Re: More than 40 minutes to delete an obj tree?
- Reply: Michael Grimm : "Re: More than 40 minutes to delete an obj tree?"
- In reply to: Bjoern A. Zeeb: "More than 40 minutes to delete an obj tree?"
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Sat, 17 Jan 2026 11:54:23 UTC
> On 17. Jan 2026, at 11:40, Bjoern A. Zeeb <bzeeb-lists@lists.zabbadoz.net> wrote: > > Hi, > > I did a universe built yesterday of stable/14. > > That accumulates 83G of data in the obj tree (I wanted to know how > much space we need) > > % du -hcsx * > 27G amd64.amd64 > 6.9G arm.armv7 > 13G arm64.aarch64 > 8.2G i386.i386 > 5.9G powerpc.powerpc > 7.7G powerpc.powerpc64 > 5.8G powerpc.powerpc64le > 8.0G riscv.riscv64 > 83G total > > I didn't run "time" on the rm -rf of that obj tree but I caught it > seconds before it was done: > > load: 0.56 cmd: rm 91967 [zio->io_cv] 2541.72r 1.52u 81.62s 4% 3972k > > Tha machine in questing is a 14.3-STABLE GENERIC and while it is shared > it's been pretty much idle during that time. > > Is it normal that it takes so much real time to cleanup a universe build > these days or is that a ZFS tuning issue? It feels a bit excessive. > > /bz > Different factors may be involved and both [d]tracing and patch level analysis are needed to understand it. For workaround, zfs destroy should be a bit faster than rm -rf (unless there is deduplication involved….). rgds, toomas