I’ve used Bacula since at least January 2004 (so nearly 20 years). I liked it so much I dropped my deployment-in-motion of another tool (if you search lightly, you can find out which one). I liked it so much, I wrote a PostgreSQL backend for it.
This post is not for Bacula novices. This post is for those who have already deployed a Bacula instance and are rather familiar with the process. I recommend this older post, which still applies today. For a basic introduction, please try Bacula on FreeBSD with ZFS.
This post will combine several of my favorites tools:
- FreeBSD (13.2-RELEASE)
- Bacula (9.6.7 – I am far behind on my upgrades here)
- ZFS (2.1.9)
- ZFS quota
- ZFS reservation (perhaps)
- ZFS snapshots (via syncoid)
- jails
The primary goal of this post is setting Maximum Volume Bytes and Maximum Volumes on the new Bacula Pools and describing the methods used to come up with these values.
Background – moving and downsizing
I am moving an existing Bacula Storage Daemon (bacula-sd). I am also downsizing in the disk space available. I’ll be moving some backups via Bacula copy jobs, but that will not be covered by this post.
In configuring the new bacula-sd, I will be creating new Pools. In order to conserve disk space, I configure Bacula to recycle Volumes. Bacula does everything it can to avoid overwriting existing backups. I impose a restriction upon Bacula by limiting the maximum number of Volumes in a pool. How do I determine those numbers? Estimation, and historical data. In this posts, I’ll provide some sample queries I used in setting these values.
In this particular case, I’m moving an existing bacula-sd from one host (knew) to another (r730-03). I run my bacula instances in jails. This simplifies many administration tasks, including moving to another host.
The storage
I’ve created three ZFS filesystems to receive the backups:
[16:50 r730-03 dvl ~] % zfs list -r data01/bacula-volumes NAME USED AVAIL REFER MOUNTPOINT data01/bacula-volumes 440K 8.84T 96K /jails/bacula-sd-04/usr/local/bacula/volumes data01/bacula-volumes/DiffFile 152K 8.84T 96K /jails/bacula-sd-04/usr/local/bacula/volumes/DiffFile data01/bacula-volumes/FullFile 96K 8.84T 96K /jails/bacula-sd-04/usr/local/bacula/volumes/FullFile data01/bacula-volumes/IncrFile 96K 8.84T 96K /jails/bacula-sd-04/usr/local/bacula/volumes/IncrFile
I have set these non-default values for these datasets:
- recordsize 1M
- compression lz4
- atime off
I plan to set these so the above mentioned filesystems do not fill up.
- quota
- reservation
Pools
These are the Pools I created for this new bacula-sd:
Pool {
Name = FullFile-04
Pool Type = Backup
Recycle = yes
AutoPrune = yes
Volume Retention = 1 years
Storage = bacula-sd-04-FullFile
Next Pool = Fulls
Maximum Volume Bytes = 20G
Maximum Volume Jobs = 1
Maximum Volumes = 2000
LabelFormat = "FullAuto-04-"
}
Pool {
Name = DiffFile-04
Pool Type = Backup
Recycle = yes
AutoPrune = yes
Volume Retention = 6 weeks
Storage = bacula-sd-04-DiffFile
Next Pool = Differentials
Maximum Volume Bytes = 20G
Maximum Volume Jobs = 1
Maximum Volumes = 240
LabelFormat = "DiffAuto-04-"
}
Pool {
Name = IncrFile-04
Pool Type = Backup
Recycle = yes
AutoPrune = yes
Volume Retention = 3 weeks
Storage = bacula-sd-04-IncrFile
Next Pool = Incrementals
Maximum Volume Bytes = 20G
Maximum Volume Jobs = 1
Maximum Volumes = 450
LabelFormat = "IncrAuto-04-"
}
The Name and LabelFormat include a -04 suffix to make them unique across all Volumes. This choice is personal preference and not logically linked to anything. The 04 corresponds to bacula-sd-04.
How did I get to those numbers?
I started with my data, in the Catalog, and looked at my backups for the past month. These queries are listed here mainly for my future use and I hope they are also useful to you.
In all cases, you’ll need to modify the queries if you want to use a different data range. I run full backups on the first Sunday, diffs on all other Sundays, and incrementals on all other remaining days. I limit Volume usage to a single Job.
Full jobs, run at the start of the month
Note that I am excluding Copy jobs in this query. They go to another host.
select * from job where starttime between '2023-08-06 00:00:00' and '2023-08-06 23:59:59' and level = 'F' and type != 'C' order by name;
How much disk spaces did the full backups use?
select sum(jobbytes) from job where starttime between '2023-08-06 00:00:00' and '2023-08-06 23:59:59'
and level = 'F' and type != 'C';
What jobmedia entries are involved?
