server_gunicorn.py

#
# Copyright 2025 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#      http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""Launcher for the prediction_executor based encoder server.

Uses the servingframework to create a request server which
performs the logic for requests in separate processes and uses a local TFserving
instance to handle the model.
"""

from collections.abc import Sequence
import os

from absl import app
from absl import logging
import jsonschema
import yaml

from serving.serving_framework import inline_prediction_executor
from serving.serving_framework import server_gunicorn
from serving.serving_framework.triton import server_health_check
from serving.serving_framework.triton import triton_server_model_runner
from serving import predictor


def main(argv: Sequence[str]) -> None:
  if len(argv) > 1:
    raise app.UsageError('Too many command-line arguments.')
  if 'AIP_HTTP_PORT' not in os.environ:
    raise ValueError(
        'The environment variable AIP_HTTP_PORT needs to be specified.'
    )
  http_port = int(os.environ.get('AIP_HTTP_PORT'))
  options = {
      'bind': f'0.0.0.0:{http_port}',
      'workers': 3,
      'timeout': 120,
  }
  model_rest_port = int(os.environ.get('MODEL_REST_PORT'))
  health_checker = server_health_check.TritonServerHealthCheck(model_rest_port)
  # Get schema validators.
  local_path = os.path.dirname(__file__)
  with open(
      os.path.join(local_path, 'vertex_schemata', 'instance.yaml'), 'r'
  ) as f:
    instance_validator = jsonschema.Draft202012Validator(yaml.safe_load(f))
  # with open(
  #     os.path.join(local_path, 'vertex_schemata', 'prediction.yaml'), 'r'
  # ) as f:
  #   prediction_validator = jsonschema.Draft202012Validator(yaml.safe_load(f))
  prediction_validator = None  # TODO(b/430426684): Restore validation.
  predictor_instance = predictor.MedSiglipPredictor()
  logging.info('Launching gunicorn application.')
  server_gunicorn.PredictionApplication(
      inline_prediction_executor.InlinePredictionExecutor(
          predictor_instance.predict,
          triton_server_model_runner.TritonServerModelRunner,
      ),
      health_check=health_checker,
      options=options,
      input_validator=instance_validator,
      prediction_validator=prediction_validator,
  ).run()


if __name__ == '__main__':
  app.run(main)