ci: make SynapseML releases approved and resumable - #2628
Rana Singh (ranadeepsingh) wants to merge 25 commits into
Conversation
|
Azure Pipelines: There may be pipelines that require an authorized user to comment /azp run to run. |
|
Hey Rana Singh (@ranadeepsingh) 👋! We use semantic commit messages to streamline the release process. Examples of commit messages with semantic prefixes:
To test your commit locally, please follow our guild on building from source. |
c763701 to
2822d0b
Compare
## Summary Refresh the release automation on current master, align it with the Fabric release guide and the live Publish-Official pipeline, and add proof-oriented coverage for release identifiers, artifact verification, GitHub workflows, and BBC-VHD edits. ## Prompting Intent Refresh microsoft/SynapseML PR microsoft#2628 using the SynapseML PR readiness loop, follow the internal Fabric release guide, and derive automation from actual prior OSS, Internal, Publish-Official, and BBC-VHD releases rather than relying on stale examples. ## Linked Sources - Fabric release guide: https://msdata.visualstudio.com/A365/_wiki/wikis/Osmos%20Team%20Wiki/130638/SynapseML-Fabric-Release-Guide-v2 - Release automation PR: microsoft#2628 - Derivative tag automation: microsoft#2540 - Live v1.1.3 release: https://github.com/microsoft/SynapseML/releases/tag/v1.1.3 - Publish-Official pipeline: https://msdata.visualstudio.com/A365/_build?definitionId=35879 - Historical BBC-VHD release PR: https://msdata.visualstudio.com/A365/_git/BBC-VHD/pullrequest/1805064 ## Rationale Keep ESRP, review, White-Glove, and train decisions human-gated while automating deterministic mechanics. Tag the exact reviewed merge, dispatch downstream GitHub automation explicitly because GITHUB_TOKEN pushes do not recurse, gate release notes on public artifacts, use the live pipeline parameter contract, fail loudly on incomplete network evidence, and roll back paired BBC-VHD writes rather than leaving partial release state. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
/azp run |
|
Azure Pipelines: There may be pipelines that require an authorized user to comment /azp run to run. |
|
Azure Pipelines: Successfully started running 1 pipeline(s). |
There was a problem hiding this comment.
Pull request overview
This PR adds release-engineering automation to make SynapseML-to-Fabric releases more deterministic and reviewable, including a GitHub Actions “prepare → tag merged commit → publish notes” flow plus supporting Python tooling to derive/verify release identifiers and safely bump BBC-VHD pins.
Changes:
- Introduces a release matrix generator (
release_matrix.py) and an end-to-end verifier (verify_release.py) with accompanying tests. - Adds
bump_bbcvhd.py(plus tests) to safely update BBC-VHD component pins while preserving line endings and rolling back on failure. - Adds/updates GitHub workflows for release prepare/tagging and manual, artifact-gated release notes; enables
workflow_dispatchfor PR validation so release PRs created viaGITHUB_TOKENcan be validated.
Show a summary per file
| File | Description |
|---|---|
| scripts/test_bump_version.py | Extends bump-version tests for denylisted repo-relative paths; normalizes path keys to POSIX. |
| scripts/bump-version.py | Improves Windows console encoding robustness; adds path-based denylist support; normalizes path handling via as_posix(). |
| scripts/release/release_matrix.py | Adds a single-source-of-truth release matrix for tags and artifact versions + CLI rendering. |
| scripts/release/verify_release.py | Adds live verification of GitHub/ADO tags and Maven/PyPI/UPack/Azure Artifacts presence. |
| scripts/release/bump_bbcvhd.py | Adds deterministic, rollback-safe BBC-VHD component pin updater with CRLF/LF preservation. |
| scripts/release/README.md | Documents how to use the new release tooling and where it fits in the guide. |
| scripts/release/test_release_matrix.py | Adds regression/contract tests for the release matrix derivations and CLI validation. |
| scripts/release/test_verify_release.py | Adds unit tests for verifier networking/error handling and run plan behavior. |
| scripts/release/test_bump_bbcvhd.py | Adds tests for BBC-VHD bump idempotency, rollback, and newline preservation. |
| scripts/release/test_release_workflows.py | Adds tests asserting key workflow contract properties (manual gating, dispatch behavior). |
| scripts/release/test_prev_tag.sh | Adds a repo-taglist regression script for “previous primary tag” selection logic. |
| .github/workflows/release-prepare.yml | Adds a reviewed release PR generator and a merged-commit tagger that dispatches downstream orchestration. |
| .github/workflows/release-notes.yml | Adds a manual, artifact-gated GitHub Release publisher with explicit previous-tag selection. |
| .github/workflows/pr-validation.yml | Enables workflow_dispatch so validations can be dispatched for bot-opened release PR branches. |
Review details
💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Suppressed comments (1)
scripts/release/verify_release.py:260
--skip internaldoes not currently skip the internal PyPI feed checks (thesynapseml_internalpackage); it only skips internal git tag checks. This makes--skip internalbehave inconsistently compared with--skip public.
