. i don't know if i should incirminate the hba, the das... the psu of the das has been replaced last year. dionysos-diagnostics-20261101-0940.zip

just here to drop a thanks i was using the script since a long time. Today i noticed it was not working anymore (in fact it was an old version 1.4) and probably for months. So the plugin is very welcome.

@SimonF i think i asked you a few months ago about xe driver. At this time if i remember correctly it was not available because of unraid kernel (before V7). I didn't have time to look at it since. But i have a new server (running in proxmox) with intel gear which has integrated gpu+npu.cpu. I tried to use it for local inference without great success. Anyway i found a repo about a tool to monitor intel gpu with xe driver (also i915) details of compatible hardware are on the repo. I compiled it ti run it in a docker container and it seems to be working great. Maybe you can do something with it. the repo is Topswatch

i am making progress. I use openarc and have loaded a qwen3 model using igpu from my core ultra. I use openwebui as gui. As i begin with ai i can see that i have a ton to learn to make a better usage (how to choose a model, system prompt, skills, agent, knowledge base,....) but it looks promising. There is valuable information on openwebui communities. You should try openarc and explore their github. Another option would be to go with openvino model server but i find their documentation quite confusing, even the name of the different components are confusing

can't answer as i lack some details about your setup but i'd look at network source in the vm settings. For example if you are using bridging for your unraid network interface, you would want to have br0 as your network source in your vm settings.

i plan to buy a B70 too. But for the moment i try to build an ai stack with the igpu and the npu of my core ultra 285HX. I tried several solutions with ipex-llm, ollama-intel. But it's very limited ipex-llm is no more under dev; ollama-intel seems to be compiled with an old version of ollama and is not able to load recent models. This week i found Openarc. Looks very promising but for me it's a steep learning curve. It's based on openvino and models from huggingface. So far the container runs and i've been able to connect openwebui. The drivers are correctly loaded and the available devices are correctly reported (CPU,iGPU,NPU) But i have not been able to download a model. Apparently you need to install huggingface-cli tool to manage models. I'm still looking for better alternatives to run local ai on intel hardware.

I don't expose unraid itself of course. Unraid and other management interfaces are behind a vpn whenever i need access outside of my lan. I expose things like plex, jellyfin, nextcloud, immich,.... and for those when compatible i use authentik for managing user authentication. All those services are in a dmz behind a reverse proxy sitting on my opnsense router. If you decide to host those services for your family members they need to be exposed because if not what is the utility of a nextcloud instance accessible only inside your lan ? Of course i have still work to do enforce better security (for example some of my older containers may not use docker secrets). I have also have to review my docker networks to ensure they are bound to only the interface i want. My post was more about that subject. Things you can't control despite doing things right when you setup an exposed service. Vulnerabilities can be introduced by the image and the way way it's built. And if i understood correctly you can build an image locally (without modifying the dockerfile) and it will be more up to date than the one on the public registry. From what i've read so far you would need to clone the repo of the app and replace the image in your compose file with build. That each time you execute compose up the image is built locally with the most up to date base image and dependency packages. But i need to read more about the subject and also figure out which vulnerabilities are the most dangerous; because some require access to the host others don't,...

you can't have that. Docker returns only running containers because to be attached to a network a container needs to be running.

Unraid is a great way to start with docker containers. It's easy as with some plugins unraid can take care of backup and updates automatically. But some apps require the use of several services with dependencies and unraid alone is not sufficient to manage updates. You can manage that with compose plugin and watchtower and some docker labels When you have many services running it 's not so efficient. To enhance my experience i started to use portainer but the way it manages compose files is a pain (at least for me) I recently discovered dockhand and arcane. They are both apps for managing containers, stacks and images. And you can define where you want your compose files to reside. My preference is for dockhand for the gui but it's a pain when you need to control stacks on another host through the hawser agent: for example you define a stack on the dockhand host and the compose file resides here until it's deployment. Then it's created on the remote host or the existing compose file on the remote host is updated. That makes it impossible to simply run a docker command on the remote host in a container that need an init for example. With arcane it's better handled but imho dockhand has the best gui. Both these 2 apps have vulnerability scanning of images which is great but wtf the number of vuln found in most of the images i run. I have to say that lsio images are those with the fewest vulns. I'm still in the process of understanding the way those vulns impact the security of my homelab but for now it's quite scary as i publish many of the services i use for my family members. So for now i'm in the process of trying to improve my security posture. Until now i've always used prebuilt images. I'm curious to know if many of you build their own images and if so how do you implemented it in your homelab. How do you build new image for an app according to the result of a scan ? What do you have to modify in a compose file to use ownbuilt images instead of prebuilt ? How to manage dependencies ? Can all these steps be automated ? I hope to have some hints to move forward.

you can use vlan in unraid and don't have an address on br0.x interface. You simply have to define ip assignment to none for the sub interface. But i guess in that case containers in that vlan can't use a bridge. I use it to define macvlan network. One thing that came into my mind is that by default docker listen to all host interfaces so if you define a bridge in a vlan interface and you expose a port in a docker on another bridge you have to restrict listening interface when you define the bridge. It's done with the "com.docker.network.bridge.host_binding_ipv4" option in the bridge ipam driver. custom networks bridge have the advantage of simplifying communication between containers. If you have several containers running in the same custom network they can communicate with each others by their name. For example if you have a server that need a database you can define the database host as postgres:5432 You can define firewall rules for your containers see docker docs (iptables).