device delete progress
We need to have a way to determine the progress of a device delete operation. Also for a balance of a RAID-1 that has more than 2 devices it would be good to know how much space is used on each device.
Could btrfs fi df be extended to show information separately for each device?
Signature · 2 lines
My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/
Russell Coker posted on Sun, 21 Sep 2014 12:09:11 +1000 as excerpted:
Quoted 7 lines
We need to have a way to determine the progress of a device delete operation.
Also for a balance of a RAID-1 that has more than 2 devices it would be good to know how much space is used on each device.Could btrfs fi df be extended to show information separately for each device?
btrfs fi show should give you at least some minimal per-device stats, today. Enough to at least have some idea of the progress of a balance when adding/removing devices.
Longer term, there has been discussion of extending/changing the fi df format and making it far more verbose, including the information found in btrfs fi show as well, and making everything potentially per-device. I hadn't paid a whole lot of attention to the details, however.
Alternatively, leave df more or less as it is (perhaps extending it a bit but attempting not to kill existing scripts using it) and put the detail in a new btrfs filesystem usage. This sounds rather more reasonable to me.
Either way the idea is to give people a single command that combines the current output of fi show and fi df, ideally displaying per-device and filesystem totals both, in enough verbosity to avoid the unintuitive and arcane btrfs specific knowledge required today to interpret it.
I had originally presumed that such a change would happen before the experimental labels came off, but it didn't. I don't know the timetable for it now, or even if it's still planned, as IIRC the discussion died away back in the btrfs-progs 3.12 era and I expected to see it in 3.14 and it wasn't there, so I don't know...
Signature · 3 lines
Duncan - List replies preferred. No HTML msgs. "Every nonfree program has a lord, a master -- and if you use the program, he is your master." Richard Stallman
Quoted 7 lines · Duncan
Russell Coker posted on Sun, 21 Sep 2014 12:09:11 +1000 as excerpted:
Quoted 5 lines
We need to have a way to determine the progress of a device delete operation.
Also for a balance of a RAID-1 that has more than 2 devices it would be good to know how much space is used on each device.
Quoted 6 lines
Quoted 2 lines
Could btrfs fi df be extended to show information separately for each device?
btrfs fi show should give you at least some minimal per-device stats, today. Enough to at least have some idea of the progress of a balance when adding/removing devices.
Up to now this is the only way to see the progress.
Quoted 15 lines
Longer term, there has been discussion of extending/changing the fi df format and making it far more verbose, including the information found in btrfs fi show as well, and making everything potentially per-device. I hadn't paid a whole lot of attention to the details, however.
Alternatively, leave df more or less as it is (perhaps extending it a bit but attempting not to kill existing scripts using it) and put the detail in a new btrfs filesystem usage. This sounds rather more reasonable to me.
Either way the idea is to give people a single command that combines the current output of fi show and fi df, ideally displaying per-device and filesystem totals both, in enough verbosity to avoid the unintuitive and arcane btrfs specific knowledge required today to interpret it.
Quoted 5 lines
I had originally presumed that such a change would happen before the experimental labels came off, but it didn't. I don't know the timetable for it now, or even if it's still planned, as IIRC the discussion died away back in the btrfs-progs 3.12 era and I expected to see it in 3.14 and it wasn't there, so I don't know...
The upcomming 3.17 release should contain all the 'new df' patches that were unfortunatelly delayed. There are some minor issues to be fixed but I hope it'll be ready close to the kernel 3.17 release.
David Sterba posted on Mon, 22 Sep 2014 11:49:55 +0200 as excerpted:
Quoted 3 lines
The upcomming 3.17 release should contain all the 'new df' patches that were unfortunatelly delayed. There are some minor issues to be fixed but I hope it'll be ready close to the kernel 3.17 release.
Will that fix my "unknown" chunk profiles too?
I expected either userspace 3.16 or kernel 3.17 to fix that, but I'm running both (well, kernel 3.17 git, since the release hasn't happened yet) and still seeing the unknown chunk profiles.
Either that, or if userspace 3.16 and kernel 3.17 should have fixed it, I'm triggering a bug somewhere.
Signature · 3 lines
Duncan - List replies preferred. No HTML msgs. "Every nonfree program has a lord, a master -- and if you use the program, he is your master." Richard Stallman
Duncan posted on Mon, 22 Sep 2014 16:29:26 +0000 as excerpted:
Quoted 7 lines
David Sterba posted on Mon, 22 Sep 2014 11:49:55 +0200 as excerpted:
Quoted 3 lines
The upcomming 3.17 release should contain all the 'new df' patches that were unfortunatelly delayed. There are some minor issues to be fixed but I hope it'll be ready close to the kernel 3.17 release.
Will that fix my "unknown" chunk profiles too?
Never mind. I see that in the 3.16.1 announcement, which I imagine I'll get next update.
Signature · 3 lines
Duncan - List replies preferred. No HTML msgs. "Every nonfree program has a lord, a master -- and if you use the program, he is your master." Richard Stallman
Quoted 3 lines · Russell Coker
We need to have a way to determine the progress of a device delete operation. Also for a balance of a RAID-1 that has more than 2 devices it would be good to know how much space is used on each device.
btrfs replace does do this; but in the use case where the user wants to permanently shrink a filesystem by only removing a device, then it might be nice to have a status for that too considering it might be an 8TB helium drive and take a while.
Chris Murphy