Skip to content

Issue #2139: switch tarball install flow to CloudFront indexes #4644

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
rwilliamspbg-ops wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Issue #2139: switch tarball install flow to CloudFront indexes #4644

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
2b3f242
Issue #2139: switch tarball install flow to CloudFront indexes
rwilliamspbg-ops Apr 17, 2026
3f40074
Merge branch 'main' into users/ryan/issue-2139-cloudfront-tarball
rwilliamspbg-ops Apr 17, 2026
7e6a5c2
Merge branch 'main' into users/ryan/issue-2139-cloudfront-tarball
rwilliamspbg-ops Apr 26, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 6 additions & 6 deletions RELEASES.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -551,12 +551,12 @@ such as setting environment variables.

Release tarballs are uploaded to the following locations:

| Tarball index | S3 bucket | Description |
| ----------------------------------------- | ---------------------------------------------------------------------------------------- | -------------------------------------------------- |
| https://repo.amd.com/rocm/tarball/ | (not publicly accessible) | Stable releases |
| https://rocm.nightlies.amd.com/tarball/ | [`therock-nightly-tarball`](https://therock-nightly-tarball.s3.amazonaws.com/index.html) | Nightly builds from the default development branch |
| https://rocm.prereleases.amd.com/tarball/ | (not publicly accessible) | ⚠️ Prerelease builds for QA testing ⚠️ |
| https://rocm.devreleases.amd.com/tarball/ | [`therock-dev-tarball`](https://therock-dev-tarball.s3.amazonaws.com/index.html) | ⚠️ Development builds from project maintainers ⚠️ |
| Tarball index | Legacy bucket name | Description |
| ----------------------------------------------------------------- | ------------------------ | -------------------------------------------------- |
| [https://repo.amd.com/rocm/tarball/](https://repo.amd.com/rocm/tarball/) | n/a | Stable releases |
| [https://rocm.nightlies.amd.com/tarball/](https://rocm.nightlies.amd.com/tarball/) | therock-nightly-tarball | Nightly builds from the default development branch |
| [https://rocm.prereleases.amd.com/tarball/](https://rocm.prereleases.amd.com/tarball/) | n/a | ⚠️ Prerelease builds for QA testing ⚠️ |
| [https://rocm.devreleases.amd.com/tarball/](https://rocm.devreleases.amd.com/tarball/) | therock-dev-tarball | ⚠️ Development builds from project maintainers ⚠️ |

### Manual tarball extraction

Expand Down
142 changes: 76 additions & 66 deletions build_tools/install_rocm_from_artifacts.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,9 +114,6 @@
"""

import argparse
import boto3
from botocore import UNSIGNED
from botocore.config import Config
from datetime import datetime
from fetch_artifacts import main as fetch_artifacts_main
from pathlib import Path
Expand All @@ -126,20 +123,13 @@
import subprocess
import sys
import tarfile
from urllib.request import urlopen
from typing import Optional

PLATFORM = platform.system().lower()
s3_client = boto3.client(
"s3",
verify=False,
config=Config(max_pool_connections=100, signature_version=UNSIGNED),
)
# S3 bucket names for TheRock releases.
# NOTE: These buckets will be restricted to CloudFront-only access in the future.
# When that happens, direct S3 API calls (list_objects, download_fileobj) will fail
# and this script will need to be updated to use CloudFront URLs instead.
NIGHTLY_BUCKET_NAME = "therock-nightly-tarball"
DEV_BUCKET_NAME = "therock-dev-tarball"
# Public CloudFront tarball indexes.
NIGHTLY_TARBALL_INDEX_URL = "https://rocm.nightlies.amd.com/tarball/"
DEV_TARBALL_INDEX_URL = "https://rocm.devreleases.amd.com/tarball/"


def parse_nightly_version(version: str) -> Optional[datetime]:
Expand Down Expand Up @@ -168,64 +158,75 @@ def extract_version_from_asset_name(
return None


def _extract_asset_names_from_index(index_html: str) -> set[str]:
"""Extract .tar.gz asset names from a CloudFront tarball index page."""
# Match both href="name" and href=".../name" variants.
href_matches = re.findall(
r'href=["\'](?:[^"\']*/)?([^"\']+\.tar\.gz)["\']', index_html
)
# Match JavaScript object lists containing entries like:
# {"name": "therock-dist-linux-gfx110X-all-7.13.0a20260417.tar.gz", ...}
json_name_matches = re.findall(r'"name"\s*:\s*"([^"]+\.tar\.gz)"', index_html)
return set(href_matches + json_name_matches)


def _fetch_tarball_asset_names(index_url: str) -> set[str]:
"""Fetch available tarball asset names from a CloudFront index."""
with urlopen(index_url) as response:
html = response.read().decode("utf-8", errors="replace")
return _extract_asset_names_from_index(html)


def list_available_nightly_gpu_families(platform_str: str = PLATFORM) -> set[str]:
"""
Query S3 to find all GPU families that have nightly releases.
Useful for error messages when an invalid GPU family is specified.
"""
prefix = f"therock-dist-{platform_str}-"

paginator = s3_client.get_paginator("list_objects_v2")
families: set[str] = set()

for page in paginator.paginate(Bucket=NIGHTLY_BUCKET_NAME, Prefix=prefix):
for obj in page.get("Contents", []):
# Extract family from: therock-dist-linux-{family}-{version}.tar.gz
match = re.match(rf"{prefix}([\w-]+)-", obj["Key"])
if match:
families.add(match.group(1))
for asset_name in _fetch_tarball_asset_names(NIGHTLY_TARBALL_INDEX_URL):
# Extract family from: therock-dist-linux-{family}-{version}.tar.gz
match = re.match(rf"{prefix}([\w-]+)-", asset_name)
if match:
families.add(match.group(1))

return families


def _fetch_and_sort_nightly_releases(
artifact_group: str,
platform_str: str = PLATFORM,
) -> list[dict]:
) -> list[dict[str, datetime | str | None]]:
"""
Fetch and sort nightly releases from S3 bucket for a given artifact group.
Fetch and sort nightly releases from CloudFront for a given artifact group.

