gigl.src.common.vertex_ai_launcher#

Shared functionality for launching Vertex AI jobs for training and inference.

Attributes#

logger

Functions#

`build_vertex_ai_job_name`(job_name, component)	Build the Vertex AI CustomJob display name from a raw GiGL job name.
`launch_graph_store_enabled_job`(...)	Launch a graph store enabled job on Vertex AI with separate storage and compute pools.
`launch_single_pool_job`(vertex_ai_resource_config, ...)	Launch a single pool job on Vertex AI.

Module Contents#

gigl.src.common.vertex_ai_launcher.build_vertex_ai_job_name(job_name, component)[source]#

Build the Vertex AI CustomJob display name from a raw GiGL job name.

Parameters:

job_name (str) – Raw GiGL applied task identifier.
component (gigl.src.common.constants.components.GiGLComponents) – The GiGL component being launched.

Returns:

The component-prefixed Vertex AI CustomJob display name.

Raises:

ValueError – If component is not a launchable Vertex AI component.

Return type:

str

gigl.src.common.vertex_ai_launcher.launch_graph_store_enabled_job(vertex_ai_graph_store_config, job_name, task_config_uri, resource_config_uri, compute_commmand, compute_runtime_args, storage_command, storage_args, resource_config_wrapper, cpu_docker_uri, cuda_docker_uri, component)[source]#

Launch a graph store enabled job on Vertex AI with separate storage and compute pools.

Parameters:

vertex_ai_graph_store_config (snapchat.research.gbml.gigl_resource_config_pb2.VertexAiGraphStoreConfig) – The Vertex AI graph store configuration
job_name (str) – Raw GiGL applied task identifier
task_config_uri (gigl.common.Uri) – URI to the task configuration
resource_config_uri (gigl.common.Uri) – URI to the resource configuration
compute_commmand (str) – Command to run in the compute container
compute_runtime_args (collections.abc.Mapping[str, str]) – Runtime arguments for the compute process
storage_command (str) – Command to run in the storage container
storage_args (collections.abc.Mapping[str, str]) – Arguments to pass to the storage command
resource_config_wrapper (gigl.src.common.types.pb_wrappers.gigl_resource_config.GiglResourceConfigWrapper) – Wrapper for the resource configuration
cpu_docker_uri (Optional[str]) – Docker image URI for CPU execution
cuda_docker_uri (Optional[str]) – Docker image URI for GPU execution
component (gigl.src.common.constants.components.GiGLComponents) – The GiGL component (Trainer or Inferencer)

Return type:

None

gigl.src.common.vertex_ai_launcher.launch_single_pool_job(vertex_ai_resource_config, job_name, task_config_uri, resource_config_uri, process_command, process_runtime_args, resource_config_wrapper, cpu_docker_uri, cuda_docker_uri, component, vertex_ai_region)[source]#

Launch a single pool job on Vertex AI.

Parameters:

vertex_ai_resource_config (snapchat.research.gbml.gigl_resource_config_pb2.VertexAiResourceConfig) – The Vertex AI resource configuration
job_name (str) – Raw GiGL applied task identifier
task_config_uri (gigl.common.Uri) – URI to the task configuration
resource_config_uri (gigl.common.Uri) – URI to the resource configuration
process_command (str) – Command to run in the container
process_runtime_args (collections.abc.Mapping[str, str]) – Runtime arguments for the process
resource_config_wrapper (gigl.src.common.types.pb_wrappers.gigl_resource_config.GiglResourceConfigWrapper) – Wrapper for the resource configuration
cpu_docker_uri (Optional[str]) – Docker image URI for CPU execution
cuda_docker_uri (Optional[str]) – Docker image URI for GPU execution
component (gigl.src.common.constants.components.GiGLComponents) – The GiGL component (Trainer or Inferencer)
vertex_ai_region (str) – The Vertex AI region to launch the job in

Return type:

None

gigl.src.common.vertex_ai_launcher.logger[source]#