[OpenVZ] Ploop 1.12.2
- Thread starter Geek
- Start date
Any customer that comes to us asking for huge inode limits we offer to put them on ploop with customized block sizing. We've only done this a few times and thankfully they've been fine for now.
We have a few other users, though, that have had a rough time and we, nor they, can figure out why. Their FS has gone read-only for no good reason. We've migrated them to development nodes thinking maybe the underlying hardware needed a FSCK and it still happened. The VPS was only ever being powered off to do force FSCK's.
In the end we simply offered to extract the users data and put them on the old setup. They agreed, I handled it in all of 5 minutes, and they've not had a single issue in the last few months.
I really do like the idea of it, I'm just curious why they insisted on using their own format, instead of just wrapping QCOW2. We've got some QCOW in use and it's great. Performance is great, the tools work as they should, etc. The biggest thing ploop has over QCOW is the shrinking option, but given @Geek's own experiences, it doesn't work very well anyway.
Francisco
Yep, I remain of similar opinion. I had a little time to fart around with updated resize on the new kernel and, well, I'm now -13gb wasted space on that device. And that's just after a ploop-baloon discard. Don't get me started on the snapshots.
I briefly flirted with the idea of Ploop on ZFS but without the quota support, it's just not ideal for what I provide.
@serverian, did you have any luck with your ext4 corruption from earlier this morning? Some of our Russian friends are surgical with Ploop... I mean they just tore into it. Pavel Odintsov has a tool to mount ploops in userspace. Googleable by name. Might come in handy with your recovery. GL.
P.S. -- where did your thread go? I only can only find it on a tab I didn't refresh this morning.
Last edited by a moderator:
ZFS + linux is extremely RAM whorish for a lack of a nicer term. The ARC isn't integrated into linux's base memory management which means when ZFS decides it's going to blow up its RAM usage, it's going to mark the RAM as actually used, even if all it's doing is caching.
While there are some locks for this all (zrc_arc_max & zfs_arc_meta_limit), on linux it doesn't really listen to them. It does for a while but if something starts pushing (rsyncing a large inode folder for instance), it isn't going to age out via LRU, it's just going to push past it.
They have plans to fix this, and did merge some patches to make ARC not thrash like a spoiled child, but it's still not going to handle large work loads very well.
Francisco
