granny-devops 0.7.0__tar.gz → 0.8.0__tar.gz

{granny_devops-0.7.0 → granny_devops-0.8.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: granny-devops
-Version: 0.7.0
+Version: 0.8.0
 Summary: Cloud tools collection -- AWS infrastructure, CDN, and DevOps automation
 Author-email: Martin Wieser <martin.wieser@pseekoo.com>
 License: MIT License
@@ -263,6 +263,10 @@ granny analyze gpus --provider aws --profile prod --profile dev
 granny analyze credits                             # available balances
 granny analyze costs                               # MTD + month-end forecast
 
+# AWS Capacity Blocks for ML -- discover available H100/H200/A100 blocks
+granny analyze capacity-blocks --instance-type p5.48xlarge --hours 24
+granny analyze capacity-blocks --instance-type p5e.48xlarge --count 2 --hours 168
+
 # Cloudflare account resources (Workers, D1, R2, KV)
 granny cloudflare d1 create my-app
 granny cloudflare r2 create my-app-media

{granny_devops-0.7.0 → granny_devops-0.8.0}/README.md

@@ -131,6 +131,10 @@ granny analyze gpus --provider aws --profile prod --profile dev
 granny analyze credits                             # available balances
 granny analyze costs                               # MTD + month-end forecast
 
+# AWS Capacity Blocks for ML -- discover available H100/H200/A100 blocks
+granny analyze capacity-blocks --instance-type p5.48xlarge --hours 24
+granny analyze capacity-blocks --instance-type p5e.48xlarge --count 2 --hours 168
+
 # Cloudflare account resources (Workers, D1, R2, KV)
 granny cloudflare d1 create my-app
 granny cloudflare r2 create my-app-media

{granny_devops-0.7.0 → granny_devops-0.8.0}/granny/__init__.py

@@ -1,6 +1,6 @@
 """Granny -- Cloud tools collection for AWS infrastructure and DevOps automation."""
 
-__version__ = "0.7.0"
+__version__ = "0.8.0"
 __all__ = [
     "get_secret",
     "load_secrets_into_env",

{granny_devops-0.7.0 → granny_devops-0.8.0}/granny/analyze/__init__.py

@@ -17,9 +17,11 @@ def __getattr__(name: str):
     if name in {
         "GpuInstance",
         "GpuReservation",
+        "CapacityBlockOffering",
         "aws_profiles",
         "list_aws_gpu_instances",
         "list_aws_gpu_reservations",
+        "list_aws_capacity_block_offerings",
         "gcp_projects",
         "list_gcp_gpu_instances",
         "list_gcp_gpu_reservations",

{granny_devops-0.7.0 → granny_devops-0.8.0}/granny/analyze/gpus.py

@@ -16,7 +16,7 @@ from __future__ import annotations
 import logging
 import re
 from dataclasses import dataclass, field
-from datetime import datetime, timezone
+from datetime import datetime, timedelta, timezone
 
 logger = logging.getLogger(__name__)
 
@@ -148,6 +148,24 @@ class GpuReservation:
     commitment_term: str | None
 
 
+@dataclass
+class CapacityBlockOffering:
+    """An available EC2 Capacity Block for ML offering (not yet purchased)."""
+
+    account: str
+    region: str
+    zone: str
+    offering_id: str
+    instance_type: str
+    instance_count: int
+    duration_hours: int
+    start_time: str
+    end_time: str
+    upfront_fee: float
+    currency: str
+    tenancy: str
+
+
 # ---------------------------------------------------------------------------
 # AWS
 # ---------------------------------------------------------------------------
@@ -315,6 +333,97 @@ def list_aws_gpu_reservations(
     return results
 
