perf(robot-server): Discriminate command unions

[SyntaxColoring]

Overview

Now that we have Pydantic 1.9, we can close RSS-128. This dramatically improves error reporting when you misspell something in a Protocol Engine command, and it also improves performance anywhere we're parsing Protocol Engine commands.

Test Plan

For "this doesn't break anything," I'm mostly leaning on integration tests.

• To do: Also push this to an OT-2 to make sure the new typing.Annotated usage hasn't broken anything across Python versions.

For improved error reporting:

• I've manually used Postman to check POST /runs/commands
• I've manually used Postman to check POST /maintenance_runs/commands
• I've manually used Postman to check POST /commands

For improved performance:

• I've manually used Postman on a dev server to check the speed of POST /runs with a JSON protocol. Seems ~20-30% better.
• The slowest test in tests/integration/http_api/persistence/test_compatibility.py goes from ~23s to ~16s.

Changelog

• Annotate our command unions as being discriminated by their commandType field. See the Pydantic documentation: https://docs.pydantic.dev/1.10/usage/types/#discriminated-unions-aka-tagged-unions
• Unfortunately, before Pydantic v2, there appear to be bugs that prevent this from working when the union appears inside a pydantic.Generic model. To work around this, I unfortunately had to make some non-generic duplicates of our RequestModel, which is unfortunate.

Review requests

• Are there any HTTP endpoints I missed where we're parsing Protocol Engine commands?
• Can you think of a better workaround for the pydantic.Generic thing?

Risk assessment

Low—if this broke anything, I think integration tests would cover it, unless there's weirdness across Python versions.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (ca70ab0) 67.96% compared to head (ef02312) 68.08%.
Report is 2 commits behind head on edge.

❗ Current head ef02312 differs from pull request most recent head be08e43. Consider uploading reports for the commit be08e43 to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             edge   #14286      +/-   ##
==========================================
+ Coverage   67.96%   68.08%   +0.12%     
==========================================
  Files        2467     2509      +42     
  Lines       70000    71445    +1445     
  Branches     8868     9063     +195     
==========================================
+ Hits        47573    48645    +1072     
- Misses      20356    20689     +333     
- Partials     2071     2111      +40     

Flag	Coverage Δ
app	65.38% <ø> (ø)
g-code-testing	96.48% <ø> (ø)
labware-library	41.10% <ø> (ø)
notify-server	89.13% <ø> (ø)
protocol-designer	34.93% <ø> (ø)
step-generation	86.90% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
...entrons/protocol_engine/commands/command_unions.py	100.00% <ø> (ø)
robot-server/robot_server/commands/router.py	100.00% <ø> (ø)
...server/robot_server/commands/stateless_commands.py	100.00% <ø> (ø)
...server/robot_server/runs/router/commands_router.py	100.00% <ø> (ø)

... and 42 files with indirect coverage changes

[jbleon95]

Are the changes made to the command schema compatible with the make command-schema target? Mostly in that it doesn't overwrite that information, but also if it would (or can) add that discriminator information automatically.

[SyntaxColoring]

Are the changes made to the command schema compatible with the make command-schema target? Mostly in that it doesn't overwrite that information, but also if it would (or can) add that discriminator information automatically.

Yep, changes to 8.json in this PR were generated automatically via make -C api command-schema COMMAND_SCHEMA_VERSION=8 && make format-js.

[jbleon95]

Looks great, glad we could finally get to this with the Python updates