Set ASTRAPE_TCN_TRAINER_URL and the scheduler ships raw series to that
service and writes back the artifact it returns, instead of training in
process. This host has no AVX2 and torch dies with SIGILL in Adam, so remote
is the working path here; local training stays for hosts that can do it.
run() now picks between _train_local and _train_remote. The remote path never
builds examples locally -- it only reads the database -- and the far side
rebuilds them with the same gibil. The artifact is written last and
atomically, because the oracle daemon loads it on every forecast. A remote
failure raises SystemExit so the unit fails visibly rather than leaving
yesterday's model in place while reporting success. Events record trained_by,
trainer_url and job_id.
Also stops deriving the model, state and log paths from __file__. Where this
package is installed says nothing about where the deployment keeps them; they
now resolve against the working directory, which every unit sets.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The checkout is now astrape/astrape-oracle, not astrape/astrape-oracle/<package>. The package
contents move to the repository root and pyproject maps the package name to
'.' via package-dir, with the subpackages listed explicitly.
The systemd units move out to /srv/astrape/deploy/systemd, outside every
repository, because they describe this deployment rather than any one
component.
Verified both install modes: editable, which is how the deployment runs, and
a plain wheel install, which is what the container build does.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The repos now sit directly under /srv/astrape, which is also every unit's
WorkingDirectory. Python prepends the working directory to sys.path for
'python -m', so the gibil checkout at /srv/astrape/gibil was shadowing the
installed package: the repo root has no __init__.py, so 'gibil' resolved as a
namespace portion with __file__ None, and submodules only resolved because a
namespace path recomputes itself on access.
Everything is installed into the venv, so nothing needs the working directory
on sys.path. WorkingDirectory still matters for env/ and models/.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Each repo carries the units for the daemons it owns. WorkingDirectory stays
/srv/astrape so env/ and models/ resolve, while the code comes from the
editable installs under repos/.
The old astrape-db.service was misnamed -- it runs the weather forecast
ingest, and collided with what would have been an astrape-db repo. It is
astrape-weather-forecast.service here.
The training unit forces ATEN_CPU_CAPABILITY=default, because this host has no
AVX2 and torch otherwise dies with SIGILL in Adam, and caps threads with Nice
so a nightly run does not starve the 5-second Modbus poll.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The forecasting runtime gibil pilots: weather ingest (Open-Meteo
forecasts and the Ecowitt station), the oracle and weather stores, the
snapshot daemon, the forecast evaluator, and the TCN training
scheduler.
classes/datasets.py holds the shapes the stores return for display.
They sit beside the stores deliberately: a store's return type belongs
to the store, and keeping them with the renderers would mean this repo
importing the dashboard, which is backwards.
Extracted from a single repository where the forecasting engine, the
database plumbing, the hardware client, the ingest daemons and the web
UI all imported each other freely. Verified that every module imports
with gibil resolved from its own installed checkout.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
rpotter6298