 
+# Regions where Capacity Blocks for ML are generally available.
+# describe-capacity-block-offerings only works here; other regions return
+# UnsupportedOperation. Update as AWS rolls out to new regions.
+_AWS_CAPACITY_BLOCK_REGIONS: tuple[str, ...] = (
+    "us-east-1",
+    "us-east-2",
+    "us-west-2",
+    "eu-west-1",
+    "eu-west-2",
+    "eu-north-1",
+    "eu-central-1",
+    "ap-northeast-1",
+    "ap-southeast-2",
+)
+
+
+def list_aws_capacity_block_offerings(
+    instance_type: str,
+    instance_count: int = 1,
+    duration_hours: int = 24,
+    start_window_days: int = 30,
+    profiles: list[str] | None = None,
+    regions: list[str] | None = None,
+) -> list[CapacityBlockOffering]:
+    """Find available EC2 Capacity Block offerings for ML GPU instances.
+
+    Args:
+        instance_type: GPU instance type, e.g. ``p5.48xlarge`` or ``p5e.48xlarge``.
+        instance_count: How many instances the block must hold.
+        duration_hours: Block length in hours (AWS supports 1, 6, 12, 24, 48,
+            72, 168, 336, 504, 672 ... — pass what you need).
+        start_window_days: Search for blocks starting between now and this many
+            days out.
+        profiles: AWS profiles to query (default: all configured profiles).
+        regions: Regions to query (default: the known Capacity-Block-enabled
+            regions; pass an explicit list to override).
+
+    Returns:
+        Offerings sorted cheapest-first.
+    """
+    import boto3
+    from botocore.exceptions import BotoCoreError, ClientError
+
+    profiles = profiles or aws_profiles()
+    regions = regions or list(_AWS_CAPACITY_BLOCK_REGIONS)
+    now = datetime.now(timezone.utc)
+    end = now + timedelta(days=start_window_days)
+
+    out: list[CapacityBlockOffering] = []
+    for profile in profiles:
+        for region in regions:
+            try:
+                session = boto3.Session(profile_name=profile if profile != "default" else None)
+                ec2 = session.client("ec2", region_name=region)
+                paginator = ec2.get_paginator("describe_capacity_block_offerings")
+                for page in paginator.paginate(
+                    InstanceType=instance_type,
+                    InstanceCount=instance_count,
+                    CapacityDurationHours=duration_hours,
+                    StartDateRange=now,
+                    EndDateRange=end,
+                ):
+                    for o in page.get("CapacityBlockOfferings", []):
+                        try:
+                            fee = float(o.get("UpfrontFee", 0) or 0)
+                        except (TypeError, ValueError):
+                            fee = 0.0
+                        out.append(
+                            CapacityBlockOffering(
+                                account=profile,
+                                region=region,
+                                zone=o.get("AvailabilityZone", ""),
+                                offering_id=o.get("CapacityBlockOfferingId", ""),
+                                instance_type=o.get("InstanceType", instance_type),
+                                instance_count=o.get("InstanceCount", instance_count),
+                                duration_hours=o.get(
+                                    "CapacityBlockDurationHours", duration_hours
+                                ),
+                                start_time=_iso(o.get("StartDate")) or "",
+                                end_time=_iso(o.get("EndDate")) or "",
+                                upfront_fee=fee,
+                                currency=o.get("CurrencyCode", "USD"),
+                                tenancy=o.get("Tenancy", "default"),
+                            )
+                        )
+            except (BotoCoreError, ClientError) as e:
+                logger.debug("AWS %s/%s capacity-block offerings: %s", profile, region, e)
+    out.sort(key=lambda o: (o.upfront_fee, o.start_time))
+    return out
+
+
 # ---------------------------------------------------------------------------
 # GCP
 # ---------------------------------------------------------------------------

{granny_devops-0.7.0 → granny_devops-0.8.0}/granny/cli/analyze.py

@@ -125,13 +125,16 @@ def gpus(
 
     instances: list = []
     reservations: list = []
+    scanned: dict[str, list[str]] = {}
 
     for p in targets:
         try:
             if p == "aws":
+                aws_targets = list(profiles) or g.aws_profiles()
+                scanned["aws"] = aws_targets
                 instances.extend(
                     g.list_aws_gpu_instances(
-                        profiles=list(profiles) or None,
+                        profiles=aws_targets,
                         regions=list(regions) or None,
                         include_stopped=include_stopped,
                     )
@@ -139,33 +142,35 @@ def gpus(
                 if include_reserved:
                     reservations.extend(
                         g.list_aws_gpu_reservations(
-                            profiles=list(profiles) or None,
+                            profiles=aws_targets,
                             regions=list(regions) or None,
                         )
                     )
             elif p == "gcp":
+                gcp_targets = list(projects) or g.gcp_projects()
+                scanned["gcp"] = gcp_targets
                 instances.extend(
                     g.list_gcp_gpu_instances(
-                        projects=list(projects) or None,
+                        projects=gcp_targets,
                         include_stopped=include_stopped,
                     )
                 )
                 if include_reserved:
                     reservations.extend(
-                        g.list_gcp_gpu_reservations(projects=list(projects) or None)
+                        g.list_gcp_gpu_reservations(projects=gcp_targets)
                     )
             elif p == "azure":
+                az_targets = list(subscriptions) or [sid for sid, _ in g.azure_subscriptions()]
+                scanned["azure"] = az_targets
                 instances.extend(
                     g.list_azure_gpu_instances(
-                        subscriptions=list(subscriptions) or None,
+                        subscriptions=az_targets,
                         include_stopped=include_stopped,
                     )
                 )
                 if include_reserved:
                     reservations.extend(
-                        g.list_azure_gpu_reservations(
-                            subscriptions=list(subscriptions) or None
-                        )
+                        g.list_azure_gpu_reservations(subscriptions=az_targets)
                     )
         except RuntimeError as e:
             click.echo(f"[{p}] skipped: {e}", err=True)
@@ -178,6 +183,7 @@ def gpus(
         click.echo(
             json.dumps(
                 {
+                    "scanned": scanned,
                     "instances": [asdict(i) for i in instances],
                     "reservations": [asdict(r) for r in reservations],
                 },
@@ -187,6 +193,13 @@ def gpus(
         )
         return
 
