I’m really curious what the eMMC problem was. My first guess was that it has an internal capacitor or similar to power the remapping state and ensure that everything is committed to persistent storage during abrupt power failures and this was empty, but I don’t know why plugging it into a different adaptor would have charged it (unless the controller in the other machine cut power if it didn’t appear to be a working device).
You might also want some adjacent things in rc.conf, in particular tmpmfs, which makes /tmp a memory disk. I also learned, looking at these options, that these no longer use md devices (UFS filesystem on an in-memory block device) but instead default to tmpfs and fall back to md if it doesn’t work. This probably changed ages ago, but the last time I looked at these options was probably around 15 years ago. It’s generally better to use tmpfs, since it’s not a real filesystem on a fake device, it’s just memory allocations to hold files and so has less overhead (for example, it doesn’t need to round file chunks up to a block device block size or maintain full inodes), so great to see that is now the default.
I’m really curious what the eMMC problem was. My first guess was that it has an internal capacitor or similar to power the remapping state and ensure that everything is committed to persistent storage during abrupt power failures and this was empty, but I don’t know why plugging it into a different adaptor would have charged it (unless the controller in the other machine cut power if it didn’t appear to be a working device).
You might also want some adjacent things in
rc.conf, in particulartmpmfs, which makes/tmpa memory disk. I also learned, looking at these options, that these no longer use md devices (UFS filesystem on an in-memory block device) but instead default to tmpfs and fall back to md if it doesn’t work. This probably changed ages ago, but the last time I looked at these options was probably around 15 years ago. It’s generally better to use tmpfs, since it’s not a real filesystem on a fake device, it’s just memory allocations to hold files and so has less overhead (for example, it doesn’t need to round file chunks up to a block device block size or maintain full inodes), so great to see that is now the default.