Differences between revisions 4 and 10 (spanning 6 versions)
⇤ ← Revision 4 as of 2010-07-26 17:05:30 →
Size: 916
Editor: SamatJain
Comment: Write-intent bitmaps
← Revision 10 as of 2020-09-23 06:34:48 → ⇥
Size: 2231
Editor: SamatJain
Comment: btrfs RAID0
Deletions are marked like this. Additions are marked like this.
Line 2: Line 2:

<<TableOfContents>>
Line 5: Line 7:
{{{#!highlight sh {{{#!highlight sh numbers=off
Line 7: Line 9:
sudo mdadm --create /dev/md5 --level=5 --raid-devices=3 /dev/sd[def]1 sudo mdadm --create /dev/md5 --level=5 --assume-clean --raid-devices=3 /dev/sd{d,e,f}1
Line 10: Line 12:
sudo mdadm --create /dev/md6 --level=10 -p f2 --raid-devices=2 /dev/sd[ik]1 sudo mdadm --create /dev/md6 --level=10 -p f2 --assume-clean --raid-devices=2 /dev/sd{i,k}1

# Save configuration (not necessary)
/usr/share/mdadm/mkconf > /new/etc/mdadm/mdadm.conf
Line 13: Line 18:
== Write-intent bitmaps == == Performance ==

=== Stripe cache size ===

Increase 1 MiB default to 16 MiB (№ 4K blocks):

{{{#!highlight sh numbers=off
echo 4096 | sudo tee /sys/block/md0/md/stripe_cache_size
echo 4096 | sudo tee /sys/block/md1/md/stripe_cache_size
}}}

=== Write-intent bitmaps ===
Line 31: Line 47:
{{{#!highlight sh {{{#!highlight sh numbers=off
Line 34: Line 50:

== Recovery ==

=== Inactive array ===

If an array comes out as inactive, e.g.:

{{{
md8 : inactive sdp1[0](S) sds1[4](S) sdx1[2](S) sdj1[1](S)
      7814054094 blocks super 1.2
}}}

Stop the array before we continue to work on it.

{{{#!highlight sh numbers=off
mdadm --stop /dev/md8
}}}

First, use examine the array components and figure out which component is out of sync (it's best to do this manually). Then, with the remaining components, reassemble the array. E.g. if /dev/sds1 was the bad device:

{{{#!highlight sh numbers=off
mdadm --assemble /dev/md8 /dev/sd{p,x,j}1 --force
}}}

to restart the array. If supposedly bad device is fine, go ahead and re-add it:

{{{#!highlight sh numbers=off
mdadm --manage /dev/md8 --add /dev/sds1
}}}

== btrfs ==

{{{#!highlight sh numbers=off
# btrfs RAID0
mkfs.btrfs -d raid0 /dev/sd{b,c}1
}}}

Contents

  1. Creation
  2. Performance
    1. Stripe cache size
    2. Write-intent bitmaps
  3. Diagnosis
    1. Examine array component
  4. Recovery
    1. Inactive array
  5. btrfs

Creation

# RAID 5 array creation
sudo mdadm --create /dev/md5 --level=5 --assume-clean --raid-devices=3 /dev/sd{d,e,f}1

# Create a high-performance RAID10 array (2x read speed compared to RAID1)
sudo mdadm --create /dev/md6 --level=10 -p f2 --assume-clean --raid-devices=2 /dev/sd{i,k}1

# Save configuration (not necessary)
/usr/share/mdadm/mkconf > /new/etc/mdadm/mdadm.conf

Performance

Stripe cache size

Increase 1 MiB default to 16 MiB (№ 4K blocks): 

echo 4096 | sudo tee /sys/block/md0/md/stripe_cache_size
echo 4096 | sudo tee /sys/block/md1/md/stripe_cache_size

Write-intent bitmaps

Write-intent bitmaps speed-up RAID resyncs significantly, at the expense of a little write performance. 

   1 # Create write-intent bitmap
   2 mdadm --grow --bitmap=internal /dev/mdX
   3 
   4 # Remove write-intent bitmap
   5 mdadm --grow --bitmap=none /dev/mdX

Diagnosis

Examine array component

Displays superblock information, including last event, RAID UUID, other components, etc. 

mdadm --examine /dev/sda1

Recovery

Inactive array

If an array comes out as inactive, e.g.: 

md8 : inactive sdp1[0](S) sds1[4](S) sdx1[2](S) sdj1[1](S)
      7814054094 blocks super 1.2

Stop the array before we continue to work on it. 

mdadm --stop /dev/md8

First, use examine the array components and figure out which component is out of sync (it's best to do this manually). Then, with the remaining components, reassemble the array. E.g. if /dev/sds1 was the bad device: 

mdadm --assemble /dev/md8 /dev/sd{p,x,j}1 --force

to restart the array. If supposedly bad device is fine, go ahead and re-add it: 

mdadm --manage /dev/md8 --add /dev/sds1

btrfs

# btrfs RAID0
mkfs.btrfs -d raid0 /dev/sd{b,c}1

CategoryCheatSheet

SamatsWiki: CheatSheet/LinuxSoftwareRAID (last edited 2021-11-01 21:59:06 by SamatJain)