def pip(self, package: str, version: str) -> str:
if "pip" in self.skip or "ado" in self.skip:
return SKIPPED
# Azure Artifacts normalises pypi names: synapseml_internal -> synapseml-internal
return (
OK
if version
in self._feed_versions("Synapse-Conda", "pypi", package.replace("_", "-"))
else MISSING
)
- Files reviewed: 14/14 changed files
- Comments generated: 2
- Review effort level: Lite
## Summary Make --skip internal consistently skip SynapseML-Internal tags, UPacks, and wheels while retaining OSS artifact checks, and document every skip scope with regression coverage. ## Prompting Intent Resolve all current-head automated review findings on microsoft/SynapseML PR microsoft#2628 and preserve an explicit, safe release-verification CLI contract. ## Linked Sources - Release automation PR: microsoft#2628 - Internal artifact skip review: microsoft#2628 (comment) - Skip help review: microsoft#2628 (comment) ## Rationale A release operator who opts out of Internal verification must not still query or fail on Internal packages. Passing artifact scope explicitly keeps OSS checks active, avoids hiding public release gaps, and makes combined skip behavior predictable from both CLI help and maintainer documentation. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
/azp run |
|
Azure Pipelines: Successfully started running 1 pipeline(s). |
There was a problem hiding this comment.
Review details
Suppressed comments (1)
Previously missed (1) — in code that hasn't changed since the last review.
scripts/release/verify_release.py:75
- On Windows,
subprocess.run(..., shell=True)expects a command string, but this code passes a list. That combination can fail (or behave differently) on win32, which would break the fallback ADO auth path even whenazis installed. Build a platform-appropriate command (string forshell=True, list otherwise).
out = subprocess.run(
[
"az",
"account",
"get-access-token",
"--resource",
ADO_RESOURCE,
"--query",
"accessToken",
"-o",
"tsv",
],
capture_output=True,
text=True,
shell=(sys.platform == "win32"),
)
- Files reviewed: 14/14 changed files
- Comments generated: 0 new
- Review effort level: Lite
## Summary Build the Azure CLI authentication command as a string only when Windows requires shell execution, while retaining an argument list and shell-free execution on other platforms. Add regression coverage for both command shapes. ## Prompting Intent Resolve the current-head suppressed Copilot review finding on microsoft/SynapseML PR microsoft#2628 and prove the fallback authentication path used by release verification on Windows. ## Linked Sources - Release automation PR: microsoft#2628 - Fabric release guide: https://msdata.visualstudio.com/A365/_wiki/wikis/Osmos%20Team%20Wiki/130638/SynapseML-Fabric-Release-Guide-v2 ## Rationale Windows resolves the Azure CLI through a command shim, so the verifier needs shell execution there; subprocess requires that shell command to be a string. Keeping list-based execution elsewhere preserves safe argument boundaries, while an actual token-less Windows replay proves the fallback can authenticate and inspect the live release. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
Addressed the current-head suppressed Windows auth finding in 6e80f27.
|
|
/azp run |
|
Azure Pipelines: Successfully started running 1 pipeline(s). |
There was a problem hiding this comment.
Review details
Suppressed comments (1)
Previously missed (1) — in code that hasn't changed since the last review.
scripts/release/verify_release.py:313
- The Maven publication gate here verifies only
synapseml-core_{scala}(viapublic_maven()), but the workflow-generated release notes advertisecom.microsoft.azure:synapseml_2.12:${VERSION}(see.github/workflows/release-notes.yml:137) and the docs use thesynapseml_2.12coordinate (website/doctest.py:25). This can letverify_release.pyreport COMPLETE even if the primary install coordinate is missing (or vice versa). Consider aligning the verification to check the same Maven artifact(s) that users are instructed to consume (e.g., verify bothsynapseml_{scala}andsynapseml-core_{scala}, or switch to the canonical one).
