qwak-sdk 0.1.0__py3-none-any.whl → 0.2.20rc0__py3-none-any.whl

qwak_sdk/commands/models/deployments/deploy/realtime/_logic/serving_strategy_mapper.py

@@ -0,0 +1,105 @@
+from __future__ import annotations
+
+from typing import Dict, List
+
+from _qwak_proto.qwak.auto_scaling.v1.auto_scaling_pb2 import AutoScalingConfig
+from _qwak_proto.qwak.deployment.deployment_pb2 import (
+    RealTimeConfig,
+    ServingStrategy,
+    TrafficConfig,
+    Variation,
+)
+from _qwak_proto.qwak.deployment.deployment_service_pb2 import (
+    GetDeploymentStatusResponse,
+)
+from _qwak_proto.qwak.ecosystem.v0.ecosystem_pb2 import UserContextEnvironmentDetails
+from qwak.exceptions import QwakException
+
+from qwak_sdk.commands.audience._logic.config.v1.audience_config import AudienceConfig
+from qwak_sdk.commands.models._logic.variations import (
+    create_variation_from_variation_config,
+)
+from qwak_sdk.commands.models.deployments.deploy._logic.deploy_config import (
+    DeployConfig,
+)
+from qwak_sdk.commands.models.deployments.deploy._logic.variations import (
+    get_variations_for_deploy,
+)
+
+
+def create_realtime_serving_strategy(
+    auto_scaling: AutoScalingConfig,
+    audiences: List[AudienceConfig],
+    fallback_variation: str,
+    variation_name: str,
+    variations: List[Variation],
+) -> ServingStrategy:
+    return ServingStrategy(
+        realtime_config=(
+            RealTimeConfig(
+                auto_scaling_config=auto_scaling,
+                traffic_config=TrafficConfig(
+                    selected_variation_name=variation_name,
+                    variations=variations,
+                    audience_routes_entries=[
+                        audience.to_audience_route_entry(index)
+                        for index, audience in enumerate(audiences)
+                    ],
+                    fallback_variation=fallback_variation,
+                ),
+            )
+        )
+    )
+
+
+def create_realtime_serving_strategy_from_deploy_config(
+    deploy_config: DeployConfig,
+    model_traffic: GetDeploymentStatusResponse,
+    environment_name_to_config: Dict[str, UserContextEnvironmentDetails],
+) -> Dict[str, ServingStrategy]:
+    serving_strategies = {}
+    errors = []
+    variation_name = deploy_config.realtime.variation_name or "default"
+    fallback_variation = deploy_config.realtime.fallback_variation
+    auto_scaling = (
+        deploy_config.auto_scaling.to_autoscaling_api()
+        if deploy_config.auto_scaling
+        else None
+    )
+    for env_name, env_config in environment_name_to_config.items():
+        env_variations_response = model_traffic.environment_to_model_traffic.get(
+            env_config.id
+        )
+        existing_variation = (
+            env_variations_response.variations if env_variations_response else []
+        )
+        try:
+            variations = []
+            if not deploy_config.realtime.audiences:
+                variations = get_variations_for_deploy(
+                    variation_name=variation_name,
+                    existing_variations=existing_variation,
+                    requested_variations=list(
+                        map(
+                            create_variation_from_variation_config,
+                            deploy_config.realtime.variations,
+                        )
+                    ),
+                    environment_name=env_name,
+                )
+
+            serving_strategies[env_config.id] = create_realtime_serving_strategy(
+                auto_scaling,
+                deploy_config.realtime.audiences,
+                fallback_variation,
+                variation_name,
+                variations,
+            )
+
+        except QwakException as e:
+            errors.append(e.message)
+
+    if errors:
+        raise QwakException("\n".join(errors))
+
+    return serving_strategies

