Resources: compute requirements of Tasks. This class is immutable once created (to ensure some validations are done whenever properties chang

Mock PVC object.

One column of the displayed cluster table

Mock Pod object.

Base exception for Mithril API errors.

Base exception for Hyperbolic API errors.

The client to interact with a remote cluster through Skylet.

AWS CloudWatch logging agent. This agent forwards logs from SkyPilot clusters to AWS CloudWatch using Fluent Bit. It supports authentication

Managed job status, designed to be in serverless style. The ManagedJobStatus is a higher level status than the JobStatus. Each managed job su

Test policy server that runs in a background thread with automatic port assignment. Uses pre-bound sockets to avoid TOCTOU port races when multip

Amazon Web Services.

Mock cloud class for testing.

Base exception for Hyperbolic catalog errors.

Internal daemon that runs an event in the background.

An executor that creates a new thread for each task and destroys it after the task is completed. Note(dev): We raise an error instead of

A timer class that can be used as a context manager to measure execution time.

A wrapper around filelock.FileLock. This implements a distributed lock that works across multiple processes when they share the same filesyst

Middleware to handle authentication by delegating to OAuth2 Proxy.

A client for the Verda Cloud API.

An http client, a wrapper for the requests library. For each request, it adds the authentication header with an access token. If the access t

Backend: runs on cloud virtual machines, managed by Ray. Changing this class may also require updates to: * Cloud providers' templates unde

Captures the state of the job from the scheduler's perspective. A newly created job will be INACTIVE. The following transitions are valid: -

Mock retry context for testing resumable functionality.

A region.

Configuration for S3-compatible storage providers.

Manager for SSH Node Pool configurations.

A zone, typically grouped under a region.

Autoscaling decisions. |------------------------------------------------------------------------| | Operator | TargetType |

An multiprocessing executor that supports bursting worker processes.

Google Cloud Platform.

Job status enum.

Raised when a lock acquisition times out.

A reference to a secret in task YAML.

Filter for requests. Args: status: a list of statuses of the requests to filter on. cluster_names: a list of cluster names to fil

Raised when the Seeweb API key is missing or empty.

Minimalist provisioner for Seeweb ECS.

Information about the Slurm partitions.

Comprehensive scale tests for SkyPilot operations.

Proxy that makes a gRPC stub honor the current SkyPilotContext cancel. Each method becomes cancellable: when the active context is cancelled

Stand-in for the iterator returned by ``method(req, timeout=...)`` for unary-stream RPCs.

A context manager that releases the lock when exiting. This proxy is returned by acquire() and ensures proper cleanup when used in a with sta

A pickle-able handle to a cluster created by CloudVmRayBackend. The handle object will last for the whole lifecycle of the cluster. - (requi

Normalized ``enable_docker`` config produced by :func:`normalize_enable_docker_config`.

Example class for generating documentation content from a given path. Attributes: path (Path): The path to the main directory or fil

Information about a git clone.

Information extracted from a git URL.

Adds logging prefix to newlines to align multi-line messages.

Pod status.

A custom ProcessPoolExecutor with additional supports for skypilot. The additional supports include: 1. Disposable workers: support control w

Enum for the different types of stores.

A configuration for the Verda Cloud API.

Autostop configuration.

Middleware to handle HTTP Basic Auth.

This is a simplified wrapper around the Content Library APIs. It is used to access services exposed by Content Library Service.

SkyPilot config that supports setting/getting values with nested keys.

Main loop for a job controller process. Many jobs will be handled by this, each by a single JobController.

A simple wrapper that creates a new process for each task. This is a workaround for Python 3.10 since `max_tasks_per_child` of ProcessPoolExe

Modes for the ``enable_docker`` config.

Default image and volume names for a Docker sidecar mode.

A handler that awares environment variables. This handler dynamically reflects the log level from environment variables.

Configuration for external authentication proxy. Attributes: enabled: Whether external proxy authentication is enabled. header_na

Manages a vpc containing the network configuration for the cluster's nodes.

Infrastructure information parsed from infra string. When a field is None, it means the field is not specified.

Instance model class.

A collection of messages.

Information about a Slurm node from sinfo.

Handles automated Okta OAuth2 login flow using Selenium

Environment variables for SkyPilot.

PostgreSQL advisory lock implementation. Uses PostgreSQL advisory locks to implement distributed locking that works across multiple machines

Extended TestScale for production-scale data injection.

Custom SFTTrainer that tracks timing statistics.

Custom SFTTrainer that tracks timing statistics.

A single prompt with its ground truth answer.

A batch of prompts.