"maven",
tp.key,
f"synapseml-core_{tp.scala}",
tp.oss_maven_version,
c.public_maven(tp.scala, tp.oss_maven_version),
- Files reviewed: 14/14 changed files
- Comments generated: 0 new
- Review effort level: Lite
## Summary Gate release completeness on both the user-facing synapseml_<scala> Maven coordinate and the release-guide synapseml-core_<scala> coordinate for every selected Spark target. Add a failing-install-coordinate regression and document the two-coordinate contract. ## Prompting Intent Resolve the current-head suppressed Copilot review finding on microsoft/SynapseML PR microsoft#2628 by aligning artifact verification with generated release notes, installation docs, and actual published releases. ## Linked Sources - Release automation PR: microsoft#2628 - Fabric release guide: https://msdata.visualstudio.com/A365/_wiki/wikis/Osmos%20Team%20Wiki/130638/SynapseML-Fabric-Release-Guide-v2 - Live v1.1.3 release: https://github.com/microsoft/SynapseML/releases/tag/v1.1.3 ## Rationale The aggregate synapseml artifact is the coordinate users install and the one release notes advertise, while the guide explicitly points maintainers to synapseml-core. Requiring both prevents a GitHub Release from reporting complete when either the public install contract or the guide's Maven evidence is absent. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
Azure Pipelines: Successfully started running 1 pipeline(s). |
There was a problem hiding this comment.
Copilot review overview
🟡 Changes recommended
Two critical issues remain in ESRP staging: stale explicitly versioned artifacts can be relabeled, and mandatory arguments break the supported Spark 4.1 pipeline.
Get a fresh assessment by requesting another Copilot review.
Review tier: Lite
Findings: 2
| filename = ( | ||
| path.name | ||
| if suffix.startswith(f"-{version}.") | ||
| or suffix.startswith(f"-{version}-") | ||
| else f"{module}-{version}{suffix}" | ||
| ) |
|
|
||
| def main(argv=None): | ||
| parser = argparse.ArgumentParser(description=__doc__) | ||
| parser.add_argument("--version", required=True) |
Automates the mechanical parts of the SynapseML Fabric release so the remaining human work is decision-making and approvals. Fixes to scripts/bump-version.py (it currently FAILS on master): - PR microsoft#2589 (CDN migration) rewrote "docs/Reference/R Setup.md" to per-module archive names (synapseml-core-1.1.3.zip) and added VerifyRCodegen.scala with hardcoded version strings. Neither is matched by the existing patterns, so the script hard-fails and would block the next release. Added a generic "-{V}.zip" line anchor plus file anchors for VerifyRCodegen. - _detect_version() read docusaurus.config.js without an encoding, raising UnicodeDecodeError on a cp1252 console. - analyze() and the EXPECTED_FILES manifest check compared str(rel), which yields backslashes on Windows and never matches the "/" paths in the anchor tables, producing false "not updated" warnings. - Non-ASCII status output crashed on a cp1252 console *after* files had been rewritten, leaving a half-applied bump behind a non-zero exit. - Added DENYLIST_PATHS for files whose basename is too common to denylist safely. New tooling in scripts/release/: - release_matrix.py derives every tag, UPack version, pip version and BBC-VHD value from one input version. One release spans 7 tags per repo and 4 mutually inconsistent naming conventions; notably the OSS UPack package mangles spark dots to dashes (1.1.3-spark4-0) while the Internal package preserves them (1.1.3-0-spark4.0). - verify_release.py checks every tag and artifact against the matrix. Worth running even on a green publish pipeline, because several of its publish steps use continueOnError: true. - bump_bbcvhd.py applies a release to a BBC-VHD component, replacing the most error-prone hand-edit in the process. New workflows: - release-prepare.yml opens the version-bump PR. Two of the last four bumps landed as unsigned direct pushes to master with no PR. - release-notes.yml publishes the GitHub Release on a vX.Y.Z tag. v1.1.1 has a tag but no Release, which made v1.1.3's auto-generated notes span two releases; the workflow pins the diff base to the previous primary tag so notes stay correct regardless. Expected values in the tests are transcribed from live v1.1.1 and v1.1.3 data, so a failure means the tooling has drifted from what was actually shipped. