Brian Thorne

This looks pretty handy - https://crashedmind.github.io/PlantUMLHitchhikersGuide/kubernetes/kubernetes.html

Now that anonlink builds and deploys a universal wheel we probably don't need `g++` or `libffi-dev`. Unsure if need `gmp-dev`, `mpfr-dev,` `mpc1-dev` anymore too.

This looks like it could address the multi-stage build: https://tech.zarmory.com/2018/09/docker-multi-stage-builds-for-python-app.html Another consideration - perhaps more important - is the container build time should be reduced.

Build time has been reduced in #506 would still be nice to have a smaller base image by using multi-stage. cc @wilko77

https://datamade.us/blog/transactional-testing

Given the pr for supporting run deletion introduced a cascading delete on the permutation mask table, which is on the flask error, it could be the database for the integration...

Entirely unhelpful with regard to solving the problem... As impressive as the grep/sed command is you can probably use `docker port tools_nginx_1 "8851"` to get the port number you want.

@muelsy would you care to weigh in?

Possible prometheus metrics with [types](https://prometheus.io/docs/concepts/metric_types/): - number_of_mappings counter - mappings_count{status="progress"} gauge - mappings_count{status="failed"} gauge - matching_rate gauge (e.g. 22931985) - matching_job_last_success_duration_seconds gauge - matching_job_last_success_duration_seconds gauge - matching_job_last_duration gauge - matching_job_last_size...

Logging is now configurable, however the defaults are not ideal. Currently the api produces non-colored output like this: ``` 2020-02-04T21:52:05.003340Z [info ] request to access run result authorized [entityservice.views.run.results] pid=378e01e6922775785142ca4677337cb8376bcde73f077e96...