Conversation

Notices

1. Embed this notice
   Wolf480pl (wolf480pl@mstdn.io)'s status on Thursday, 04-Jan-2024 05:13:17 JST Wolf480pl

   TIL you can't resize dm-cache's underlying volume >.<

   • Embed this notice
     feld (feld@bikeshed.party)'s status on Thursday, 04-Jan-2024 05:13:11 JST feld

     in reply to

     • Daniel Taylor
     @wolf480pl @RandomDamage ZFS has RAID-Z expansion now btw but I will keep telling people to stop using anything but mirrors or your performance will suck and any recovery from a disk failure will suck even worse; your data will be damn near inaccessible while it's grinding to resilver. Enjoy your 5 IOPS for 48 hours while it's replacing your dead 14 TB Drive
   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Thursday, 04-Jan-2024 05:13:12 JST Wolf480pl

     in reply to

     • Daniel Taylor

     @RandomDamage ZFS is very limited options of adding disks and reshaping RAID compared to md(4). And I don't think stacking ZFS on md is a good idea...

   • Embed this notice
     Daniel Taylor (randomdamage@infosec.exchange)'s status on Thursday, 04-Jan-2024 05:13:14 JST Daniel Taylor

     in reply to

     @wolf480pl zfs has caching support, AFAIK

     It takes some getting used to, but it is nice

   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Thursday, 04-Jan-2024 05:13:15 JST Wolf480pl

     in reply to

     Maybe using lvmcache wasn't the best idea...

   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Thursday, 04-Jan-2024 05:13:15 JST Wolf480pl

     in reply to

     Can we have overlayfs but copy-up is triggered by reads too, not just writes?

   • Embed this notice
     feld (feld@bikeshed.party)'s status on Thursday, 04-Jan-2024 06:06:29 JST feld

     in reply to

     • Daniel Taylor
     @wolf480pl @RandomDamage hardware RAID5/6 has the write hole issue and also doesn't do any data verification with checksums which allows it to go much faster, yes, and it can go sequentially.
     
     There is sequential resilver for ZFS now but you still have to do a scrub afterwards to confirm data integrity (automatic, but you can cancel it). In my tests it goes much faster, closer to what it was like with hardware RAID, and then at the end you can do a lower priority scrub which won't affect IOPS as much as the resilver does.
     
     Either way, I don't think of mirrors as "wasted space". It's your insurance policy. Plus it makes your storage really fast. With RAIDZ your storage is only as fast as the slowest disk in each VDEV. Hardware RAID5/6 is about the same.
     
     Being able to migrate data to/from my NVMEs at 2000MB/s+ is really nice. Plus that gives me plenty of overhead so when I run backups no workloads are impacted. If I used RAIDZ/5/6 I would be limited to about 600MB/s last I calculated...
   • Embed this notice
     Daniel Taylor (randomdamage@infosec.exchange)'s status on Thursday, 04-Jan-2024 06:06:31 JST Daniel Taylor

     in reply to

     • feld

     @feld @wolf480pl striped and mirrored is The Way no matter what you are using to strap the disks together

     Which does mean adding disks in pairs, which can be annoying for homelab work

   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Thursday, 04-Jan-2024 06:06:31 JST Wolf480pl

     in reply to

     • feld
     • Daniel Taylor

     @RandomDamage @feld
     at previous job we had hardware RAID5 and rebuilds did take a few hours, never more than a day though (because we didn't have any 14TB drives). And the iops during a rebuild were bearable.

     At home I have a RAID1 now, but if I was adding more disks I'd go for RAID 5 or 6, so as not to waste half the space.

   • Embed this notice
     Quad (quad@akko.quad.moe)'s status on Thursday, 04-Jan-2024 23:34:25 JST Quad

     in reply to

     • Quad
     @wolf480pl for that exact reason my homeassistant container has three major versions of python installed simultaneously, since i can't be bothered to clean them up
     
     sometimes it can cause a lot of annoyance to actually update your stuff regularly.
     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Quad (quad@akko.quad.moe)'s status on Thursday, 04-Jan-2024 23:34:26 JST Quad

     in reply to

     • Quad
     @wolf480pl I haven't really had minor updates break my venvs though.
     
     But whenever I update homeassistant it wants THIS EXACT version of python so i have to nuke the whole venv, recreate it with the specific version it wants, then reinstall it in the venv
   • Embed this notice
     Quad (quad@akko.quad.moe)'s status on Thursday, 04-Jan-2024 23:34:27 JST Quad

     in reply to
     @wolf480pl yeah, the thing that breaks most often for me is homeassistant, because it's ultra picky about its python version.
     
     a specific version of home assistant might run on python3.10, but will crash on both 3.9 and 3.11
   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Thursday, 04-Jan-2024 23:34:29 JST Wolf480pl

     in reply to

     • Quad

     @quad maybe if I had a regular routine of "I have a free evening, let's update my NAS for fun".

     But I usually update stuff at the least convenient moment because I have to, and if that breaks something it's another level of yak shaving...

   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Thursday, 04-Jan-2024 23:34:29 JST Wolf480pl

     in reply to

     • Quad

     @quad I guess a part of this dread of updates comes from using python venvs which obviously break every time python minor version changes.

   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Thursday, 04-Jan-2024 23:34:31 JST Wolf480pl

     in reply to

     • Quad

     @quad nothing is bug-free. What I care about is that no new bugs are added after I get things working

   • Embed this notice
     Quad (quad@akko.quad.moe)'s status on Thursday, 04-Jan-2024 23:34:31 JST Quad

     in reply to
     @wolf480pl frankly i don't mind on linux.
     
     i update on my own schedule and and if something breaks i can usually tell at a glance which component it is and the error message is likely decent.
     
     on windows it's torture when something randomly self-destructs, and the piece of software is a proprietary monolith with no decent logs
     
     on linux, being held back by a debian bug for 2 years tends to annoy me more than having to do a fix every 2 years. some of my containers even run arch linux due to that
   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Thursday, 04-Jan-2024 23:34:33 JST Wolf480pl

     in reply to

     • Quad

     @quad yup, it happens more often than I thought

   • Embed this notice
     Quad (quad@akko.quad.moe)'s status on Thursday, 04-Jan-2024 23:34:33 JST Quad

     in reply to
     @wolf480pl i think there was a debian verison that shipped with a broken npm version which was entirely useless. but of course it's debian so you just had to wait 2 years for the next release if you wanted to apt install a working npm
     
     debian's old packages don't really mean stable as in bug/crash free. it just means nothing changes
   • Embed this notice
     Wolf480pl (wolf480pl@mstdn.io)'s status on Thursday, 04-Jan-2024 23:34:35 JST Wolf480pl

     in reply to

     hmm ok apparently this was an lvm2 limitiation that was removed in a later version?

     https://gitlab.com/lvmteam/lvm2/-/commit/5ec24dfb0bb0c9e48d59d6fbb83262a7087964e1

     Of course LVM 2.03.11 that's in Debian 11 is the last one that had the limitation.

   • Embed this notice
     Quad (quad@akko.quad.moe)'s status on Thursday, 04-Jan-2024 23:34:35 JST Quad

     in reply to
     @wolf480pl a classic
     
     Q: when was this bug solved
     
     A: 1 day after debian froze their packages
     Haelwenn /элвэн/ :triskell: likes this.