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Refresh the release automation on current master, align it with the Fabric release guide and the live Publish-Official pipeline, and add proof-oriented coverage for release identifiers, artifact verification, GitHub workflows, and BBC-VHD edits. ## Prompting Intent Refresh microsoft/SynapseML PR microsoft#2628 using the SynapseML PR readiness loop, follow the internal Fabric release guide, and derive automation from actual prior OSS, Internal, Publish-Official, and BBC-VHD releases rather than relying on stale examples. ## Linked Sources - Fabric release guide: https://msdata.visualstudio.com/A365/_wiki/wikis/Osmos%20Team%20Wiki/130638/SynapseML-Fabric-Release-Guide-v2 - Release automation PR: microsoft#2628 - Derivative tag automation: microsoft#2540 - Live v1.1.3 release: https://github.com/microsoft/SynapseML/releases/tag/v1.1.3 - Publish-Official pipeline: https://msdata.visualstudio.com/A365/_build?definitionId=35879 - Historical BBC-VHD release PR: https://msdata.visualstudio.com/A365/_git/BBC-VHD/pullrequest/1805064 ## Rationale Keep ESRP, review, White-Glove, and train decisions human-gated while automating deterministic mechanics. Tag the exact reviewed merge, dispatch downstream GitHub automation explicitly because GITHUB_TOKEN pushes do not recurse, gate release notes on public artifacts, use the live pipeline parameter contract, fail loudly on incomplete network evidence, and roll back paired BBC-VHD writes rather than leaving partial release state. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Make --skip internal consistently skip SynapseML-Internal tags, UPacks, and wheels while retaining OSS artifact checks, and document every skip scope with regression coverage. ## Prompting Intent Resolve all current-head automated review findings on microsoft/SynapseML PR microsoft#2628 and preserve an explicit, safe release-verification CLI contract. ## Linked Sources - Release automation PR: microsoft#2628 - Internal artifact skip review: microsoft#2628 (comment) - Skip help review: microsoft#2628 (comment) ## Rationale A release operator who opts out of Internal verification must not still query or fail on Internal packages. Passing artifact scope explicitly keeps OSS checks active, avoids hiding public release gaps, and makes combined skip behavior predictable from both CLI help and maintainer documentation. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Build the Azure CLI authentication command as a string only when Windows requires shell execution, while retaining an argument list and shell-free execution on other platforms. Add regression coverage for both command shapes. ## Prompting Intent Resolve the current-head suppressed Copilot review finding on microsoft/SynapseML PR microsoft#2628 and prove the fallback authentication path used by release verification on Windows. ## Linked Sources - Release automation PR: microsoft#2628 - Fabric release guide: https://msdata.visualstudio.com/A365/_wiki/wikis/Osmos%20Team%20Wiki/130638/SynapseML-Fabric-Release-Guide-v2 ## Rationale Windows resolves the Azure CLI through a command shim, so the verifier needs shell execution there; subprocess requires that shell command to be a string. Keeping list-based execution elsewhere preserves safe argument boundaries, while an actual token-less Windows replay proves the fallback can authenticate and inspect the live release. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Gate release completeness on both the user-facing synapseml_<scala> Maven coordinate and the release-guide synapseml-core_<scala> coordinate for every selected Spark target. Add a failing-install-coordinate regression and document the two-coordinate contract. ## Prompting Intent Resolve the current-head suppressed Copilot review finding on microsoft/SynapseML PR microsoft#2628 by aligning artifact verification with generated release notes, installation docs, and actual published releases. ## Linked Sources - Release automation PR: microsoft#2628 - Fabric release guide: https://msdata.visualstudio.com/A365/_wiki/wikis/Osmos%20Team%20Wiki/130638/SynapseML-Fabric-Release-Guide-v2 - Live v1.1.3 release: https://github.com/microsoft/SynapseML/releases/tag/v1.1.3 ## Rationale The aggregate synapseml artifact is the coordinate users install and the one release notes advertise, while the guide explicitly points maintainers to synapseml-core. Requiring both prevents a GitHub Release from reporting complete when either the public install contract or the guide's Maven evidence is absent. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Make the SBT launcher download fail fast with actionable HTTP errors and transient retries, and invoke the release verifier through the guaranteed Python 3 executable on Ubuntu. Lock both workflow contracts with tests. ## Prompting Intent Resolve both current-head suppressed Copilot findings on microsoft/SynapseML PR microsoft#2628 without changing the release approval or publication flow. ## Linked Sources - Release automation PR: microsoft#2628 - Fabric release guide: https://msdata.visualstudio.com/A365/_wiki/wikis/Osmos%20Team%20Wiki/130638/SynapseML-Fabric-Release-Guide-v2 ## Rationale A failed launcher download should stop at the network boundary rather than creating a corrupt tool that fails later, and release publication must not depend on an optional python alias. Explicit curl failure semantics, bounded retries, and python3 make failures early and deterministic while preserving all human gates. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Insert the release test directory explicitly before importing verify_release and bump_bbcvhd, matching the existing release-matrix test pattern. This keeps collection independent of pytest's default path-prepend behavior. ## Prompting Intent Resolve both current-head Copilot review threads on microsoft/SynapseML PR microsoft#2628 and prove the release tests collect under alternate pytest import modes. ## Linked Sources - Release automation PR: microsoft#2628 - Verifier import review: microsoft#2628 (comment) - BBC-VHD import review: microsoft#2628 (comment) ## Rationale Relying on pytest's default import mode makes test collection sensitive to runner configuration. Explicitly locating sibling modules is already the repository convention for release_matrix and allows these suites to run consistently from the repository root and under importlib collection. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Normalize whitespace around rebuild-counter values and reject empty target keys directly in parse_iterations. Add positive whitespace and negative empty-target regression cases. ## Prompting Intent Resolve the current-head suppressed Copilot finding on microsoft/SynapseML PR microsoft#2628 and make release counter errors actionable at the CLI parsing boundary. ## Linked Sources - Release automation PR: microsoft#2628 - Fabric release guide: https://msdata.visualstudio.com/A365/_wiki/wikis/Osmos%20Team%20Wiki/130638/SynapseML-Fabric-Release-Guide-v2 ## Rationale Rebuild counters are operator-entered recovery controls for immutable artifacts. Empty targets should fail as malformed KEY=N input rather than surfacing later as an unknown target, while spacing around a numeric value should not turn an otherwise valid recovery command into an error. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Replace GNU sort -V in the release tag-history regression helper with Python 3 semantic-version ordering, reuse the sorted primary-tag stream, and support environments where the Python 3 executable is named either python3 or python. ## Prompting Intent Resolve the current-head suppressed Copilot portability finding on microsoft/SynapseML PR microsoft#2628 while preserving validation against the repository's complete historical tag list. ## Linked Sources - Release automation PR: microsoft#2628 - Live v1.1.3 release: https://github.com/microsoft/SynapseML/releases/tag/v1.1.3 ## Rationale Contributor-side release evidence should run on macOS/BSD as well as GNU systems. Python 3 is already a release-tooling dependency and provides deterministic numeric tuple ordering without relying on platform-specific sort flags; consuming the full stream also avoids early-pipeline termination behavior. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Make the release matrix the executable contract for Spark 3.5, 4.0, and 4.1; emit explicit public and Internal Maven queue commands; add full versus Internal-only release scopes; propagate rebuild counters; and fail release verification unless every expected Maven, pip, UPack, and tag artifact is present and commit-consistent. ## Prompting Intent Refresh PR microsoft#2628 on current master and make SynapseML-to-Fabric releases easy to operate, complete across supported Spark branches, and verifiably published to Maven. Follow historical releases and the SynapseML Fabric Release Guide v2, while producing an interoperable contract for the SynapseML-Internal follow-up. ## Linked Sources - GitHub pull request: microsoft#2628 - SynapseML Fabric Release Guide v2: https://msdata.visualstudio.com/A365/_wiki/wikis/Osmos%20Team%20Wiki/130638/SynapseML-Fabric-Release-Guide-v2 - Historical release reference: https://github.com/microsoft/SynapseML/releases/tag/v1.1.3 ## Rationale Use one validated matrix instead of hand-maintained branch instructions so Maven, pip, UPack, tags, and BBC-VHD values cannot drift. Queue Maven publication explicitly because the downstream packaging pipeline does not publish Maven. Keep Internal-only hotfixes from republishing immutable public artifacts, and verify live artifact files plus peeled tag commits because green pipelines alone do not prove a complete release. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Apply the repository-pinned Black 22.3.0 layout to the Fabric certificate test helper so PR validation accepts the rebased release automation changes. ## Prompting Intent Keep PR microsoft#2628 green after its release hardening update by fixing the exact formatting mismatch reported by the Python Style Check, without changing behavior or formatting unrelated files. ## Linked Sources - GitHub pull request: microsoft#2628 - Failed Python Style Check: https://github.com/microsoft/SynapseML/actions/runs/32958770630/job/98146240564 ## Rationale Apply only the formatter diff emitted by Black 22.3.0 because the local newer Black version accepted a layout that the repository-pinned CI version rejects. A focused follow-up preserves commit history and avoids unrelated repository-wide formatting churn. