Comment # 6
on bug 1109
from Frank Ch. Eigler
> It's an important usability issue, it absolutely must match Docker
> behaviour. The whole point of the fuzzy matching is to provide the
> behaviour that users expect
According to a little experiment here, docker-1.6 (f21) actually behaves
in the way that I proposed: ambiguous container prefixes are rejected.
Try:
% for i in `seq 20`
do
docker run --rm busybox sleep 128 </dev/null &
done
over and over until you get some containers with the same first-two bytes 0xAB.
Then try
% docker inspect AB
you will see docker reject it. Then do:
% docker kill ABFIRST ABSECOND # but not ABLAST
Then wait a moment, then
% docker inspect AB
will now be accepted and describe ABLAST. I haven't inspected docker golang
code to confirm.
> (same reason why a docker: prefix on all name was never going to fly).
Then you suffer collisions between docker and non-docker container
namespaces: something irrelevant to docker end-users but significant
to a sysadmin with both lxc & docker on the system.
(Again I don't mind heuristics, but they should be unambiguous.)