Trigger Pipelines

Event-driven post-training workflows

Fire a webhook when a checkpoint or model lands. No polling required.

After a training job writes a checkpoint or final weights, downstream work usually follows: evaluation, quantization, conversion to a serving format, or deployment. Polling the bucket adds latency, wastes API calls, and complicates orchestration.

Tigris Object Notifications replace the polling loop with a push model. An HTTP POST fires the moment a new object lands, carrying the bucket, key, size, and ETag. Your pipeline handler starts immediately.

Object notifications →

Benefits

Instant pipeline triggers

No polling interval means no wasted time between a checkpoint landing and your pipeline starting. The webhook fires the moment the PutObject completes, so eval, conversion, or deployment kicks off immediately.

Prefix filtering

Filter to exactly the events you care about — for example, only objects under checkpoints/ or final/ — so your handler isn't invoked on intermediate artifacts it doesn't need.

WHERE `key` REGEXP "^final/" AND `Event-Type` = "OBJECT_CREATED_PUT"

Fan-out to multiple steps

A single webhook handler can fan out to parallel downstream tasks: run evals against the new checkpoint, kick off ONNX or TensorRT conversion, and trigger a rolling deploy to your inference fleet — all from one event.

Example: eval → convert → deploy

Configure a notification rule through the Tigris Dashboard, pointing at an HTTPS endpoint you control. When weights land at final/, your handler:

1. Runs evaluation benchmarks against the new model
2. Converts to optimized serving format (ONNX, TensorRT)
3. Triggers a rolling deploy to inference endpoints

note

Notifications are delivered at least once and can arrive out of order across regions. Use the Last-Modified timestamp on the object (not eventTime) to sequence events correctly, and design your handler to be idempotent.