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Probe Maven and PyPI artifact URLs with HEAD first, falling back to GET only when a server reports HEAD as unsupported with HTTP 405 or 501. Add regression coverage for successful HEAD probes, unsupported-method fallback, and missing artifacts in both paths. ## Prompting Intent Address the current-head review finding on PR microsoft#2628 while preserving the verifier's fail-closed semantics and making complete multi-branch Maven verification faster and less bandwidth-intensive. ## Linked Sources - GitHub pull request: microsoft#2628 - Review finding: microsoft#2628 (comment) - SynapseML Fabric Release Guide v2: https://msdata.visualstudio.com/A365/_wiki/wikis/Osmos%20Team%20Wiki/130638/SynapseML-Fabric-Release-Guide-v2 ## Rationale Use HEAD for the common path so large JAR bodies are never opened unnecessarily. Fall back only for the standard unsupported-method responses, rather than masking authentication, server, or network failures, so operational errors remain actionable and missing artifacts still fail release verification. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary Add a discoverable SynapseML release skill with focused preflight, automation-boundary, recovery, and rollout references. Make the documented dry run work from a fresh checkout, keep historical review evidence out of version replacement, and add scoped verification that reports only Internal rows for an Internal-only patch. ## Prompting Intent The engineer asked to document the current SynapseML-to-Fabric release process through an agent skill in PR microsoft#2628. The skill needed to explain what can be previewed, which actions publish or tag state, where human approval remains mandatory, and how to recover without moving immutable tags or packages. The change also needed to pass the SynapseML PR loop and its six-round review gauntlet. ## Linked Sources - GitHub pull request: microsoft#2628 - Derivative tag automation: microsoft#2540 - SynapseML Fabric Release Guide v2: https://msdata.visualstudio.com/A365/_wiki/wikis/Osmos%20Team%20Wiki/130638/SynapseML-Fabric-Release-Guide-v2 - Public release reference: https://github.com/microsoft/SynapseML/releases/tag/v1.1.3 - Review evidence: reviews/release-skill/ ## Rationale Keep the main skill short and put commands and recovery detail in focused references so agents can load only what a release step needs. Use the release matrix and verifier as executable sources of truth instead of duplicating coordinates. Add an explicit Internal-only verification scope while preserving the previous nonzero-patch inference, and record the resolved scope in output so stored evidence cannot hide omitted OSS rows. Keep tag creation, publication, credentials, and rollout approvals behind explicit human gates. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 476d113f-dd35-40c6-bc79-005dcccd7b79
## Summary Move the six release-automation review reports into `review/pr-2628/`, rename them with the PR-number prefix, add an index README, and remove machine-local artifact paths. Exclude both the new `review` root and the historical `reviews` root from version replacement, with tests for directory and nested-file paths. ## Prompting Intent The engineer asked to move and rename PR microsoft#2628's review artifacts under `review/pr-2628/` and to follow the organization used by existing checked-in PR review evidence. ## Linked Sources - GitHub pull request: microsoft#2628 - Existing review-layout example: https://github.com/microsoft/SynapseML/tree/master/reviews/pr-2666 - Review evidence index: review/pr-2628/README.md ## Rationale Use the requested singular root while keeping the established PR-numbered subdirectory, filename prefix, and README pattern. Leave the existing `reviews/pr-2666/` history in place to avoid unrelated churn. Denylist both roots because both contain immutable reports with historical version strings that must not be changed by a future release bump. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 476d113f-dd35-40c6-bc79-005dcccd7b79
## Summary Move PR microsoft#2628 review evidence from the singular `review/pr-2628/` directory to `reviews/pr-2628/`. Add a repository-wide `AGENTS.md` rule for PR-numbered and feature-numbered review directories, and remove the now-unused singular-root version-bump exclusion. ## Prompting Intent The engineer clarified that committed reviews must live under `reviews/pr-<number>/`, or under the defined feature-numbered directory when there is no PR. They also asked to record this rule in `AGENTS.md` so future reviews do not use topic-named or singular-root directories. ## Linked Sources - GitHub pull request: microsoft#2628 - Existing review-layout example: https://github.com/microsoft/SynapseML/tree/master/reviews/pr-2666 - Repository guidance: AGENTS.md ## Rationale Use the plural `reviews` root because it matches the checked-in PR microsoft#2666 precedent and gives every review set a stable work-item identifier. Keep the PR and feature forms explicit in repository guidance, and remove the temporary singular-root handling rather than preserving two accepted layouts. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 476d113f-dd35-40c6-bc79-005dcccd7b79
## Summary Convert malformed UTF-8 or JSON HTTP 200 responses into verifier runtime errors that name the failing URL. Add regressions for HTML and truncated JSON bodies. ## Prompting Intent Resolve the current-head Copilot finding on PR microsoft#2628. Release operators must be able to identify which GitHub or Azure DevOps endpoint returned a malformed success response instead of receiving a context-free JSON decoder error. ## Linked Sources - GitHub pull request: microsoft#2628 - Copilot review finding: microsoft#2628 (comment) ## Rationale Catch only response-decoding failures at the HTTP helper boundary and retain the existing fail-closed behavior. Including the URL there gives every caller the missing backend context without logging request headers or credentials. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 476d113f-dd35-40c6-bc79-005dcccd7b79
## Summary Add sealed source-bound release plans, explicit queue approval, durable recovery state, producer evidence, and guarded notes/BBC-VHD handoffs. Harden tag workflows, release versioning, PyPI collisions and ESRP staging. Keep ordinary CI on snapshot coordinates. Add history-preserving publisher retries, persistent local ledger claims and bounded lock inspection. Keep same-coordinate Maven retries disabled without whole-namespace absence proof. ## Prompting Intent The engineer asked for safer, more effective end-to-end release automation, repeatable command-line steps and a release skill, while allowing Internal to publish independently of new OSS releases. Human approval, reviewed source, immutable packages and existing runtime compatibility must remain intact. ## Linked Sources - Public release automation: microsoft#2628 - Operator commands: scripts/release/README.md - Release skill: .github/skills/synapseml-release/SKILL.md - Review findings and resolutions: reviews/pr-2628/ - Repository rules: AGENTS.md ## Rationale Bind intent once and preserve it through queueing, recovery and verification instead of recalculating package coordinates at every step. Keep inventory separate from producer evidence and require original approvals for staged rollout. This permits independent Internal releases without weakening source, destination or immutability checks. This is release/build infrastructure; no library API or dependency pin changes are included. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 476d113f-dd35-40c6-bc79-005dcccd7b79
## Summary Use portable noninteractive SBT probe invocations. Preserve actionable blob inspection failures and cache each ESRP module directory's resolved path. Add failing-before regressions and retain the current-head review resolutions. ## Prompting Intent The engineer asked to continue the coordinated release improvements and commit them. The PR loop requires fixing current-head CI and review findings rather than treating historical passing runs as evidence for new code. ## Linked Sources - Public PR: microsoft#2628 - CI failure: https://github.com/microsoft/SynapseML/actions/runs/33964603972 - Blob diagnostic: microsoft#2628 (comment) - ESRP resolution: microsoft#2628 (comment) - Review and resolution history: reviews/pr-2628/pr-2628-attempt-2-review-6-claude-opus-5.md ## Rationale An explicit SBT task and closed stdin work with both native and wrapper launchers. Keep query failure distinct from artifact absence while preserving the underlying process error. Resolve each admitted source directory once without dropping containment checks. This is release infrastructure, not a library feature or dependency change. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 476d113f-dd35-40c6-bc79-005dcccd7b79
## Summary Add no-overwrite plan output, shared duplicate-rejecting plan decoding, and bounded read-only or explicitly approved release polling. Preserve durable state and restore unsent intent when its save exhausts the scheduling window. Keep fixed release tooling and fixtures unchanged across successive version bumps. Include focused regressions, operator documentation and review reports. ## Prompting Intent Refresh and complete the release automation PR for reliable future releases. Rebase onto the current target, address review findings, and commit evidence without bypassing approval or immutable-publication safeguards. Treat the requested confidence as an evidence standard, not a bug-free guarantee. ## Linked Sources - Release automation PR: microsoft#2628 - Operator contract: scripts/release/README.md - Release procedure: .github/skills/synapseml-release/SKILL.md - Branch and readiness rules: AGENTS.md and .github/skills/synapseml-pr-loop/SKILL.md - Review findings and resolutions: reviews/pr-2628/pr-2628-attempt-3-review-*.md ## Rationale Extend the existing plan-driven state machine rather than introduce another release coordinator. Waiting must not approve, retry, adopt, widen scope or cancel remote builds. Exclusive output and stable fixture exclusions keep successive releases separate without invalidating historical evidence. Shared strict decoding closes the producer admission gap as well as local plan reads. This is release/build infrastructure, covered by the PR's area/build label. