On some workloads, a process creating backups should not become so
aggressive on resources that it prevents normal activity of a machine
from running. It is especially important when writing to disk.
One way to make backups have less impact on the main workload is to
force backups to a lower ionice level.
With the new ionicelevel configuration, it becomes possible for users to
activate the use of ionice and to define the level within the
best-effort class to better suit their needs.
Since the luks headers backup was introduced in the handler, cryptsetup
has gained the "luksHeaderBackup" command. We should use this instead
of dd, as it is likely more reliable. Plus, it supports backing up luks
version 2 headers, which the current code doesn't handle.
The previous sfdisk-based command would also list partitions, leading to
partition-based luks volumes being duplicated. Instead, reuse the same
command used in the partition backup component.
borg will by default create its cache directory in ~/.cache/borg. This
means that during backup runs, borg will read and write quite
extensively from/to this directory.
In some situations, it is rather undesirable to have this amount of IO
activity in this location and it would make sense to tell borg to place
its cache elsewhere. This can help for example with placing the cache on
a hard drive where the added IO load will not have as big of an impact
on other running activity for the system.
This change also makes sure that the cache directory environment
variable is cleared out when the configuration option is unset. This
should avoid unpleasant surprises when this environment variable is set
to some unknown value in the context where backupninja is called, which
could lead to borg reading and writing to random places on the system.
This enables the handlers to be used for hosts which use a non-standard
port, such as is offered by some commercial services. Default to port
22, so it's optional and backwards-compatible.
It seems like only the short output of hwinfo --disk is provided to hardware.txt. I'd suggest we also add disk to the flags for the "complete" output.
That's especially useful when you need to indicate the serial number of a dead drive in order for it to be replaced. Once the disk is dead, using hardware.txt can be the only way to retrieve this info.
This fixes cases where the day component of the when parameter resolved
to a non-zero padded integer under 10, such as "when = 1st at 01"
The comparison against the nowday variable failed consistently since
the date command format used always returned a zero-padded two digit
integer.
The solution chosen is to avoid generating nowday with zero-padding and
remove any zero-padding from whenday parameter using a more thorough
sed pattern which now also removes any non-alphanumeric characters from
the string.
