Garrett Wollman
@wollman@mastodon.social
I fix networks and other stuff with computers inside for MIT CSAIL. Bi cyclist, cis male. More frequency = better mobility. Formerly @garrett_wollman on Twitter, still @gwollman.bsky.social on the other site.
Posts
Latest notes
@dexter If you `zfs set readonly=on mydataset` or likewise with `atime=off`, it doesn't change any of the child datasets. For example:
# zfs get atime rootvg/root
NAME PROPERTY VALUE SOURCE
rootvg/root atime on default
# zfs set atime=off rootvg
# zfs get atime rootvg/root
NAME PROPERTY VALUE SOURCE
rootvg/root atime on temporary
Pre-#OpenZFS on #FreeBSD the right thing happened and atime=off was inherited down the tree.
ZFS fun at work today. `zfs rollback` is very very slow. What turned out to be 1000x faster was:
* delete all intermediate snapshots
* zfs rename badfs badfs-bad
* zfs clone badfs-bad@good-snapshot badfs
* zfs promote badfs
* zfs destroy -r badfs-bad
My guess is that `zfs rollback` is synchronous and we were stuck waiting for the blocks to be freed before it would return. (I chose an intermediate snapshot as the first rollback target just in case, and it took about 45 minutes.)
This may just be old-mean-yells-at-cloud territory here but I really do think that code written 35 years ago has a better chance of compiling and running and actually working than code written 15 or 20 years ago. The code from 35 years ago probably just has one or two dependencies. The code from 15 years ago has 50 dependencies, two or three of which just outright no longer exist and another ten still "exist" but were abandoned without replacement.