qwak_sdk/commands/models/deployments/deploy/realtime/ui.py

@@ -0,0 +1,179 @@
+from typing import Optional
+
+import click
+from qwak.inner.const import QwakConstants
+
+from qwak_sdk.commands.models.deployments.deploy._logic.deploy_config import (
+    DeployConfig,
+    PurchaseOption,
+)
+from qwak_sdk.commands.models.deployments.deploy._logic.deployment_response_handler import (
+    client_deployment,
+)
+from qwak_sdk.commands.models.deployments.deploy.realtime._logic.deploy_executor import (
+    RealtimeDeployExecutor,
+)
+from qwak_sdk.inner.tools.cli_tools import QwakCommand
+from qwak_sdk.inner.tools.config_handler import config_handler
+from qwak_sdk.inner.tools.logger.logger import (
+    get_qwak_logger,
+    set_qwak_logger_stdout_verbosity_level,
+)
+
+logger = get_qwak_logger()
+PENDING_STATUSES = [
+    "INITIATING_DEPLOYMENT",
+    "PENDING_DEPLOYMENT",
+    "INITIATING_UNDEPLOYMENT",
+    "PENDING_UNDEPLOYMENT",
+]
+DEPLOYED_STATUSES = ["SUCCESSFUL_DEPLOYMENT"]
+UNDEPLOYED_STATUSES = ["SUCCESSFUL_UNDEPLOYMENT"]
+FAILED_STATUSES = [
+    "INVALID_DEPLOYMENT",
+    "FAILED_INITIATING_DEPLOYMENT",
+    "FAILED_DEPLOYMENT",
+    "FAILED_UNDEPLOYMENT",
+]
+
+
+@click.command("realtime", help="Deploy realtime model.", cls=QwakCommand)
+@click.option(
+    "--build-id",
+    help="Build ID, If not specified latest successful build will deployed.",
+    type=str,
+)
+@click.option("--model-id", help="Model ID", type=str, required=True)
+@click.option(
+    "--timeout",
+    type=int,
+    help="Inference request timeout in MS, (e.g. 2000).",
+)
+@click.option(
+    "--server-workers",
+    type=int,
+    help="Number of workers of http server. (e.g. 4).",
+)
+@click.option(
+    "--daemon-mode/--no-daemon-mode",
+    type=bool,
+    default=True,
+    help="Configure Gunicorn daemon mode. Default is True",
+)
+@click.option(
+    "--pods",
+    type=int,
+    help="Number of pods to deploy.",
+)
+@click.option(
+    "--cpus",
+    type=float,
+    help="Number of CPU cores, can be fraction of a core (e.g. 0.4).",
+)
+@click.option(
+    "--memory",
+    type=int,
+    help="Memory amount in Mib (e.g. 256).",
+)
+@click.option(
+    "--gpu-type",
+    metavar="NAME",
+    required=False,
+    help=f"Type of GPU to use on the deployments ({', '.join([x for x in QwakConstants.GPU_TYPES])}).",
+    type=click.STRING,
+)
+@click.option(
+    "--gpu-amount",
+    metavar="NAME",
+    required=False,
+    type=int,
+    help="Amount of GPU to use on the deployments.",
+)
+@click.option(
+    "--iam-role-arn",
+    type=str,
+    help="Custom IAM Role ARN.",
+)
+@click.option(
+    "--max-batch-size",
+    type=int,
+    help="Max batch size in prediction. (defaults to 0 which is dynamic)",
+)
+@click.option(
+    "--variation-name",
+    required=False,
+    type=str,
+    help="The variation name",
+)
+@click.option(
+    "--environment-name",
+    required=False,
+    type=str,
+    help="Environment to deploy on (if not specified uses your default environment",
+    multiple=True,
+)
+@click.option(
+    "--sync",
+    is_flag=True,
+    default=False,
+    help="Waiting for deployments to be ready",
+)
+@click.option(
+    "-v",
+    "--verbose",
+    count=True,
+    help="Log verbosity level - v: INFO, vv: DEBUG [default: WARNING]",
+)
+@click.option(
+    "-f",
+    "--from-file",
+    help="Deploy by run_config file, Command arguments will overwrite any run_config.",
+    required=False,
+    type=click.Path(exists=True, resolve_path=True, dir_okay=False),
+)
+@click.option(
+    "--out-conf",
+    help="Extract deploy conf from commands arguments, the commands will not run it wil only output valid yaml "
+    "structure",
+    default=False,
+    is_flag=True,
+)
+@click.option(
+    "--purchase-option",
+    required=False,
+    type=click.Choice([i.value for i in PurchaseOption], case_sensitive=False),
+    help="Indicate the instance deployments type, whether it is spot/ondemand. Default is spot",
+)
+@click.option(
+    "--deployments-timeout",
+    type=int,
+    help="The number of seconds the deployments can be in progress before it is considered as failed. This is useful "
+    "in cases where a deployments has a high number of pods to replace, scarce resources (gpu), or large build "
+    "size. Default is 600 seconds (10 minutes).",
+)
+@click.option(
+    "-E",
+    "--env-vars",
+    required=False,
+    metavar="NAME=VALUE",
+    multiple=True,
+    help="A parameter for the deployments, of the form -E name=value",
+)
+def realtime(
+    verbose: bool, from_file: Optional[str], out_conf: bool, sync: bool, **kwargs
+):
+    set_qwak_logger_stdout_verbosity_level(verbose + 1)
+    deploy_realtime(from_file, out_conf, sync, **kwargs)
+
+
+def deploy_realtime(from_file: Optional[str], out_conf: bool, sync: bool, **kwargs):
+    config: DeployConfig = config_handler(
+        config=DeployConfig,
+        from_file=from_file,
+        out_conf=out_conf,
+        sections=("realtime", "autoscaling"),
+        **kwargs,
+    )
+    if not out_conf:
+        deploy_executor = RealtimeDeployExecutor(config)
+        client_deployment(deploy=deploy_executor, sync=sync)

