Please review my Bluetooth run script

Please review my Bluetooth run script

[arasv]

Hi there,

I'm using the runit scripts in Busybox v1.11.1. I've pretty well moved
all my old init.d scripts over to the runit format and I'm very happy
with it.

I had a little bit of trouble with my Bluetooth Blue-Z startup script
but I have a working version.

What I want to know is if this is the "proper" way to go about it. The
original Blue-Z startup script starts quite a few daemons all at once,
and it kills off each of them with a "killall". I've managed to split the
old init.d script into a "run" and "finish" script. I also have a "down"
file so Bluetooth is only started when a BT dongle is added (triggered via
udev).

Here is my (simplified!) run script:

/etc # cat /etc/sv/bluetooth/run
#!/bin/sh
set -e

PATH=/sbin:/bin:/usr/sbin:/usr/bin

[ -e /etc/default/bluetooth ] && . /etc/default/bluetooth

# start all the Bluetooth services
hcid -f $HCID_CONFIG
sdpd
hidd $HIDD_OPTIONS --server || true
hid2hci --tohci > /dev/null 2>&1 || true
rfcomm -f $RFCOMM_CONFIG bind all || true
dund $DUND_OPTIONS
pand $PAND_OPTIONS

# now wait until we are told to finish by sv
trap 'exit 2' 1 2 3 15
exec sh -c "while [ 1 ]; do sleep 100; done"

exit 0


Here is the finish script:

/etc # cat /etc/sv/bluetooth/finish
#!/bin/sh

set -e

killall pppd > /dev/null 2>&1 || true
killall pand > /dev/null 2>&1 || true
killall dund > /dev/null 2>&1 || true
rfcommm release all > /dev/null 2>&1 || true
killall hidd > /dev/null 2>&1 || true
killall sdpd > /dev/null 2>&1 || true
killall hcid > /dev/null 2>&1 || true

exit 0



Does this seem reasonable?

Aras Vaichas



______________________________________________________________________
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email 
______________________________________________________________________

Re: Please review my Bluetooth run script

[Laurent Bercot]

> What I want to know is if this is the "proper" way to go about it. The
> original Blue-Z startup script starts quite a few daemons all at once,
> and it kills off each of them with a "killall".

 The original point of service supervision, what caused it to be
implemented (with daemontools first, then with runit) in the first place,
is that some daemons need to be automatically restarted if something
somehow causes them to die.

 Your current "run script" does not accomplish this.
 It starts the various Bluetooth daemons, and kills them, just like a
System V script would do. Those daemons are not supervised; what is
supervised is actually the following script:

   exec sh -c "while [ 1 ]; do sleep 100; done"

 I'm sure you will agree that supervising this script isn't particularly
useful. :)

 In other words, you've successfully converted your System V scripts
to the runsv model, but you're not taking advantage of what service
supervision can do for you. As is, you'd still be better off using a
start/stop SysV script, and do without your akward "bluetooth" service
which isn't a real service.

 My advice would be to have separate services for hcid, sdpd, hidd,
rfcommm (if it's a long-lived daemon), dund, pand and pppd. All those
daemons need to be kept alive and running; as an additional benefit,
you don't have to lose their logs by sending them to /dev/null, you
can make use of runsv's logging facility to send each daemon's log
into its own multilog or whatever your favorite logging system is.

 Then, once you have set up one service for each Bluetooth daemon
that needs to be up, you can have a "bluetooth start" script that
does a "sv up" on all those services at once, and a "bluetooth stop"
script that does a "sv down" on them. And you can even control
each daemon separately if the need arises.

-- 
 Laurent

Re: Please review my Bluetooth run script

[Aras Vaichas]

Laurent Bercot wrote:
>> What I want to know is if this is the "proper" way to go about it. The
>> original Blue-Z startup script starts quite a few daemons all at once,
>> and it kills off each of them with a "killall".
>>     
>
>  The original point of service supervision, what caused it to be
> implemented (with daemontools first, then with runit) in the first place,
> is that some daemons need to be automatically restarted if something
> somehow causes them to die.
>
>  Your current "run script" does not accomplish this.
>  It starts the various Bluetooth daemons, and kills them, just like a
> System V script would do. Those daemons are not supervised; what is
> supervised is actually the following script:
>
>    exec sh -c "while [ 1 ]; do sleep 100; done"
>
>  I'm sure you will agree that supervising this script isn't particularly
> useful. :)
>   
Oh yeah, I was never happy about that bit. The run script for the
bluetooth was a first-pass conversion from the Sys-V model. I used the
"mountd" run script as an example of starting multiple daemons from the
one place.

