Good point.
It seems impossible to get a reliable estimate of expected write cycles for SD cards, the general consensus seems to be 3-5k cycles. Let's assume the card is half full, so there is 16GB available for routine writing. Even if one assumes only 1,000 cycles that is a total of 16000 GB on a 32GB card. A pi that I have that runs my home automation and other stuff and is on continuously was rebuilt in April, and looking at the journal, which goes right back to that date, it is 3.8GB. I believe that includes all the logging that is done by systemd tasks, so it includes syslog and all the other standard stuff. That sounds about right, being 15GB/day. At that rate I think it would take about 2500 years to get to 16000GB.
I think you really only need to worry about wearing it out if you are routinely writing videos to the card, or other similar intensive tasks.
My experience has been that most cards die following unexpected power down, where I guess something like interrupting it in the middle of a write operation has messed something up.
Well...yes and no.
Generally speaking, a good rule of thumb is between 10k to 100k write cycles depending on quality and class. The thing is that the memory performance can degrade super quick - it's obviously not just like everything runs perfectly from 0 to 99,999 and then the 100kth write fails. So I'd say once you get to the 75-80% lifespan utilisation point (and this is just generalised speculation on my part, it varies wildly) you tend to see degradation affecting data write cycles significantly, which can mean a lot of rewrite cycles which just exacerbates the problem.
But also remember the heat and stress - cold writing 100k cycles is a lot different than hot writing, and in the use case of Node-RED/FlowFuse/middleware, this is where the complaints/notes I made above about class, quality, and production type come into force. SD card endurance varies wildly by NAND type, controller, over-provisioning, and thermal conditions - consumer TLC cards under sustained hot writes are the worst case, while industrial or high-endurance MLC/SLC cards are obviously way better.
Of course I'm probably preaching to the choir here - but I don't want to assume every single person building a NR/FF instance is going to get the nuance instead of just buying an Amazon deal and plugging the card in before they wonder a year later why nothing works.
