skypilot-nightly 1.0.0.dev20250502__py3-none-any.whl → 1.0.0.dev20251203__py3-none-any.whl

sky/catalog/data_fetchers/fetch_hyperbolic.py

@@ -0,0 +1,136 @@
+"""Script to fetch Hyperbolic instance data and generate catalog."""
+import argparse
+import csv
+import json
+import os
+import sys
+from typing import Any, Dict
+
+import requests
+
+ENDPOINT = 'https://api.hyperbolic.xyz/v2/skypilot/catalog'
+API_KEY_PATH = os.path.expanduser('~/.hyperbolic/api_key')
+
+REQUIRED_FIELDS = [
+    'InstanceType', 'AcceleratorName', 'AcceleratorCount', 'vCPUs', 'MemoryGiB',
+    'StorageGiB', 'Price', 'Region', 'GpuInfo', 'SpotPrice'
+]
+
+
+class HyperbolicCatalogError(Exception):
+    """Base exception for Hyperbolic catalog errors."""
+    pass
+
+
+def get_api_key(api_key=None) -> str:
+    """Get API key from arg, env var, or file."""
+    if api_key:
+        return api_key
+    if api_key := os.environ.get('HYPERBOLIC_API_KEY'):
+        return api_key
+    try:
+        with open(API_KEY_PATH, 'r', encoding='utf-8') as f:
+            return f.read().strip()
+    except FileNotFoundError as exc:
+        raise HyperbolicCatalogError(
+            'No API key found. Please either:\n'
+            '1. Pass --api-key\n'
+            '2. Set HYPERBOLIC_API_KEY environment variable\n'
+            '3. Create ~/.hyperbolic/api_key file') from exc
+
+
+def get_output_path() -> str:
+    """Get output path for catalog file."""
+    current_dir = os.getcwd()
+    if os.path.basename(current_dir) == 'hyperbolic':
+        return 'vms.csv'
+    hyperbolic_dir = os.path.join(current_dir, 'hyperbolic')
+    os.makedirs(hyperbolic_dir, exist_ok=True)
+    return os.path.join(hyperbolic_dir, 'vms.csv')
+
+
+def validate_instance_data(instance: Dict[str, Any]) -> None:
+    """Validate instance data has all required fields."""
+    missing_fields = [
+        field for field in REQUIRED_FIELDS if field not in instance
+    ]
+    if missing_fields:
+        raise HyperbolicCatalogError(
+            f'Instance data missing required fields: {missing_fields}')
+
+
+def create_catalog(api_key=None) -> None:
+    """Generate Hyperbolic catalog CSV file."""
+    try:
+        response = requests.get(
+            ENDPOINT,
+            headers={'Authorization': f'Bearer {get_api_key(api_key)}'},
+            timeout=30)
+        response.raise_for_status()
+
+        try:
+            data = response.json()
+        except json.JSONDecodeError as e:
+            raise HyperbolicCatalogError(
+                f'Invalid JSON response from API: {response.text}') from e
+
+        if 'vms' not in data:
+            raise HyperbolicCatalogError(
+                f'Missing "vms" field in API response: {data}')
+
+        instances = data['vms']
+        if not isinstance(instances, list):
+            raise HyperbolicCatalogError(
+                f'Expected list of instances, got {type(instances)}')
+
+        if not instances:
+            raise HyperbolicCatalogError('No instances found in API response')
+
+        # Validate each instance
+        for instance in instances:
+            validate_instance_data(instance)
+
+    except requests.exceptions.RequestException as e:
+        raise HyperbolicCatalogError(
+            f'Failed to fetch instance data: {e}') from e
+
+    output_path = get_output_path()
+    try:
+        with open(output_path, 'w', newline='', encoding='utf-8') as f:
+            writer = csv.DictWriter(f, fieldnames=REQUIRED_FIELDS)
+            writer.writeheader()
+
+            for instance in instances:
+                entry = instance.copy()
+                # Convert GpuInfo to string format
+                entry['GpuInfo'] = json.dumps(entry['GpuInfo'],
+                                              ensure_ascii=False).replace(
+                                                  '"', "'")  # pylint: disable=invalid-string-quote
+                writer.writerow(entry)
+    except (IOError, OSError) as e:
+        raise HyperbolicCatalogError(
+            f'Failed to write catalog file to {output_path}: {e}') from e
+
+
+def main() -> int:
+    """Main entry point."""
