yellow team mailing list archive

[Francesco Banconi <francesco.banconi@xxxxxxxxxxxxx>]

Hi Serge,

I was trying to prototype an *lxc-ip* script that takes a container's
name and returns its ip address. Gary Poster spoke to you about the idea
earlier this week.
Today I've investigated ways to get the ip.  This email summarizes my
results. I'd really appreciate your comments and suggestions.

The first method taken into consideration was something like the following:

1. get the ip address of the dnsmasq virtual interface (checking for
lxcbr0 and falling back to virbr0 if the former does not exist)

2. use host command or similar to query the name server for the given
lxc name

Using this method I am encountering dns cache problems, e.g. if you stop
and then restart a container, dnsmasq returns the old ip address until
the started instance obtains the new one. I've tried to clear the dns
cache sending the HUP signal to the process, but it doesn't work.

Another story could be (as discussed with Gary):

1. get the ip of the container, e.g. parsing lxc-ps and looking for the
init process (lxc-info is another option but requires to be run as root,
and I think a goal for this script is to run without root access).

2. use the ip to obtain the time the instance was started (using
/proc/<pid>)

3. compare the lxc start time to the modification time of the leases
file (rootfs/var/lib/dhcp3/dhclient.eth0.leases or similar)

4. if leases_time > start_time, get the ip from the leases
   otherwise, wait until we have a new entry in leases
   (or just exit with an error, or both, depending on a flag).

I know this path is a little bit tricky, and I'd like to hear from you
before starting to follow it.

Thank you,

-- 
Francesco Banconi