flat-zoo-31952
10/21/2021, 5:37 PMThis whole thing fits very naturally as a graph of dependencies. We could model it this way and execute nodes in the graph as their dependencies are satisfied, and if nodes fail we can retry N times before marking them completely failed.
I built a little test executor for this using networkx and it looked like it could work pretty well. It basically worked like:
- build graph, eg base image depends on container, provider image depends on base image, provider upload depends on provider image etc.
Then, to execute:
- gather all stages that are runnable (dependencies are satisfied)
- submit to threadpool to run
- sleep
- walk the graph again and find any stages that are now runnable and submit to threadpool to run
- sleep
- continue until we can't progress any more or everything is doneIf only he knew there was tooling that already basically did this, but in a little more refined way. 😀 Looks like I've got another person I can make a good case for Pants to.
curved-television-6568
10/21/2021, 5:43 PM