>  My advice would be to have separate services for hcid, sdpd, hidd,
> rfcommm (if it's a long-lived daemon), dund, pand and pppd. All those
> daemons need to be kept alive and running; as an additional benefit,
> you don't have to lose their logs by sending them to /dev/null, you
> can make use of runsv's logging facility to send each daemon's log
> into its own multilog or whatever your favorite logging system is.
>   
We're running an embedded Linux system, so I was trying to keep the
number of running processes to a minimum. I guess the runsv is fairly
light?

I think I will follow your advice and separate them.

As a side note, the logging could be handy during development, but we
don't keep any permanent logs because we run a flash filing system.

>  Then, once you have set up one service for each Bluetooth daemon
> that needs to be up, you can have a "bluetooth start" script that
> does a "sv up" on all those services at once, and a "bluetooth stop"
> script that does a "sv down" on them. And you can even control
> each daemon separately if the need arises.
>   
So are you recommending that I use a Sys-V script to start all the runsv
services?
e.g.
case "$1" in
  start)
    sv up hcid
    ...
    sv up sdpd
        ;;
  stop)
    sv down
     ...
    sv  down hcid
    ;;;
esac

Would I be able to use a runsv style of script to do this? Then I have
the same problem with "ending" the run script because the run script
doesn't end in a daemon process which is why I had the sleep loop in the
original run script.

I currently use UDEV to do an "sv up bluetooth" when the Bluetooth
device is inserted. I guess I could just as easily do "sv up hcid; sv up
sdpd; ... ".

No-one actually calls "sv down" because it's an embedded system with no
physical user interface, the services are only stopped when the device
is power cycled, which is fairly common and expected.

Aras

______________________________________________________________________
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email 
______________________________________________________________________

Re: Please review my Bluetooth run script

[Laurent Bercot]

> We're running an embedded Linux system, so I was trying to keep the
> number of running processes to a minimum. I guess the runsv is fairly
> light?

 Ah, it's good too see that embedded developers use smart supervision
systems and good-quality, small-sized, software. :)
 The runsv process is quite light, you don't need to worry about it.
To be convinced, compile runit statically with the small libc of your
choice, and look at the size of the executables and the memory
footprints.

 Note: if you're not using logging at all, and you need to optimize
as much as you can, you might consider switching to daemontools.
runsv embeds the logging facility, while supervise does not, so I
think supervise should be a bit lighter. But it's a very minor
optimization, and sticking to runit might give you more benefits.


> So are you recommending that I use a Sys-V script to start all the runsv
> services?

 I'm certainly not advocating a SysV scheme. Actually I'm not advocating
any scheme at all. A simple one-time script will do. The SysV stuff
is an overcomplicated way of running one-time scripts. :)


> I currently use UDEV to do an "sv up bluetooth" when the Bluetooth
> device is inserted. I guess I could just as easily do "sv up hcid; sv up
> sdpd; ... ".
> 
> No-one actually calls "sv down" because it's an embedded system with no
> physical user interface, the services are only stopped when the device
> is power cycled, which is fairly common and expected.

 Well, as I understand it, you want to start your Bluetooth daemons as
soon as your embedded thingie is powered: then you don't need to write
any script at all. Just have your Bluetooth service directories, with
appropriate run scripts, in /var/service or wherever runsvdir is run.
 runsvdir will spawn runsv processes, which will automatically start
your daemons (unless you have "down" files in the service directories).

 If you need to send rfcomm commands, perform them in the one-time
initialization (i.e. /etc/runit/1), since rfcomm is a short-lived
program. If you need to send them _after_ the daemons are up, it's
a bit trickier; but you should already have devised a way to
execute a one-time script once the services have been started, just
add your rfcomm lines to that script.

-- 
 Laurent