T20150930.md

Turin, Wed 30 Sep - 2015

Deliverables:

• We currently are able to deploy plancton daemon on every docker-compatible host (at least Python2.7 is required).
• We currently can upgrade/modify plancton daemon and its configuration without access the hosts.

Goals achieved:

1. Centos6/SCL6 based containers parrot+CVMFS aware, capable to have access to the experiment software.
2. Condor cluster using containers as workers.
3. An easily deployable daemonized python script capable to:
   1. perform a continuous check of resource availability on the host, and take decisions following specific rules.
   2. use latest image in the registry.
   3. create and manage containers according to user usage on the host.
4. a container pilot as a container entry-point: start the condor daemon(s) -> wait for a job -> eventually execute it, if any -> exit the container.

In progress:

1. apply new procedure to decide how many containers deploy based on cpu-shares, cpucounts.
2. unification of configuration files (condor, plancton, docker-container).
3. mount condor configuration files directly into the container (currently the pilot downloads them).
4. easy deployment reliable procedure (e.g. bash <(curl -sSL https://raw.githubusercontent.com/mconcas/plancton/master/install))
5. reorganization of repositories depending on production environment.

In the future:

1. test a real/test job on the cluster.
2. interface for jobs submission.
3. Minor fixes:
   • shrink current container size.
   • correct, if it's possible, issues with condor_status who lists persistently dead containers for a while.