fix(ingestion): install the repository into the Dagster+ image

[jirhiker]

Fixes ModuleNotFoundError: No module named 'db' in database_connectivity.

What the diagnostic showed

Materializing ingestion_heartbeat inside the container:

cwd                /opt/dagster/app
app_root_contents  ... automated_ingestion, db, domain, services ...
db_on_path         false
domain_on_path     false

db/ is present in the working directory and still not findable. So
/opt/dagster/app is simply not on sys.path in the process that executes a
step — while it is in the process that loads the code location, which is why
the location lists all four assets and ingestion_heartbeat runs.

Why my first two attempts failed

Both treated the symptom, and both were wrong about the mechanism:

1. Inserting the repository root from automated_ingestion/__init__.py —
   assumed the package's own location could bootstrap the path.
2. PYTHONPATH=/opt/dagster/app — assumed the variable reached that
   process.

Neither worked, and both left the image depending on path luck.

The fix

The build template already runs dagster_cloud_post_install.sh after the
repository is copied and the requirements installed. Using it to pip install --no-deps . puts db, domain, services, core, and schemas in
site-packages, where they resolve regardless of working directory, PYTHONPATH,
or which process is importing.

--no-deps because the pinned, hashed requirements are already installed and
this must not resolve on top of them. (Appending . to requirements.txt
instead does not work — pip refuses a directory requirement alongside hashes.)

Verification

Built the image locally and ran under the failing condition — cwd /, no
PYTHONPATH:

db findable: True
domain findable: True
db from: /usr/local/lib/python3.13/site-packages/db/__init__.py
SUCCESS: transducer_observation 15.469751

Also

The sys.path insert is removed rather than kept as redundant insurance. It
encoded a theory that turned out to be wrong, and leaving it in would imply the
mechanism still matters. 42 tests pass; the test that imports db from an
unrelated working directory stays, since it is the behaviour that broke.

🤖 Generated with Claude Code

[github-actions]

Coverage

✅ 79.26% total — gate is 75%.

Coverage for the Python files changed in this PR

Name	Stmts	Miss	Cover	Missing
automated_ingestion/__init__.py	0	0	100%
TOTAL	0	0	100%

[github-actions]

Your pull request is automatically being deployed to Dagster Cloud.

Location	Status	Link	Updated
ocotillo-automated-ingestion		View in Cloud	Aug 19, 2026 at 03:18 AM (UTC)