qwak_sdk/commands/models/deployments/deploy/streaming/_logic/deploy_executor.py

@@ -0,0 +1,24 @@
+from _qwak_proto.qwak.deployment.deployment_pb2 import KubeDeploymentType
+from _qwak_proto.qwak.deployment.deployment_service_pb2 import DeployModelResponse
+
+from qwak_sdk.commands.models.deployments.deploy._logic.base_deploy_executor import (
+    BaseDeployExecutor,
+)
+from qwak_sdk.commands.models.deployments.deploy._logic.deployment_message_helpers import (
+    get_env_to_deployment_message,
+)
+
+
+class StreamDeployExecutor(BaseDeployExecutor):
+    def deploy(self) -> DeployModelResponse:
+        env_deployment_messages = get_env_to_deployment_message(
+            self.config,
+            KubeDeploymentType.STREAM,
+            self.deploy_client,
+            self.ecosystem_client,
+        )
+        return self.deploy_client.deploy_model(
+            model_id=self.config.model_id,
+            build_id=self.config.build_id,
+            env_deployment_messages=env_deployment_messages,
+        )

qwak_sdk/commands/models/deployments/deploy/streaming/_logic/serving_strategy_mapper.py

@@ -0,0 +1,38 @@
+from __future__ import annotations
+
+from _qwak_proto.qwak.deployment.deployment_pb2 import (
+    KafkaConfig,
+    ServingStrategy,
+    StreamConfig,
+)
+
+from qwak_sdk.commands.models.deployments.deploy._logic.deploy_config import (
+    DeployConfig,
+)
+
+DEFAULT_VARIATION_NAME = "default"
+
+
+def create_streaming_serving_strategy_from_deploy_config(
+    deploy_config: DeployConfig,
+) -> ServingStrategy:
+    return ServingStrategy(
+        stream_config=StreamConfig(
+            kafka=KafkaConfig(
+                consumer=KafkaConfig.Consumer(
+                    bootstrap_server=deploy_config.stream.kafka.consumer.bootstrap_servers,
+                    topic=deploy_config.stream.kafka.consumer.topic,
+                    group=deploy_config.stream.kafka.consumer.group,
+                    timeout=deploy_config.stream.kafka.consumer.timeout,
+                    auto_offset_type=deploy_config.stream.kafka.consumer.auto_offset_reset.upper(),
+                    max_batch_size=deploy_config.stream.kafka.consumer.max_batch_size,
+                    max_poll_latency=deploy_config.stream.kafka.consumer.max_poll_latency,
+                ),
+                producer=KafkaConfig.Producer(
+                    bootstrap_server=deploy_config.stream.kafka.producer.bootstrap_servers,
+                    topic=deploy_config.stream.kafka.producer.topic,
+                    compression_type=deploy_config.stream.kafka.producer.compression_type.upper(),
+                ),
+            )
+        )
+    )