+    parser = argparse.ArgumentParser(
+        description='Fetch Hyperbolic instance data')
+    parser.add_argument('--api-key', help='Hyperbolic API key')
+    args = parser.parse_args()
+
+    try:
+        create_catalog(args.api_key)
+        print(f'Hyperbolic Service Catalog saved to {get_output_path()}')
+        return 0
+    except HyperbolicCatalogError as e:
+        print(f'Error: {e}', file=sys.stderr)
+        return 1
+    except (requests.exceptions.RequestException, json.JSONDecodeError, IOError,
+            OSError) as e:
+        print(f'Unexpected error: {e}', file=sys.stderr)
+        return 1
+
+
+if __name__ == '__main__':
+    sys.exit(main())

sky/{clouds/service_catalog → catalog}/data_fetchers/fetch_lambda_cloud.py

@@ -49,6 +49,7 @@ GPU_TO_MEMORY = {
     'V100': 16384,
     'H100': 81920,
     'GH200': 98304,
+    'B200': 184320,  # 180 GB
     'GENERAL': None
 }
 

sky/catalog/data_fetchers/fetch_nebius.py

@@ -0,0 +1,338 @@
+"""A script that queries Nebius API to get instance types and pricing info.
+
+This script takes about 1 minute to finish.
+"""
+import csv
+from dataclasses import dataclass
+import decimal
+import json
+import logging
+import os
+import re
+from typing import Any, Dict, List, Optional
+
+from sky.adaptors import nebius
+from sky.adaptors.nebius import billing
+from sky.adaptors.nebius import compute
+from sky.adaptors.nebius import iam
+from sky.adaptors.nebius import nebius_common
+
+logger = logging.getLogger(__name__)
+
+TIMEOUT = 10
+PARENT_ID_TEMPLATE = 'project-{}public-images'
+ACCELERATOR_MANUFACTURER = 'NVIDIA'
+
+
+@dataclass
+class PresetInfo:
+    """Represents information about a specific compute preset,
+    including its pricing.
+
+    Attributes:
+        region (str): The geographical region where the preset is available.
+        fullname (str): The full name of the preset, a combination of platform
+            and preset name.
+        name (str): The name of the preset.
+        platform_name (str): The name of the platform the preset belongs to.
+        gpu (int): The number of GPUs in the preset.
+        vcpu (int): The number of virtual CPUs in the preset.
+        gpu_memory_gibibytes (int): size of gpu memory in GiB.
+        memory_gib (int): The amount of memory in GiB in the preset.
+        accelerator_manufacturer (str | None): The manufacturer of the
+            accelerator (e.g., "NVIDIA"), or None if no accelerator.
+        accelerator_name (str | None): The name of the accelerator
+            (e.g., "H100"), or None if no accelerator.
+        price_hourly (decimal.Decimal): The hourly price of the preset.
+        spot_price (decimal.Decimal): The spot (preemptible) price
+            of the preset.
+    """
+
+    region: str
+    fullname: str
+    name: str
+    platform_name: str
+    gpu: int
+    vcpu: int
+    gpu_memory_gibibytes: int
+    memory_gib: int
+    accelerator_manufacturer: Optional[str]
+    accelerator_name: Optional[str]
+    price_hourly: decimal.Decimal
+    spot_price: decimal.Decimal
+
+
+def _format_decimal(value: decimal.Decimal) -> str:
+    """Formats a decimal value to a string with at least two decimal places,
+    removing trailing zeros and ensuring a two-digit decimal part.
+
+    Args:
+        value (decimal.Decimal): The decimal value to format.
+
+    Returns:
+        str: The formatted string representation of the decimal.
+    """
+    formatted_value = f'{value:f}'
+    integer_part, decimal_part = formatted_value.split(
+        '.') if '.' in formatted_value else (formatted_value, '')
+    if len(decimal_part) < 2:
+        decimal_part += '0' * (2 - len(decimal_part))
+
+    return f'{integer_part}.{decimal_part}'
+
+
+def _estimate_platforms(platforms: List[Any], parent_id: str,
+                        region: str) -> List[PresetInfo]:
+    """Collects specifications for all presets on the given platforms to form a
+    batch price request. It then sends the request and processes the responses
+    to create a list of PresetInfo objects.
