Running ZFS on consumer SSDs is absolute no go, you need datacenter-rated ones for power loss protection. Price goes brrrrt €€€€€

I too had an idea for a ssd-only pool, but I scaled it back and only use it for VMs / DBs. Everything else is on spinning rust, 2 disks in mirror with regular snapshots and off-site backup.

Now if you don’t care about your data, you can just spin up whatever you want in a 120€ 2TB ssd. And then cry once it starts failing under average load.

Edit: having no power loss protection with ZFS has an enormous (negative) impact on performance and tanks your IOPS.

This, just pgdump properly and test the restore against a different container. Bonus points for spinning as new app instance and checking if it gets along with the restored db.

And remember, friends don’t let friends use latest. Pin the versions in your manifests and version control everything.

I know what you mean. Most people mean well, some are a bit too aggressive, but probably also mean well. I honestly sometimes roll my eyes when I start reading about tailscale, cloudflare tunnels etc. The main thing is not to expose anything you don’t absolutely need to expose.

For access from the outside the most you should need is a random high port forwarded for ssh into a dedicated host (can be a VM / container if you don’t have a spare RaspberryPi). And Wireguard on a host which updates the server package regularly. So probably not on your router, unless the vendor is on top of things.

Regarding ansible and documenting, I totally get your point. Ten years ago I was an absolute Linux noob and my flatmate had to set up an IRC bouncer on my RPi. It ran like that for a few years and I dared not touch anything. Then the SD card died and took down the bouncer, dynDNS and a few other things running on it.

It takes me a lot of time to write and test my ansible playbooks and custom roles, but every now and then I have to move services between hosts. And this is an absolute life saver. Whenever I’m really low on time and need to get something up and running, I write down things in a readme in my infra repository and occasionally I would go through my backlog when I have nothing better to do.

One word of advice. Document the steps you do to deploy things. If your hardware fails or you make a simple mistake, it will cost you weeks of work to recover. This is a bit extreme, but I take my time when setting things up and automate as good as possible using ansible. You don’t have to do this, but the ability to just scrap things and redeploy gives great peace of mind.

And right now you are reluctant to do this because it’s gonna cost you too much time. This should not be the case. I mean, just imagine things going wrong in a year or two and you can’t remember most things you know now. Document your setup and write a few scripts. It’s a good start.

Same. I buy all my domains there. And in case someone needs a proper API and support for the dns challenge, host your DNS at DeSEC.

You don’t need 8 drives when they are 8 times larger than your current ones. I went from planning for 5+ drives to just downsizing to two drives in mirror. Then I can expand with another mirror.

Unless you need uptime and want to guarantee an SLA for your own services, you are much better off with a mirror or raidz1. Do regular backups (off-site, incremental) and don’t fear the disk failure.

Same, I have a bunch of “inbox” folders and drop files into my server or desktop from my phone with 3 clicks.

Just in case you missed this, you can issue valid HTTPS Certificates with the DNS challenge. I use LetsEncrypt, DeSEC and Traefik, but any other supported provider with Lego (CLI) would work.

Symfonium is great, it supports a bunch of sources and works really well. Absolutely worth supporting the dev (check his ko-fi too)!

Oh okay that’s a lot of power. For reference, I just set up an old Haswell PC as a NAS, idling at 25W (can’t get to low Package C states) and usually at 28-30 running light workloads on an SSD pool. My plan was to add a 5 disk cage and at least 3 HDDs, with Raidz2 and 5 disks being the mid term goal. Absolutely unnecessary and a huge waste. I settled on less but larger disks, and in mirror I can get 12-18 TB usable space for under 500€. Less noise and power draw too.

Look for 5W idle consumption boards + CPU combos which go down to package C6+ state. HardwareLuxx has a spreadsheet with various builds focusing on low power. Sell half your disks, go mirror or Raidz1. Invest the difference in off-site vps and or backup. Storage on any SBC is a big pain and you will hit the sata connector / IO limits very soon.

The small NUC form factors are also fine, but if your problem is power you can go very low with a good approach and the right parts. And you’ll make up for any new investments within the first year.

The problem is, the libraries and SDK used to build the app will have had vulnerabilities for sure. Same for the underlying image (unless scratch / distroless). We run extensive vulnerability scanning in our pipelines, and Go libs occasionally pop up. The Go SDK also had multiple security fixes in the last year.

For the physical hosts / bare metal I use fluentbit, with Loki as the backend. Grafana for visualization and alerts. This gives me utilization metrics and uptime monitoring. The app containers themselves I do not monitor.

How does this work, some loophole or a business customer? You can drop some info in a private message if you don’t f feel like posting in public. Re server part deals, I am not sure if this is always the case, but the current selection of disks is 90% helium (Exos etc) HDDs, a few IronWolfs which are too large (20TB) and basically that’s it. My DIY NAS is unfortunately in the apartment and I’m reluctant to try He disks due to the intensive sound profile.

I wish this was an option for Europe. Once you slap VAT and shipping, you end up paying more than for new disks. :(

Wifi pretty much excludes k*s and I assume that swarm and Nomad would be impacted by blips in the wireless connectivity. You can try how things work out with a load balancer / reverse proxy on a wired connection, which then checks the downstream services and routes the request to available instances.

Please look into Wifi-specific issues related to the various orchestration platforms before deciding to try one out. Hypervisor is usually a win win, until you try to do failover.

The best test I have is my wife complaining, that ads in Google results cannot be opened. It seems to work flawlessly for me 😂

On a more serious note, what tests are these? The thing is, the ad domain is either in the blocklist or not. Ads inside apps are hard to block (I even have adaway on my android, and some slip through as eg Instagram reuses the backend domains/endpoints for ad delivery).

I personally stepped away from compose. You mentioned that you want a more declarative setup. Give Ansible a try. It is primarily for config management, but you can easily deploy containerized apps and correlate configs, hosts etc.

I usually write roles for some more specialized setups like my HTTP reverse proxy, the arrs etc. Then I keep everything in my inventory and var files. I’m really happy and I really can tear things down and rebuild quickly. One thing to point out is that the compose module for Ansible is basically unusable. I use the docker container module instead. Works well so far and it keeps my containers running without restarting them unnecessarily.

I’ve been a bit busy so I haven’t had the time to figure out what and how much I need to compensate so the sensor data is more useful. One of the sensors seems to be detecting something reminiscent of a sine curve, so this will involve some extra high school math to find a function to cancel it out. Busy dad etc, maybe next week. In the mean time I started putting together the case and ordered the springy subwoofer legs. Here is how a simple plot of the raw acceleration looks like.

It’s obvious which one is the before and after. The second one even includes two trains arriving back to back.

Now I need to figure out a few things:

1. repeatable experiment (hammer? dropping something heavy from the same height?)
2. make the Z-axis reading more useful and compare velocities
3. add some foam/plywood and rubber feet on the disks