Comment 23 for bug 119660

Gürkan Sengün wrote:
> We've got 200 workstations for students, doctors, postdocs, professors
> and employees... the ugly hack on top works, however it's not the right
> solution, is it?
>

There's already a precedent set by other services to include scripts in
if-up.d/ and if-down.d/ to manage themselves in the event of a network
state change, so I would say that this approach is kosher enough to call
it correct. (It just didn't work for me, and I couldn't tell you why
not, so I implemented my horrible kludge instead and it works like a charm.)

I can also see that there are upstart scripts in if-up.d/ and if-down.d/
that emit "net-device-up" and "net-device-down", respectively, as a
result of a network state change. The problem is that autofs is still
started via an /etc/init.d script, and not by upstart, so these signals
may not have any bearing on the state of autofs. But given Ubuntu's
move to upstart I would image that future versions of autofs-ldap will
include an /etc/init/ upstart script that is smart enough to make use of
these signals and manage things properly.

...But as a thought experiment, I think it would be a bad idea for
services like autofs to listen directly for a "net-device-*" signal.
Systems with multiple network interfaces may, depending on their use,
bring network interfaces up and down often that would have no effect at
all on the state of their directory service or the ability for autofs to
function correctly.

Perhaps the proper solution, then, should also involve modifying the
LDAP and NIS packages to include functionality that will specifically
emit their own up or down signals, but only in the event that the
associated network interface through which directory services are bound
change state.

For instance, let's say that a system has six network interfaces: eth0
through eth5. The system is bound (via LDAP or NIS) to a server that is
only available via the eth3 interface. You don't want autofs to shut
itself down whenever it sees a "net-device-down" signal, because unless
that signal is in response to bringing eth3 down then directory binding
would continue to work properly.

The LDAP (or NIS) package could include an upstart script that listens
for "net-device-up" and "net-device-down" signals. When one is
received, check the connection to the directory server (be it LDAP or
NIS). If "net-device-down" was received and the the directory server is
now unreachable, then a "directory-binding-down" should be emitted.
Likewise, if a "net-device-up" signal was seen, then a timeout should be
started where the upstart script waits for directory binding to
succeed. If it does, then a "directory-binding-up" should be emitted.
If it times out first, emit "directory-binding-down".

Other services that rely on directory binding, such as the autofs-ldap
package, could listen specifically for the "directory-binding-down" and
"directory-binding-up" signals and know that it is now okay to take
appropriate action.

(...There's probably a better way to handle this, but this was an
off-the-hip shot.)

Michael