FWIW, I had to implement a monthly/weekly/daily backup scheme a while back and determined that the most reliable way to do it was to handle it myself rather than trying to encode it all through systemd units. The script does this basically: #!/usr/bin/env python3 from datetime import datetime now = datetime.now() if now.day == 1: # Monthly backups on the first of the month monthly_backup() if now.weekday() == 6: # Weekly backups on Sunday weekly_backup() daily_backup() # Always perform daily backups This is triggered from a daily systemd timer. -- Dan Nicholson | Endless OS Foundation -- Dan Nicholson | Endless OS Foundation On Fri, Jul 12, 2024 at 2:50 AM <t.schneider@xxxxxxxxxxx> wrote: > > There's no script (or whatsoever) in this backup functionality. > It's only rsnapshot (https://wiki.archlinux.org/title/Rsnapshot) and systemd-timer involved. > > > Am 2024-07-12 10:35, schrieb Barry Scott: > > > > On 12 Jul 2024, at 08:34, t.schneider@xxxxxxxxxxx wrote: > > Setting the time of day requires detailed knowledge about the runtime of each job. > > But this forecast is not accurate and this means any setting could result in same error. > > Agreed that is a problem with this approach. > > Or you have the code that locks wait for the lock to be freed does that help? > Are you using flock in the script, then its a easy change. > > You still cannot ensure the order, but you can ensure that they will all run sequentially. > > Personally I encode the rules for my incremental vs. full backup into my hourly backups script. > So I only have one timer. > > Barry >