Maintenance on Btrfs

[GreyGeek]

I've been a dufus! A real

Working out all these plans for making backup snapshots on my 2nd HD and a USB HD.
Then it hit me .... both of my HDs are identical. Why not convert the pair to a raid 1 setup and continue to use the USB HD as the off-site archival location for "btrfs send -f ..." backups.
Back to the manuals ....

EDIT: And then I run across this, which was followed by a howto in the same thread.

[GreyGeek]

Well, I'm getting ready to upgrade my btrfs from a single to a raid1.
First, I am going to plug in my USB HD and send&receive my @ and @home snapshots to it.
Then, I will delete on the snapshots, directories and files on the 2nd HD, which also uses btrfs, and then I will repartition it to create sdb1 and mkfs.btrfs sdb1.
After that I will turn swap off so that the 10Gb sda5 partition wasted by it can be deleted, and then use the "btrfs fi resize max /" to give sda1 the same size as sdb1.
Next I'll delete on the snapshots on my rootfs and also the snapshots directory, then run balance on sda1.
Then I'll run "btrfs device add /dev/sdb1 /", followed by using balance again but this time with the -dconvert=raid1 -mconvert=raid1 -sconvert=raid1 parameters.

My drives are 750Gb WD's. As a single "btrfs fi usage /" shows 90Gb used and about 598Gb free. With 2 disks running as a raid1 those numbers will stay at about the same values. I expect some speed penalty for making two copies of everything, even with COW because changes will be written twice. My archival copies will be on my 320gb WD Passport drive, which will be my weak link because WD's USB HD's are notoriously unreliable. That means that a 256Gb USB SSD is in my future because ro snapshots of @ and @home will, combined, take about 100Gb.

If things work out I'll sign on later today. If not, I'm going fishing.

[GreyGeek]

The 2nd HD was added to the pool and the balance started

[FONT=monospace][B][COLOR=#000000]btrfs device add -f /dev/sdb /  [/COLOR][/B][COLOR=#000000](the -f was used to force the add because sdb was already formated with btrfs)[/COLOR]

[/FONT]
[FONT=monospace][FONT=courier new][B][COLOR=#000000]btrfs balance start -dconvert=raid1 -mconvert=raid1 -sconvert=raid1 /[/COLOR][/B][/FONT]
(I was going to add -sconvert=raid1 but was given a warning so I decided to do more research on that aspect)
(The convert took 33 minutes.)

[/FONT][FONT=monospace][COLOR=#000000]:~# [B]btrfs fi show /[/B][/COLOR]
Label: none  uuid: 12980ae8-4117-4cc5-bbb8-8065e82af93d
        Total devices 2 FS bytes used 90.33GiB
        devid    1 size 691.19GiB used 97.03GiB path /dev/sda1
        devid    2 size 698.64GiB used 97.03GiB path /dev/sdb

[/FONT][FONT=monospace][COLOR=#000000]:~# [B]btrfs fi usage /[/B][/COLOR]
Overall:
    Device size:                   1.36TiB
    Device allocated:            194.06GiB
    Device unallocated:            1.17TiB
    Device missing:                  0.00B
    Used:                        180.67GiB
    Free (estimated):            603.23GiB      (min: 603.23GiB)
    Data ratio:                       2.00
    Metadata ratio:                   2.00
    Global reserve:              146.84MiB      (used: 0.00B)

Data,RAID1: Size:95.00GiB, Used:89.65GiB
   /dev/sda1      95.00GiB
   /dev/sdb       95.00GiB

Metadata,RAID1: Size:2.00GiB, Used:704.03MiB
   /dev/sda1       2.00GiB
   /dev/sdb        2.00GiB

System,RAID1: Size:32.00MiB, Used:16.00KiB
   /dev/sda1      32.00MiB
   /dev/sdb       32.00MiB

Unallocated:
   /dev/sda1     594.15GiB
   /dev/sdb      601.61GiB

[/FONT]

RAID1 on data, metadata and systems.
AND, I kept using the system live while all that was going on. (I would have had to boot into my btrfs LiveUSB in order to expand the sda1 to absorb the freed up 7.45Gb swap sda5 partition, and I decided against that. I'll think of something else to use sda5 for.)