+    scope_labels = {"aws": "AWS profiles", "gcp": "GCP projects", "azure": "Azure subscriptions"}
+    for p in targets:
+        items = scanned.get(p)
+        if items is None:
+            continue
+        click.echo(f"Scanned {scope_labels[p]}: {', '.join(items) if items else '(none)'}")
+
     if not instances and not reservations:
         click.echo("No GPU resources found.")
         return
@@ -227,6 +240,96 @@ def gpus(
             )
 
 
+@analyze.command("capacity-blocks")
+@click.option(
+    "--instance-type",
+    "instance_type",
+    required=True,
+    help="GPU instance type, e.g. p5.48xlarge, p5e.48xlarge, p5en.48xlarge, p4d.24xlarge",
+)
+@click.option("--count", "instance_count", default=1, show_default=True, help="Instances per block")
+@click.option(
+    "--hours",
+    "duration_hours",
+    default=24,
+    show_default=True,
+    help="Block duration in hours (1, 6, 12, 24, 48, 72, 168, 336, 504, 672, ...)",
+)
+@click.option(
+    "--start-window-days",
+    default=30,
+    show_default=True,
+    help="Search for blocks starting within this many days from now",
+)
+@click.option("--profile", "profiles", multiple=True, help="AWS profile (repeatable; default: all)")
+@click.option(
+    "--region",
+    "regions",
+    multiple=True,
+    help="AWS region (repeatable; default: known Capacity-Block regions)",
+)
+@click.option("--json-output", "as_json", is_flag=True, help="Output as JSON")
+def capacity_blocks(
+    instance_type: str,
+    instance_count: int,
+    duration_hours: int,
+    start_window_days: int,
+    profiles: tuple[str, ...],
+    regions: tuple[str, ...],
+    as_json: bool,
+) -> None:
+    """Find available AWS Capacity Blocks for ML (H100/H200/A100 reservations).
+
+    Scans regions where Capacity Blocks are sold and returns open offerings
+    sorted cheapest-first. Use the offering ID with the AWS CLI to purchase:
+
+      aws ec2 purchase-capacity-block --capacity-block-offering-id <id>
+          --instance-platform Linux/UNIX
+    """
+    from dataclasses import asdict
+
+    from granny.analyze.gpus import list_aws_capacity_block_offerings
+
+    offerings = list_aws_capacity_block_offerings(
+        instance_type=instance_type,
+        instance_count=instance_count,
+        duration_hours=duration_hours,
+        start_window_days=start_window_days,
+        profiles=list(profiles) or None,
+        regions=list(regions) or None,
+    )
+
+    if as_json:
+        click.echo(json.dumps([asdict(o) for o in offerings], indent=2, default=str))
+        return
+
+    if not offerings:
+        click.echo(
+            f"No Capacity Block offerings found for {instance_type} x{instance_count} "
+            f"({duration_hours}h) in the next {start_window_days} days.\n"
+            "Try: a longer --start-window-days, fewer instances, a different "
+            "duration, or another --instance-type."
+        )
+        return
+
+    click.echo(
+        f"\n{'Profile':<14} {'Region':<14} {'AZ':<16} {'Type':<16} "
+        f"{'#':>3} {'Hrs':>4} {'Start (UTC)':<22} {'Upfront':>12} {'Cur':<4} {'Offering ID'}"
+    )
+    click.echo("-" * 140)
+    for o in offerings:
+        click.echo(
+            f"{o.account[:13]:<14} {o.region:<14} {o.zone:<16} "
+            f"{o.instance_type:<16} {o.instance_count:>3} {o.duration_hours:>4} "
+            f"{o.start_time[:19]:<22} {o.upfront_fee:>12,.2f} {o.currency:<4} "
+            f"{o.offering_id}"
+        )
+    click.echo(
+        "\nPurchase with:  aws ec2 purchase-capacity-block "
+        "--capacity-block-offering-id <id> --instance-platform Linux/UNIX"
+    )
+
+
 @analyze.command()
 @click.option(
     "--provider",

{granny_devops-0.7.0 → granny_devops-0.8.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "granny-devops"
-version = "0.7.0"
+version = "0.8.0"
 description = "Cloud tools collection -- AWS infrastructure, CDN, and DevOps automation"
 readme = "README.md"
 license = { file = "LICENSE" }