tgharold's questions

Running CentOS 6.5, a brand new "minimal" server install, trying to use LVM2 thin pool feature. Packages installed are:

lvm2 (2.02.111) device-mapper-persistent-data-0.3.2-1.el6.x86_64

/etc/lvm/lvm.conf has

...
thin_check_executable = "/usr/sbin/thin_check"
thin_repair_executable = "/usr/sbin/thin_repair"
...

(And yes those files exist in the file system.)

Error message in /var/log/boot.log is:

Setting up Logical Volume Management:   /usr/sbin/thin_check: execvp failed: No such file or directory
Check of pool vg/pool failed (status:2). Manual repair required!
/usr/sbin/thin_check: execvp failed: No such file or directory
/usr/sbin/thin_check: execvp failed: No such file or directory

I'm at a loss of what to do with this.

I know that in CentOS 6.x, you can put the following line in /etc/yum.conf which tells yum to keep the last N kernels (by default this is set to five):

installonly_limit=5

However, there are cases when using experimental kernels (such as from ELRepo) where I want to keep a specific kernel around even if there have been 5+ updates to that kernel's package.

Are there edits to be made to /boot/grub/grub.conf or are there configuration options for yum (or dracut) to do this?

How do I adjust the "User" line in a ~/.ssh/config file using ERB template files in Puppet so that it contains the correct username that matches the account name?

class accounts_global::tharold {
account { 'tharold':
    ensure => present,
    }
file { "/home/tharold/.ssh/config" :
    require => Account['tharold'],
    owner   => 'tharold',
    group   => 'tharold',
    mode    => '0600',
    content => template('accounts_global/user_ssh_config.erb'),
    }
} 

Content of the user_ssh_config.erb file looks like:

Host ssh.example.com
Port 22
User tharold
IdentityFile ~/.ssh/ssh-key

The question is, what should the <%= something =%> look like to replace "User tharold" in the template file with the account name of the user? This ERB config file is going to be used for multiple users, so I need to parameterize that part of the file.

Trying to use <%= @name %> ends up putting "accounts_global::tharold" in the file.

Here's the problem in a nutshell. We need to backup some SVN repositories, Samba shares, /home directories, postgresql nightly backups, etc. But the results are going to be written out to external USB drives (which are rotated weekly).

Mounting the USB drives is not a problem (autofs works well for this), and encrypting the drive with LUKS also works flawlessly. Currently we use rdiff-backup to backup the source files to a central server (total backup size is 800GB), then rsync the results out to the USB drive each night. This works very well, even though USB drives are slow, because only a small portion of the 800GB changes each day.

Needs:

• Maintains the ownership / permissions / ACLs (rsnapshot does not do this).
• Works well with USB drives which are mounted at different locations (/mnt/offsite/DRIVEA, /mnt/offsite/DRIVEB, ..., /mnt/offsite/DRIVEZ).
• Only writes the changes to the USB drive each night (limited backup window). The combination of rdiff-backup and rsync works well for this.
• Is smart enough that files where the file modification timestamp changed, but the contents didn't, does not cause the file to be rewritten in its entirety to the backup device.
• Capable of doing diffs of files, so that only the new bits each night get stored.

Although we like rdiff-backup + rsync, we worry a lot that the rdiff-backup directory is fragile for anything other then the latest revision. Because older revisions inside the backup are done as a series of diffs and if one of those gets corrupted, you lose all older history.

Right now, the contents of /etc/rsyslog.conf which control disposition of the named log messages looks like:

# Log anything (except mail) of level info or higher.
# Don't log private authentication messages!
*.info;mail.none;authpriv.none;cron.none;;kern.none     /var/log/messages

Putting the following in front of that section correctly writes out the "named" messages to /var/log/named/named.log.

# Write named/bind messages to their own log file
:programname, isequal, "named"                          /var/log/named/named.log

The problem is that those "named" messages are still being written out to the /var/log/messages file. How would I modify the line that generates /var/log/messages to not write out "named" messages?

Note: This is rsyslog v5 as ships with RHEL/CentOS 6.

Addendum: The accepted answer from below is

# Write named/bind messages to their own log file, then discard (tilde)
:programname, isequal, "named"                          /var/log/named/named.log
:programname, isequal, "named"                          ~

How would we setup bacula so that it creates file volumes in a manner where the volume names (labels) are predictable and things work well with rsync so that we're not sending massive amounts of data across the wire?

General strategy would be to write a full backup on 1st Sunday of the month, differentials on the other Sundays, and incrementals Mon-Sat. So we would want (3) files created, one for the "full", 1-5 for the "differentials", plus 1-6 for the "incrementals".