[oshunluvr]

I've never seen a reference to -sconvert, only -d and -m. My belief is (undocumented) that the system data takes care of itself.

@here;
a RAID1 pool shows "System, RAID1"
a RAID0 pool shows "System, RAID1"
a single device shows "System, single"

[GreyGeek]

I've never seen a reference to -sconvert, only -d and -m. My belief is (undocumented) that the system data takes care of itself.

@here;
a RAID1 pool shows "System, RAID1"
a RAID0 pool shows "System, RAID1"
a single device shows "System, single"

As you would suspect, my single drive btrfs installation had data, metadata and system as "single" or DUP.

I didn't use the -sconvert=raid1 because it wanted me to force the conversion with "--force". As it turned out, the "System" became RAID1 as well without using -s...".

I just did my first ro snapshot of @, which took a fraction of a second, and then the send & receive to my usb HD mounted at /backup, and it took 10 minutes. I am now doing the send & receive of the @home backup and I suspect that it will take about 30 minutes.


Interestingly, on my sdb drive, before I added it to the pool, the Data was single and the Metadata and System were DUP. When my S&R's are done I'll see what the data, metadata and system settings are on my USB HD.

[GreyGeek]

I finished the backups to my USB 320Gb HD mounted at /dev/sdc1 as /backup.

[FONT=monospace][FONT=courier new][B][COLOR=#000000]btrfs fi usage /backup[/COLOR][/B][/FONT]
Overall:
    Device size:                 297.43GiB
    Device allocated:            179.02GiB
    Device unallocated:          118.41GiB
    Device missing:                  0.00B
    Used:                        176.34GiB
    Free (estimated):            118.51GiB      (min: 59.30GiB)
    Data ratio:                       1.00
    Metadata ratio:                   2.00
    Global reserve:              267.95MiB      (used: 0.00B)

Data,single: Size:174.00GiB, Used:173.90GiB
   /dev/sdc1     174.00GiB

Metadata,DUP: Size:2.50GiB, Used:1.22GiB
   /dev/sdc1       5.00GiB

System,DUP: Size:8.00MiB, Used:48.00KiB
   /dev/sdc1      16.00MiB

Unallocated:
   /dev/sdc1     118.41GiB


:/# [FONT=courier new][B]btrfs fi du -s /backup[/B][/FONT]
     Total   Exclusive  Set shared  Filename
 173.86GiB   173.86GiB       0.00B  /backup
[/FONT]

[dibl]

Cool -- I didn't even know I could do that!

root@Hibiscus:/# btrfs fi usage /mnt/DATA
Overall:
   Device size:                   1.82TiB
   Device allocated:            772.04GiB
   Device unallocated:            1.06TiB
   Device missing:                  0.00B
   Used:                        742.66GiB
   Free (estimated):              1.09TiB      (min: 570.64GiB)
   Data ratio:                       1.00
   Metadata ratio:                   2.00
   Global reserve:              512.00MiB      (used: 0.00B)

Data,single: Size:8.00MiB, Used:7.85MiB
  /dev/sdc        8.00MiB

Data,RAID0: Size:766.00GiB, Used:740.85GiB
  /dev/sdc      383.00GiB
  /dev/sdd      383.00GiB

Metadata,single: Size:8.00MiB, Used:0.00B
  /dev/sdc        8.00MiB

Metadata,RAID1: Size:3.00GiB, Used:924.41MiB
  /dev/sdc        3.00GiB
  /dev/sdd        3.00GiB

System,single: Size:4.00MiB, Used:0.00B
  /dev/sdc        4.00MiB

System,RAID1: Size:8.00MiB, Used:80.00KiB
  /dev/sdc        8.00MiB
  /dev/sdd        8.00MiB

