After a short chat with Jay, we found a solution - simply chain /bin/bash to the end of the second puppet run which will keep the container running and also provide the user with interactive shell.
The only requirement now is for user to run the container with correct "-h" option. Will upload new patch shortly to the JIRA Thanks, Cos On Fri, Jan 29, 2016 at 09:39AM, Konstantin Boudnik wrote: > Guys, > > I am trying to build a fully provisioned Docker container with Hadoop > installed and configured in it. The reason is pretty simple: it will give a > user an ability to quickly run a sandbox environment with some/all bits in it. > > Of course I am taking the advantage of our deploy images and puppet recipes, > but here's a snag: at the moment of image creation the FQDN of the container > could not be set, so the image will, by definition, get setup as a worker > node. I have devised a contraption where I install master packages, like > hadoop-hdfs-namenode) as a part of the image build. Then at the container > creation time I am running 'puppet apply' again and in a matter of 60-70 > seconds everything gets configured, based on the actual hostname, and the > services got started. I think this is a pretty neat workaround. compared to > what I've seen on the net with awful miles of sed statements do fix the > configs, followed by scripted service runs, etc. Nasty... > > I want to do this second run automatically, when the container get created, so > used doesn't need to do anything extra. I've tried to see if I can tail-gait > ENTRYPOINT to do the second run of the 'puppet apply', but understandably the > container stops right after puppet run is over. Any ideas how this can be > achieved? Would be good to have 1.1 official docker image once it is out. > > For reference, the initial version of this is sitting as a patch on > BIGTOP-2296 > > Looking forward for your thoughts! Thanks > Cos >
signature.asc
Description: Digital signature