+
+    Args:
+        platforms (List[Platform]): A List of compute platforms to estimate
+        prices for.
+        parent_id (str): The parent ID used for resource metadata
+        in the estimate request.
+        region (str): The region associated with the platforms.
+
+    Returns:
+        List[PresetInfo]: A list of PresetInfo objects containing details and
+        estimated prices for each preset.
+    """
+
+    calculator_service = billing().CalculatorServiceClient(nebius.sdk())
+    futures = []
+
+    for platform in platforms:
+        platform_name = platform.metadata.name
+
+        for preset in platform.spec.presets:
+            # Form the specification for the price request
+            estimate_spec = billing().ResourceSpec(
+                compute_instance_spec=compute().CreateInstanceRequest(
+                    metadata=nebius_common().ResourceMetadata(
+                        parent_id=parent_id,),
+                    spec=compute().InstanceSpec(
+                        resources=compute().ResourcesSpec(
+                            platform=platform_name,
+                            preset=preset.name,
+                        )),
+                ))
+            price_request = billing().EstimateBatchRequest(
+                resource_specs=[estimate_spec])
+
+            # Form the specification for the spot price request
+            spot_estimate_spec = billing().ResourceSpec(
+                compute_instance_spec=compute().CreateInstanceRequest(
+                    metadata=nebius_common().ResourceMetadata(
+                        parent_id=parent_id,),
+                    spec=compute().InstanceSpec(
+                        resources=compute().ResourcesSpec(
+                            platform=platform_name,
+                            preset=preset.name,
+                        ),
+                        preemptible=compute().PreemptibleSpec(priority=1),
+                    ),
+                ))
+            spot_price_request = billing().EstimateBatchRequest(
+                resource_specs=[spot_estimate_spec])
+
+            # Start future for each preset
+            futures.append((
+                platform,
+                preset,
+                calculator_service.estimate_batch(price_request,
+                                                  timeout=TIMEOUT),
+                calculator_service.estimate_batch(spot_price_request,
+                                                  timeout=TIMEOUT),
+            ))
+
+    # wait all futures to complete and collect results
+    result = []
+    for platform, preset, future, future_spot in futures:
+        platform_name = platform.metadata.name
+        result.append(
+            PresetInfo(
+                region=region,
+                fullname=f'{platform_name}_{preset.name}',
+                name=preset.name,
+                platform_name=platform_name,
+                gpu=preset.resources.gpu_count or 0,
+                vcpu=preset.resources.vcpu_count,
+                gpu_memory_gibibytes=platform.spec.gpu_memory_gibibytes,
+                memory_gib=preset.resources.memory_gibibytes,
+                accelerator_manufacturer=ACCELERATOR_MANUFACTURER
+                if platform_name.startswith('gpu-') else '',
+                accelerator_name=platform_name.split('-')[1].upper()
+                if platform_name.startswith('gpu-') else '',
+                price_hourly=decimal.Decimal(
+                    future.wait().hourly_cost.general.total.cost),
+                spot_price=decimal.Decimal(
+                    future_spot.wait().hourly_cost.general.total.cost),
+            ))
+
+    return result
+
+
+def _write_preset_prices(presets: List[PresetInfo], output_file: str) -> None:
+    """Writes the provided preset information to a CSV file.
+
+    Args:
+        presets (List[PresetInfo]): A list of PresetInfo objects to write.
+        output_file (str): The path to the output CSV file.
