Stepping Manager

steppingmanager may at some point get written.

It’s aim is to help with deploying and migrating multiple queries in production.


  • A query might get deployed via some Heroku like mechanism, with persistence, a REST API, client libraries, etc. “for free”.
  • Deployed steppings should have persistence at the input and the output to allow for replaying data on changes. It should be easy to eg. replay a load of POSTs between instances.
  • A /steppingmanager directory should know about the graph of deployments, and handle replays etc.
  • We should version control a dump of the whole graph. This should be a .py file, but with no dataclasses, only atoms (int, str, etc) - see isjsonschemasubset.
  • Handle schema changes by replaying data/connecting to old sources.
  • A worker should expose the diff between the time of the most recent event and most recently published, or some vaguely Spanner-like approach to consistency.
  • There should be some graceful error handling for queries.