On Tue, Jul 11, 2017 at 12:49:16AM +0200, Heiko Schlittermann wrote:
> > > Jose M Calhariz <[email protected]> (Mo 10 Jul 2017 
> > > 19:51:31 CEST):
> > > …
> > > > I happily ship a systemd unit for amanda if someone writes one for the
> > > > Debian package.
> > > 
> > > Here are the unit files, I created just now. 
> > > The User= setting needs to be adapted probably.
> > 
> > Fedora and CentOS (thus probably RHEL also) have some slight
> > variations.  Plus I added a couple of additions.  May not be
> > appropriate for Debian, I don't know.
> 
> Thank you. Some notes though…
> 
> > > 
> > > ,---------------------------------[the socket unit]-----------
> > > | # amanda.socket
> > > | [Unit]
> > > | Description=Amanda Server socket
> > 
> > It is refered to as "Amanda Activation socket".  I suspect
> > because it is used by both server and client.
> 
> Yes, the description can follow some formalities, "Amanda Activation
> socket" reads good. Maybe even better: "Amanda Agent Activation socket",
> to avoid confusion with the "Amanda Server", the server, that controls
> the backup machinery.
> 
> 
> > > | Wants=network.target
> > > | 
> > > | [Socket]
> > > | Accept=true
> > > | # need an IPv4 address, as current (3.3.9) amandad can't deal with ipv6 
> > > here 
> > > | ListenStream=0.0.0.0:10080
> > > | 
> > > | [Install]
> > > | WantedBy=sockets.target
> > > `----------------------------------------------------------------
> > > 
> > > ,--------------------------[the service template unit]-----------
> > > | # [email protected]
> > > | [Unit]
> > > | Description=Amanda Server service
> > 
> > Again, an alternative description "Amanda Backup Service"
> 
> Yes, but see above. "Amanda Backup Agent service"
> 
> 
> > Added "After=local-fs.target".  Debian may not even have
> > the "local-fs.target", or they may have something similar.
> 
> Do the "After=" dependencies matter for socket activated services?

I don't know.  Why do you think they may not matter?

I'm guessing the reason for the added "After=local-fs.target"
is to avoid starting amanda on the client before the backup
source is mounted or on the server before the holding disk
and/or the vtapes are mounted.

> 
> > > | 
> > > | [Service]
> > > | User=backup             <------------- change it to your needs
> > 
> > "Group=disk" is added
> 
> Tried this, but this seems to override the groups, "backup" belongs to,
> leading to "permission denied" problems when executing the amcheck
> triggered tests.
> 
I haven't checked a running dump process.  On my systems Amanda's
primary group is "Disk", though an Amanda login process has several
secondary groups.  amcheck and other amanda programs run fine here.
I wonder why we see different effects.

> > 
> > > | StandardInput=socket
> > > | StandardError=journal

Does this prevent writing to amanda's normal log/debug files?

> > > | ExecStart=/usr/lib/amanda/amandad -auth=bsdtcp amdump amindexd 
> > > amidxtaped
> > 
> > Added valid commands "amdumpd" and "ambackupd" to allow server to
> > respond to recently added client requests.  One such requests is
> > "ambackup" which allows the client to requests it own immediate
> > backup to the holding disk for later autoflush during a regular
> > amdump run.  I use it for my sometimes on/sometimes off laptop.

BTW while the "Added" comment earlier meant added by fedora/centos,
here I mean "Added" by me, not the package developers.

> 
> Maybe this could be done via an env file /etc/default/amanda having
>     SERVICES="amdump amindexd amindxtaped…"
> 
> 
> and then extending the .service unit:
> 
>     EnvironmentFile=-/etc/default/amanda
>     ExecStart=/usr/lib/amanda/amandad -auth=bsdtcp $SERVICES

I was not aware of this capability.  I think I'd pick a more specific
variable name, like AMANDAD_SERVICES.

Jon
-- 
Jon H. LaBadie                 [email protected]
 11226 South Shore Rd.          (703) 787-0688 (H)
 Reston, VA  20190              (703) 935-6720 (C)

Reply via email to