Go back up your HA installation right now.
I, like many others, was running HA OS on a raspberry pi with whatever sd card I happened to have at hand at the time when I started to play with the whole thing.
It has since became a part of my household with light automations and other stuff and with everything going on in life it’s been on a back burner to impove automations more, include more data, manage proper backups and so on. And specially the backups are the one I’ve been slacking on and now I have no one else to blame than myself.
Sd card eventually died yesterday evening and today when I started to study why the server went down but still responded to ping I was greeted with a docker error that exec format is wrong (or something along those lines). At first I thought that maybe some upgrade went sideways and decided to grab a image of the card before messing around it so I have something to come back to which failed almost immediately with IO errors on dmesg.
Gladly I could get a backup off from the card from october, so at worst I lost some automation tweaks on HA side but what I’m still afraid is that I might’ve lost Z-wave keys. Ddrescue is running on the card right now so I don’t have the final results yet but I’m not too hopeful that it’ll create a working image.
So, I did what I should’ve done already ages ago and installed HA on my proxmox server and added that to backup cycle so that’s been taken care of properly and now I’m not limited to Rpi3 performance with addons either. I still need to get a new installation for the pi with Z-wave JS due to RaZberry 7 hat but that should be pretty straightforward task, assuming I can retrieve the keys from broken card.
So: Don’t be like me, verify that you have proper backups of your stuff.
The TLC NAND chips used in most commodity memory cards these days are only good for something like 1500 write cycles per cell before they are prone to wearing out and coming back with errors. The difference between a dedicated SSD and a dinky memory card is partially the amount of extra space reserved for relocating data as the memory cells wear out, of which a consumer MicroSD card typically has little (on expensive ones) or none (on cheap ones).
I’ve heard it said, or rather seen it written, that some TLC NAND can endure “up to” as many as 3000 write cycles, but everyone is cagey about the true number and most consumer grade card vendors are tight-lipped about what kind of chips are actually in their products. So in other words, if you’re just scarfing a cheap card off of Amazon or from Microcenter or whatever, don’t expect ironclad longevity.
The one thing with flash storage writes that’ll bite you and it’ll bite you fast is logging. Unix-like systems love to incessantly write little one line additions to oodles and oodles of log files all the time, and if you want to extend your poor overworked little SD card’s lifespan you can dabble in turning some of that stuff off, once you’re positive you don’t need it for troubleshooting.
There also exist high reliability cards sold for industrial embedded applications, which will use lower capacity SLC but be able to endure upwards of 100,000 write cycles (per the marketing literature, at least). Expect capacities to realistically top out at about 64 gigs and for a single unit at that capacity expect to pay north of $100 for the privilege. It may be more appealing to use an NVME SSD at that rate and connect it with USB adapter or a hat.