+    """
+    os.makedirs(os.path.dirname(output_file))
+    # Set up the CSV writer to output to stdout
+    with open(output_file, 'w', encoding='utf-8') as out:
+        header = [
+            'InstanceType',
+            'AcceleratorName',
+            'AcceleratorCount',
+            'vCPUs',
+            'MemoryGiB',
+            'Price',
+            'Region',
+            'GpuInfo',
+            'SpotPrice',
+        ]
+        writer = csv.DictWriter(out, fieldnames=header)
+        writer.writeheader()
+        # logger.info(presets)
+        for preset in sorted(presets,
+                             key=lambda x:
+                             (bool(x.gpu), x.region, x.platform_name, x.vcpu)):
+            gpu_info = ''
+            if preset.gpu > 0 and preset.accelerator_name:
+                vram = preset.gpu_memory_gibibytes * 1024
+                gpu_info_dict = {
+                    'Gpus': [{
+                        'Name': preset.accelerator_name,
+                        'Manufacturer': preset.accelerator_manufacturer,
+                        'Count': preset.gpu,
+                        'MemoryInfo': {
+                            'SizeInMiB': vram
+                        },
+                    }],
+                    'TotalGpuMemoryInMiB': vram * preset.gpu,
+                }
+                gpu_info = json.dumps(gpu_info_dict).replace('"', '\'')
+
+            writer.writerow({
+                'InstanceType': preset.fullname,
+                'AcceleratorName': preset.accelerator_name,
+                'AcceleratorCount': preset.gpu,
+                'vCPUs': preset.vcpu,
+                'MemoryGiB': preset.memory_gib,
+                'Price': _format_decimal(preset.price_hourly),
+                'Region': preset.region,
+                'GpuInfo': gpu_info,
+                'SpotPrice': _format_decimal(preset.spot_price)
+                             if preset.spot_price else '',
+            })
+
+
+def _fetch_platforms_for_project(project_id: str) -> List[Any]:
+    """Fetches all available compute platforms for a given project.
+
+    Args:
+        project_id (str): The ID of the project to fetch platforms from.
+
+    Returns:
+        List[ComputePlatform]: A list of ComputePlatform objects available
+        in the project.
+    """
+    platform_service = compute().PlatformServiceClient(nebius.sdk())
+
+    platform_request = compute().ListPlatformsRequest(page_size=999,
+                                                      parent_id=project_id)
+    platform_response = platform_service.list(platform_request,
+                                              timeout=TIMEOUT).wait()
+
+    return platform_response.items
+
+
+def _get_regions_map() -> Dict[str, str]:
+    """Maps region codes to their full names by iterating through tenants and
+     projects.
+
+    Returns:
+        dict[str, str]: A dictionary where keys are region codes (e.g., "e00")
+                        and values are full region names (e.g., "eu-north1").
+    """
+    result = {}
+    response = iam().TenantServiceClient(nebius.sdk()).list(
+        iam().ListTenantsRequest(), timeout=TIMEOUT).wait()
+
+    for tenant in response.items:
+        projects = (iam().ProjectServiceClient(nebius.sdk()).list(
+            iam().ListProjectsRequest(parent_id=tenant.metadata.id),
+            timeout=TIMEOUT).wait())
+
+        for project in projects.items:
+            match = re.match(r'^project-([a-z0-9]{3})', project.metadata.id)
+            if match is None:
+                logger.error('Could not parse project id %s',
+                             project.metadata.id)
+                continue
+            result[match.group(1)] = project.status.region
+
+    return result
+
+
+def _get_all_platform_prices() -> List[PresetInfo]:
+    """Orchestrates fetching specifications and prices for all platforms across
+     all regions.
+
+    This function first retrieves a map of region codes to full names, then
+    iterates through each region, fetches available platforms for
+    the corresponding project ID, and finally estimates prices for all presets
+    on those platforms.
+
+    Returns:
+        List[PresetInfo]: A consolidated list of PresetInfo objects for all
+                        platforms and presets across all regions.
+    """
+
+    # Get regions codes to names
+    regions_map = _get_regions_map()
+
+    presets = []
+
+    for region_code in sorted(regions_map.keys()):
+        project_id = PARENT_ID_TEMPLATE.format(region_code)
+        region = regions_map[region_code]
+        logger.info('Processing region: %s (project: %s)...', region,
+                    project_id)
+
+        platforms = _fetch_platforms_for_project(project_id)
+        if not platforms:
+            logger.warning('No platforms found in region %s', region)
+            continue
+
+        presets.extend(
+            _estimate_platforms(platforms=platforms,
+                                parent_id=project_id,
+                                region=region))
+
+    return presets
+
+
+def main() -> None:
+    """Main function to fetch and write Nebius platform prices to a CSV file.
+
+    It initializes the SDK, fetches all platform prices, and then writes them
+    to the specified CSV file.
+    """
+
+    output_file = 'nebius/vms.csv'
+
+    # Fetch presets and estimate
+    presets = _get_all_platform_prices()
+
+    # Write CSV
+    _write_preset_prices(presets, output_file)
+
+    logger.info('Done!')
+
+
+if __name__ == '__main__':
+    main()