Add --memory option to paasta local-run to allow overriding of memory…

[EvanKrall]

… limit set in yelpsoa-configs.

In my first attempt at this branch, I tried adding a memory argument to run_docker_container, and having

-    memory = instance_config.get_mem()
+    if memory is None:
+        memory = instance_config.get_mem()

However, the actual docker command has --mem-swap={instance_config.get_mem() + 64} (via format_docker_parameters calling get_mem_swap), and also we sometimes have environment variables that reference instance_config.get_mem(), so just modifying instance_config.config_dict["mem"] seems like a more complete solution.

[nemacysts]

would we also want to add similar versions of this for the other resource options?

(i also think that we didn't add this initially since we were expecting folks to use the -y flag to a local soaconfigs with the edited values)

[EvanKrall]

would we also want to add similar versions of this for the other resource options?

Ah, at first I thought we weren't even setting CPU limits because I only saw instance_config.get_cpus() referenced once in local_run.py, but it looks likeformat_docker_parameters also sets --cpu-quota based on cpus. I'll add --cpus and --disk.

(i also think that we didn't add this initially since we were expecting folks to use the -y flag to a local soaconfigs with the edited values)

yeah maybe, but adhoc batch runs are enough of a use case and it's a little unfortunate to tell people to go edit yelpsoa-configs (or make a local checkout and run with -y) just to rerun their batch so it doesn't OOM.