qwak_sdk/commands/models/deployments/deploy/streaming/ui.py

@@ -0,0 +1,196 @@
+import click
+
+from qwak_sdk.commands.models.deployments.deploy._logic.deploy_config import (
+    AutoOffsetReset,
+    CompressionTypes,
+    DeployConfig,
+    PurchaseOption,
+)
+from qwak_sdk.commands.models.deployments.deploy._logic.deployment_response_handler import (
+    client_deployment,
+)
+from qwak_sdk.commands.models.deployments.deploy.streaming._logic.deploy_executor import (
+    StreamDeployExecutor,
+)
+from qwak_sdk.inner.tools.cli_tools import QwakCommand
+from qwak_sdk.inner.tools.config_handler import config_handler
+from qwak_sdk.inner.tools.logger.logger import (
+    get_qwak_logger,
+    set_qwak_logger_stdout_verbosity_level,
+)
+
+logger = get_qwak_logger()
+PENDING_STATUSES = [
+    "INITIATING_DEPLOYMENT",
+    "PENDING_DEPLOYMENT",
+    "INITIATING_UNDEPLOYMENT",
+    "PENDING_UNDEPLOYMENT",
+]
+DEPLOYED_STATUSES = ["SUCCESSFUL_DEPLOYMENT"]
+UNDEPLOYED_STATUSES = ["SUCCESSFUL_UNDEPLOYMENT"]
+FAILED_STATUSES = [
+    "INVALID_DEPLOYMENT",
+    "FAILED_INITIATING_DEPLOYMENT",
+    "FAILED_DEPLOYMENT",
+    "FAILED_UNDEPLOYMENT",
+]
+
+
+@click.command("stream", help="Deploy stream model.", cls=QwakCommand)
+@click.option(
+    "--build-id",
+    help="Build ID, If not specified latest successful build will deployed.",
+    type=str,
+)
+@click.option("--model-id", help="Model ID", type=str, required=True)
+@click.option("--branch", help="Branch ID", type=str)
+@click.option(
+    "--bootstrap-server",
+    envvar="BOOTSTRAP_SERVERS",
+    help="Kafka consumer/producer bootstrap server.",
+    multiple=True,
+    type=str,
+)
+@click.option(
+    "--consumer-bootstrap-server",
+    envvar="CONSUMER_BOOTSTRAP_SERVER",
+    help="Kafka consumer bootstrap server.",
+    multiple=True,
+    type=str,
+)
+@click.option(
+    "--consumer-topic",
+    envvar="CONSUMER_TOPIC",
+    help="Kafka consumer topic.",
+    type=str,
+)
+@click.option(
+    "--consumer-group",
+    envvar="CONSUMER_GROUP",
+    help="Kafka consumer group.",
+    type=str,
+)
+@click.option(
+    "--consumer-auto-offset-reset",
+    envvar="CONSUMER_AUTO_OFFSET_RESET",
+    type=click.Choice([i.value for i in AutoOffsetReset], case_sensitive=False),
+    help="Kafka consumer auto offset reset (Latest/Earliest).",
+)
+@click.option(
+    "--consumer-timeout",
+    envvar="CONSUMER_TIMEOUT",
+    help="Kafka consumer polling timeout. (The timeout should be in range of kafka admin configuration "
+    "'group.min.session.timeout.ms' and 'group.max.session.timeout.ms'.)",
+    type=int,
+)
+@click.option(
+    "--consumer-max-batch-size",
+    envvar="CONSUMER_MAX_BATCH_SIZE",
+    help="The maximum number of records returned in a single call to poll().",
+    type=int,
+)
+@click.option(
+    "--consumer-max-poll-latency",