These queries were used to create queries which follow later in this post.
select jobmedia.* from job, jobmedia where starttime between '2023-08-06 00:00:00' and '2023-08-06 23:59:59' and level = 'F' and type != 'C' and job.jobid = jobmedia.jobid;
That’s a list of ALL the jobmedia records. The following is just a list of the Volumes used.
select distinct jobmedia.mediaid from job, jobmedia where starttime between '2023-08-06 00:00:00' and '2023-08-06 23:59:59' and level = 'F' and type != 'C' and job.jobid = jobmedia.jobid;
Getting the space occupied on disk
The list of Volumes for the Full backups on the first Sunday of the month.
select media.* from job, jobmedia, media where starttime between '2023-08-06 00:00:00' and '2023-08-06 23:59:59' and level = 'F' and type != 'C' and job.jobid = jobmedia.jobid and jobmedia.mediaid = media.mediaid;
Using that to get the space used:
select sum(media.volbytes) from media where mediaid in ( select distinct jobmedia.mediaid from job, jobmedia where starttime between '2023-08-06 00:00:00' and '2023-08-06 23:59:59' and level = 'F' and type != 'C' and job.jobid = jobmedia.jobid);
About 460 GB – that’s all my full backups for a month.
What about incremental jobs?
The time periods vary with the type of Job, which relates to the Retention period involved.
Disk space used:
select sum(jobbytes) from job where starttime between '2023-07-01 00:00:00' and '2023-07-21 00:00:00' and level = 'I' and type != 'C';
About 2532814783461 bytes or 2.5 TB – I get a lot of data churn for incrementals.
Number of Jobs:
select count(*) from job where starttime between '2023-07-01 00:00:00' and '2023-07-21 00:00:00' and level = 'I' and type != 'C';
About 303
How many Volumes?
select count(distinct jobmedia.mediaid) from job, jobmedia where starttime between '2023-07-01 00:00:00' and '2023-07-21 00:00:00' and level = 'I' and type != 'C' and job.jobid = jobmedia.jobid
About 432.
differentials
select sum(jobbytes) from job where starttime between '2023-07-01 00:00:00' and '2023-08-14 00:00:00' and level = 'D' and type != 'C';
996623479547 bytes or about 1 TB.
select count(*) from job where starttime between '2023-07-01 00:00:00' and '2023-08-14 00:00:00' and level = 'D' and type != 'C';
110 jobs
select count(distinct jobmedia.mediaid) from job, jobmedia where starttime between '2023-07-01 00:00:00' and '2023-08-14 00:00:00' and level = 'D' and type != 'C' and job.jobid = jobmedia.jobid
202 Volumes.
Other full backups during the month
There are other full backups I run daily throughout the month: git and subversion repos, pfSense configuration, etc.
In this case, I’m restricting this to the pool used for Full backups.
select sum(job.jobbytes) from job where starttime between '2023-08-07 00:00:00' and '2023-08-31 23:59:59' and level = 'F' and type != 'C' and poolid=22;
Those 36 jobs uses about 140 GB / month, or about 1.7 TB / year.
How many volumes does that use?
select count(distinct jobmedia.mediaid) from job, jobmedia where starttime between '2023-08-07 00:00:00' and '2023-08-31 23:59:59' and level = 'F' and type != 'C' and job.jobid = jobmedia.jobid;
About 70 volumes or 70 * 12 = 840 a year.
select count(distinct jobmedia.mediaid) from job, jobmedia where starttime between '2023-08-07 00:00:00' and '2023-08-31 23:59:59' and level = 'F' and type != 'C' and job.jobid = jobmedia.jobid;
70 Volumes. Or 70 * 12 = 840 more Volumes in the Full Pool over the year.
The figuring
Based on these values, I came up with these totals:
- 25 full backups a month – 430 GB
- 42 Volumes used
- 1 year of full backups is 430 * 12 = about 6 TB
- 1 year of full backups uses 25 * 42 = 1050 Volumes
These are rough figures. Taking the configuration presented near the start of this post, the summary is:
| Pool Name | Maximum Volume Bytes | Maximum Volume Jobs | Maximum Volumes | Max size used | |
|---|---|---|---|---|---|
| IncrFile-04 | 20G | 1 | 500 | 10TB | |
| DiffFile-04 | 20G | 1 | 240 | 4.8TB | |
| FullFile-04 | 20G | 1 | 2000 | 40TB |
You will notice that this requires 59TB of space. I have only 8.96TB free. How will this work?
What we have done is calculated the number of volumes used and set them to be 20GB max size. Full backups will use only 6TB, not 40TB. Not all Volumes are filled to the maximum capacity. Sure, when we have large jobs yes. Some of them will be full.
Similarly for Diff and Incr jobs.
However, this could still fill up the system. Let’s avoid that with ZFS.
ZFS quotas
With this statement, this ZFS file system cannot consume more than 6 TB for Full backups:
[18:58 r730-03 dvl ~] % sudo zfs set quota=6TB data01/bacula-volumes/FullFile [18:58 r730-03 dvl ~] %
Similarly, we impose similar restrictions on the other file systems.
[18:58 r730-03 dvl ~] % sudo zfs set quota=1TB data01/bacula-volumes/DiffFile [19:17 r730-03 dvl ~] % sudo zfs set quota=2.5TB data01/bacula-volumes/IncrFile
You will notice that I’ve imposed 9.5TB of quota on a filesystem with only 9 TB free. I’m in for trouble.
[19:23 r730-03 dvl ~] % zpool list data01 NAME SIZE ALLOC FREE CKPOINT EXPANDSZ FRAG CAP DEDUP HEALTH ALTROOT data01 10.9T 1.94T 8.96T - - 0% 17% 1.00x ONLINE -
I just ordered another pair of 12 TB drives.
