Configuration

These workflows assume the app is installed and running. See Quick Start if you haven’t set up yet.

Replace <app_name> with the name you chose at install.

Scheduled retraining

Keep encoders current as your data changes.

Weekly retraining:

CALL <app_name>.core.schedule_training('sales', 'USING CRON 0 2 * * 0 UTC');

Retrain every hour, but only when the source table has grown by at least 10,000 rows:

CALL <app_name>.core.schedule_training('sales', '1 HOUR', 10000);

Remove a schedule:

CALL <app_name>.core.unschedule_training('sales');

View active schedules:

SELECT * FROM <app_name>.core.training_schedules_v;

GPU training

Training runs on CPU by default. Switch to GPU for larger datasets. GPU availability depends on your region.

CALL <app_name>.core.set_training_compute('gpu');
CALL <app_name>.core.start_services();

Switch back to CPU:

CALL <app_name>.core.set_training_compute('cpu');
CALL <app_name>.core.start_services();

No active training job can be running during a mode switch.

Encoding compute

The encoding service runs on CPU_X64_XS by default with up to 3 replicas. For heavier workloads, use a larger instance family.

CALL <app_name>.core.set_encoding_instance('CPU_X64_M');
CALL <app_name>.core.start_services();

Options: CPU_X64_XS (default), CPU_X64_S, CPU_X64_M, CPU_X64_SL, CPU_X64_L, HIGHMEM_X64_S, HIGHMEM_X64_M.

Encoder versioning

Each time you retrain, the previous version is archived automatically.

View archived versions:

SELECT * FROM <app_name>.core.encoder_versions_v
WHERE encoder_name = 'sales';

Roll back to a previous version:

CALL <app_name>.core.rollback_encoder('sales', 1);

Remove old versions (keep the 3 most recent):

CALL <app_name>.core.prune_encoder_versions('sales', 3);

Service management

Start services (runs automatically at install, call manually after compute changes):

CALL <app_name>.core.start_services();

Wait for services to reach ready:

CALL <app_name>.core.wait_for_services();

Check service health:

SELECT * FROM <app_name>.core.health_v;

Detailed diagnostics with container logs:

CALL <app_name>.core.service_status();

Stop services (call before uninstalling):

CALL <app_name>.core.stop_services();