+    envvar="CONSUMER_MAX_POLL_LATENCY",
+    help="The maximum delay between invocations of poll() when using consumer group management.",
+    type=float,
+)
+@click.option(
+    "--producer-bootstrap-server",
+    envvar="PRODUCER_BOOTSTRAP_SERVER",
+    help="Kafka producer bootstrap server.",
+    multiple=True,
+    type=str,
+)
+@click.option(
+    "--producer-topic",
+    envvar="PRODUCER_TOPIC",
+    help="Kafka producer topic.",
+    type=str,
+)
+@click.option(
+    "--producer-compression-type",
+    envvar="PRODUCER_COMPRESSION_TYPE",
+    type=click.Choice([i.value for i in CompressionTypes], case_sensitive=False),
+    help="Kafka producer compression type.",
+)
+@click.option(
+    "--pods",
+    type=int,
+    help="Number of pods to deploy.",
+)
+@click.option(
+    "--cpus",
+    type=float,
+    help="Number of CPU cores, can be fraction of a core (e.g. 0.4).",
+)
+@click.option(
+    "--memory",
+    type=int,
+    help="Memory amount in Mib (e.g. 256).",
+)
+@click.option(
+    "--iam-role-arn",
+    type=str,
+    help="Custom IAM Role ARN.",
+)
+@click.option(
+    "--sync",
+    is_flag=True,
+    default=False,
+    help="Waiting for deployments to be ready",
+)
+@click.option(
+    "--local",
+    is_flag=True,
+    default=False,
+    help="Run deployments locally.",
+)
+@click.option(
+    "-v",
+    "--verbose",
+    count=True,
+    help="Log verbosity level - v: INFO, vv: DEBUG [default: WARNING]",
+)
+@click.option(
+    "-f",
+    "--from-file",
+    help="Deploy by run_config file, Command arguments will overwrite any run_config.",
+    required=False,
+    type=click.Path(exists=True, resolve_path=True, dir_okay=False),
+)
+@click.option(
+    "--out-conf",
+    help="Extract deploy conf from commands arguments, the commands will not run it wil only output valid yaml "
+    "structure",
+    default=False,
+    is_flag=True,
+)
+@click.option(
+    "--purchase-option",
+    required=False,
+    type=click.Choice([i.value for i in PurchaseOption], case_sensitive=False),
+    help="Indicate the instance deployments type, whether it is spot/ondemand.  Default is spot",
+)
+@click.option(
+    "-E",
+    "--env-vars",
+    required=False,
+    metavar="NAME=VALUE",
+    multiple=True,
+    help="A parameter for the deployments, of the form -E name=value",
+)
+def stream(
+    verbose: bool, from_file: str, out_conf: bool, local: bool, sync: bool, **kwargs
+):
+    set_qwak_logger_stdout_verbosity_level(verbose + 1)
+    config: DeployConfig = config_handler(
+        config=DeployConfig,
+        from_file=from_file,
+        out_conf=out_conf,
+        sections=("stream",),
+        **kwargs,
+    )
+    if not out_conf:
+        deploy_executor = StreamDeployExecutor(config)
+        client_deployment(deploy=deploy_executor, sync=sync)