My thinking is that the first line of defense is the "Label Format" directive in the Pool resource. Using a label format like "${Storage}-${Job}-${Pool}-${Level}-${MediaType}" on the pool definition and using (3) different pools for (full / incremental / differential) seems like it would work, but I'm not sure how to convince Bacula to append on the proper days and recycle on the proper days.

If you want to have multiple pools for purposes of different retention levels for full / diff / incremental, can you point them all at the same storage device? When each pool goes to do its retention / purge / prune thing using the settings in the pool, is it smart enough to only touch volume files that were created by that particular pool?

One disadvantage of putting multiple pools at the same device is that a device can only service one pool at a time. (But not confirmed.)

We're trying to make heads or tails out of our Bacula setup. We have a SD (storage daemon) server with a few LVM logical volumes and we're trying to do full/diff/incr backups to it in a sensible manner. The problems is that Bacula constantly complains of "Cannot find any appendable volumes" even though we've setup volume limits, told it to recycle the oldest, etc.

First off, we have (3) devices defined on the SD server. One for "Full", one for "Incremental" and one for "Differential". Ideally, we'd like (2) full monthlies, with differentials on the intervening weeks and incrementals for each day.

I won't cover the SD configuration as the main question here is about schedule and pools.

Schedule {
  Name = SVNSchedule
  Run = Level=Full 1st sun at 21:05
  Run = Level=Differential 2nd-5th sun at 21:05
  Run = Level=Incremental mon-sat at 21:05
}

Pool {
  Name = SVNFullPool
  Storage = svn-full
  Pool Type = Backup
  Recycle = yes; AutoPrune = yes; Recycle Oldest Volume = yes
  Volume Retention = 45 days
  Maximum Volume Jobs = 1
  Label Format = "$Pool-${Year}-${Month:p/2/0/r}-${Day:p/2/0/r}-${Hour:p/2/0/r}${Minute:p/2/0/r}"
  Maximum Volumes = 2
}

Pool {
  Name = SVNDiffPool
  Storage = svn-diff
  Pool Type = Backup
  Recycle = yes; AutoPrune = yes; Recycle Oldest Volume = yes
  Volume Retention = 40 days
  Maximum Volume Jobs = 1
  Label Format = "$Pool-${Year}-${Month:p/2/0/r}-${Day:p/2/0/r}-${Hour:p/2/0/r}${Minute:p/2/0/r}"
  Maximum Volumes = 10
}

Pool {
  Name = SVNIncrPool
  Storage = svn-incr
  Pool Type = Backup
  Recycle = yes; AutoPrune = yes; Recycle Oldest Volume = yes
  Volume Retention = 20 days
  Maximum Volume Jobs = 1
  Label Format = "$Pool-${Year}-${Month:p/2/0/r}-${Day:p/2/0/r}-${Hour:p/2/0/r}${Minute:p/2/0/r}"
  Maximum Volumes = 7
}

It almost feels like we should be using the "Volume Use Duration = 1w" option on the "full" pool. But the main issue we're having is that Bacula won't automatically recycle the file volumes. At least, not during testing where we're kicking off a full backup, then kicking off another one a few hours later.

The other minor issue that I'm not happy with is how to get Bacula to create backup files with predictable names, because the current setup causes havoc with trying to rsync the files offsite.

Using the default setup of Logwatch on RHEL 6 and the caching name server (BIND 9) package, my daily logwatch reports are filled with unmatched entries regarding DNS server errors.

What do I need to change in the configuration in order to either hide those errors or do I need to modify the RHEL6 options for their caching name server package (/etc/named.caching-nameserver.conf).

--------------------- Named Begin ------------------------ 

 **Unmatched Entries**
    connection refused resolving '1.146.219.117.bb.barracudacentral.org/A/IN': 75.101.143.130#53: 1 Time(s)
    connection refused resolving '1.146.219.117.bb.barracudacentral.org/A/IN': 75.101.148.228#53: 1 Time(s)
    connection refused resolving '1.186.145.75.bb.barracudacentral.org/A/IN': 75.101.143.130#53: 1 Time(s)
    connection refused resolving '1.186.145.75.bb.barracudacentral.org/A/IN': 75.101.148.228#53: 1 Time(s)
    connection refused resolving '1.240.27.165.bb.barracudacentral.org/A/IN': 75.101.143.130#53: 1 Time(s)
    connection refused resolving '1.240.27.165.bb.barracudacentral.org/A/IN': 75.101.148.228#53: 1 Time(s)