Returns:
List of dicts with keys: version, asset_name, last_modified, size, parsed_date
Sorted by recency (newest first).
List of dicts with keys: version, asset_name, parsed_date
Sorted by recency (newest first) by parsed date then version string.
"""
prefix = f"therock-dist-{platform_str}-{artifact_group}-"

paginator = s3_client.get_paginator("list_objects_v2")
releases: list[dict] = []

for page in paginator.paginate(Bucket=NIGHTLY_BUCKET_NAME, Prefix=prefix):
for obj in page.get("Contents", []):
key = obj["Key"]
if not key.endswith(".tar.gz"):
continue
version = extract_version_from_asset_name(key, artifact_group, platform_str)
if version:
releases.append(
{
"version": version,
"asset_name": key,
"last_modified": obj["LastModified"],
"size": obj["Size"],
"parsed_date": parse_nightly_version(version),
}
)

# Sort by parsed date (newest first), falling back to last_modified
releases: list[dict[str, datetime | str | None]] = []

for asset_name in _fetch_tarball_asset_names(NIGHTLY_TARBALL_INDEX_URL):
if not asset_name.startswith(prefix):
continue
version = extract_version_from_asset_name(asset_name, artifact_group, platform_str)
if version:
releases.append(
{
"version": version,
"asset_name": asset_name,
"parsed_date": parse_nightly_version(version),
}
)

# Sort by parsed date (newest first), falling back to version text ordering.
releases.sort(
key=lambda x: (
x["parsed_date"] if x["parsed_date"] else datetime.min,
x["last_modified"],
x["version"],
),
reverse=True,
)
Expand Down Expand Up @@ -279,22 +280,28 @@ def _create_output_directory(output_dir: Path):
log(f"Created output directory '{output_dir.resolve()}'")


def _retrieve_s3_release_assets(
release_bucket, artifact_group, release_version, output_dir
):
"""
Makes an API call to retrieve the release's assets, then retrieves the asset matching the amdgpu family
"""
asset_name = f"therock-dist-{PLATFORM}-{artifact_group}-{release_version}.tar.gz"
def _download_release_asset(index_url: str, asset_name: str, output_dir: Path):
"""Download an asset from a CloudFront index and extract it."""
destination = output_dir / asset_name
download_url = f"{index_url}{asset_name}"

with open(destination, "wb") as f:
s3_client.download_fileobj(release_bucket, asset_name, f)
log(f"Downloading {download_url}")
with urlopen(download_url) as response, open(destination, "wb") as f:
shutil.copyfileobj(response, f)

# After downloading the asset, untar-ing the file
_untar_files(output_dir, destination)


def _retrieve_release_assets(
release_index_url: str, artifact_group: str, release_version: str, output_dir: Path
):
"""
Retrieve and extract the release asset matching the artifact group.
"""
asset_name = f"therock-dist-{PLATFORM}-{artifact_group}-{release_version}.tar.gz"
_download_release_asset(release_index_url, asset_name, output_dir)


def retrieve_artifacts_by_run_id(args):
"""
If the user requested TheRock artifacts by CI (run ID), this function will retrieve those assets
Expand Down Expand Up @@ -496,18 +503,18 @@ def retrieve_artifacts_by_release(args):
log("This script requires a nightly-tarball or dev-tarball version.")
log("Please retrieve the correct release version from:")
log(
"\t - https://therock-nightly-tarball.s3.amazonaws.com/ (nightly-tarball examples: 6.4.0rc20250416, 7.10.0a20251024)"
"\t - https://rocm.nightlies.amd.com/tarball/ (nightly-tarball examples: 6.4.0rc20250416, 7.10.0a20251024)"
)
log(
"\t - https://therock-dev-tarball.s3.amazonaws.com/ (dev-tarball example: 6.4.0.dev0+8f6cdfc0d95845f4ca5a46de59d58894972a29a9)"
"\t - https://rocm.devreleases.amd.com/tarball/ (dev-tarball example: 6.4.0.dev0+8f6cdfc0d95845f4ca5a46de59d58894972a29a9)"
)
log("Exiting...")
return

release_bucket = NIGHTLY_BUCKET_NAME if nightly_release else DEV_BUCKET_NAME
release_index_url = NIGHTLY_TARBALL_INDEX_URL if nightly_release else DEV_TARBALL_INDEX_URL
release_version = args.release

log(f"Retrieving artifacts from release bucket {release_bucket}")
log(f"Retrieving artifacts from release index {release_index_url}")

if args.dry_run:
asset_name = (
Expand All @@ -516,8 +523,11 @@ def retrieve_artifacts_by_release(args):
log(f"[DRY RUN] Would download: {asset_name} (version {release_version})")
return

_retrieve_s3_release_assets(
release_bucket, artifact_group, release_version, output_dir
_retrieve_release_assets(
release_index_url=release_index_url,
artifact_group=artifact_group,
release_version=release_version,
output_dir=output_dir,
)


Expand Down Expand Up @@ -577,8 +587,8 @@ def retrieve_artifacts_by_latest_release(args):
return

# Reuse existing download logic
_retrieve_s3_release_assets(
release_bucket=NIGHTLY_BUCKET_NAME,
_retrieve_release_assets(
release_index_url=NIGHTLY_TARBALL_INDEX_URL,
artifact_group=args.artifact_group,
release_version=version,
output_dir=args.output_dir,
Expand Down