Unallocated:
  /dev/sdc      545.49GiB
  /dev/sdd      545.50GiB

[GreyGeek]

It is!
Oshunluver found a neat little command in the 4.9 version of btrfs-progs:

[COLOR=#000000]:[/COLOR][COLOR=#5454FF][B]~[/B][/COLOR][COLOR=#000000]$ [B]sudo btrfs fi du -s /  [/B][/COLOR]
     Total   Exclusive  Set shared  Filename
  87.93GiB     1.01GiB    86.91GiB  /


[That was on my sda1 when it was a single.  After I converted to RAID1 using two identical HDs I got:[/FONT]


[B] sudo btrfs fi du -s /[/B]
 
     Total   Exclusive  Set shared  Filename
  85.44GiB   653.07MiB    84.80GiB  /

The btrfs-progs package is found here and it updates btrfs-tools (which is now a dummy transitional package. The 4.9 version has the "du" switch in it, the 4.4 version of btrfs-tools in the repository does not. I used btrfs-progs to convert my system to raid1.

[dibl]

The btrfs-progs package is in Debian sid:

root@hp-350g:/# apt policy btrfs-progs
btrfs-progs:
Installed: 4.9.1-1
Candidate: 4.9.1-1
Version table:
*** 4.9.1-1 500
      500 http://httpredir.debian.org/debian unstable/main amd64 Packages
      100 /var/lib/dpkg/status

[oshunluvr]

Interestingly, on my sdb drive, before I added it to the pool, the Data was single and the Metadata and System were DUP. When my S&R's are done I'll see what the data, metadata and system settings are on my USB HD.

I think, for a file system, btrfs is pretty smart. When you have a single drive, by default it duplicates the metadata and system areas. If a sector containing a file were to get trashed or die, you'd lose a file. If a metadata sector fails, you might loose a whole lot more. When you add a second drive, the second copies of meta and sys are moved to the second device. Automatic redundancy.

[GreyGeek]

I did my first reboot this morning after converting my single to a RAID1 using two internal HDs.
Here are the results:

:~$ systemd-analyze
Startup finished in 8.868s (kernel) + 3min 2.314s (userspace) = 3min 11.182s

You read that right. Six minutes + to boot to the desktop. BUT, that's not all. I opened a Konsole to run systemd-analyse:

:~$ systemd-analyze blame
Bootup is not yet finished. Please try again later.

THAT is an error msg that I have never seen before in 19 years of using Linux!

"Systemd-anlyze blame" produced normal results.

I consulted the system logs and noted:

4/27/17 7:51 AM systemd Timed out waiting for device dev-disk-by\x2duuid-5702affa\x2d252a\x2d4020\x2d9f46\x2d59461618b8aa.d evice.

A quick check of /etc/fstab revealed that although I had turned swap off and deleted its sda5 partition, I FORGOT to comment out the section in /etc/fstab that loads the swap disk.

Doing that and rebooting gave:

:~$ systemd-analyze
Startup finished in 8.516s (kernel) + 23.338s (userspace) = 31.855s

I timed 33 seconds from grub to the login screen and 15 seconds from the login screen to a functional desktop.

Ah.... things back to normal!

[oshunluvr]

IMO, that's a bug. systemd shouldn't hang for 6+ minutes and then never complete because swap is missing.

[GreyGeek]

Probably a kernel bug and not a Btrfs bug, since Btrfs doesn't use swap?

[oshunluvr]

Yeah,

[jlittle]

I timed 33 seconds from grub to the login screen and 15 seconds from the login screen to a functional desktop.

I suggest you consider getting an SSD, perhaps a smallish one just for the OS. I get about 2 s from grub to login prompt, and 3 s from login screen to desktop on a bad day. (It's 8 s for the POST, power on to grub, too much, so now I suspend.)