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 879c5a1c-4efa-4620-93ea-f8f0052112f1
## Summary Recover missing Spark and Python derivative tags at the exact recorded merge commit of a reviewed release PR. Verify existing immutable tags and actual remote refs before reporting a branch complete. Preserve open PRs and reject unknown source commits instead of guessing a moving branch tip. Add native Git regression coverage for reruns, rewritten ancestry, advanced targets, partial annotated pairs, missing or invalid evidence, remote conflicts and rejected pushes. Add outgoing-request authentication tests without changing the already-correct production token construction. Include operator guidance and six follow-up review reports with their findings and dispositions. Require exact local and remote refs across all three tagging workflows and Maven admission. Stage selected tag objects in a unique local namespace and atomically map its literal names to remote tags, preventing suffix-lookalike branches from becoming write destinations. Clean up staging refs on failure and success. Bind preparation ancestry to the exact master branch object. ## Prompting Intent Complete and refresh the release automation PRs for repeatable future releases, with evidence rather than a bug-free guarantee. Address current-head review findings, retain immutable publication and human-approval boundaries, and commit the verified changes without merging or publishing a release. ## Linked Sources - Release automation PR: microsoft#2628 - Missing-tag review: microsoft#2628 (comment) - Authentication review: microsoft#2628 (comment) - Operator contract: scripts/release/README.md - Release and branch policy: .github/skills/synapseml-release/SKILL.md and AGENTS.md - Review findings and resolutions: reviews/pr-2628/pr-2628-attempt-3-review-*.md ## Rationale An already-merged branch does not prove that its asynchronous tag workflow finished. The recorded merge SHA is the authorization boundary for recovery; creating a tag at the current branch tip could include later unreleased work. Legacy releases without that record may only verify a consistent existing pair. Atomic non-forced pushes and remote verification fail closed on concurrent changes. Pattern refspecs avoid single-ref Git destination resolution, which could otherwise update an unrelated lookalike branch before verification fails. Redacted source rendering is not evidence of a broken token header, so the authentication follow-up adds a real request assertion rather than a speculative production edit. This is release/build infrastructure covered by the PR's area/build label. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 879c5a1c-4efa-4620-93ea-f8f0052112f1
## Summary Reject an existing output symlink before resolving its path, including a link whose target does not exist. Keep the existing Ivy containment, new-output and artifact validation rules. Add twelve API/CLI regressions covering relative and absolute links with missing, directory and file targets. Preserve link identity, target contents, cache contents and explicit failure output. Document the staging constraint and retain the review finding and its evidence. ## Prompting Intent Complete, refresh and commit the coordinated release automation PRs for safe future releases. Resolve current-head review findings with reproducible evidence without promising bug-free behavior or bypassing release approvals and CI gates. ## Linked Sources - Release automation PR: microsoft#2628 - Output-symlink finding: microsoft#2628 (comment) - Operator contract: scripts/release/README.md - Review evidence: reviews/pr-2628/pr-2628-attempt-3-review-1-gpt-6-astra.md ## Rationale Resolving the path first erases the link identity and can create a missing target outside the requested staging location. Checking the original path preserves the existing staging implementation and CLI behavior while failing closed for this case. Real filesystem tests reproduced four failures before the fix. Both Linux and Windows exercise the corrected error path without skipping symlink cases. This is release/build infrastructure covered by the PR's area/build label. Live publishing and branch-integration approval remain outside this change. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 879c5a1c-4efa-4620-93ea-f8f0052112f1
## Summary Include the public Maven Release job only when both release and artifact publication are enabled. Keep the initial release admission guard so incomplete requests fail with the intended diagnostic rather than an unknown dependency. Add four flag-combination regressions and record native Azure preview evidence. Clarify that the public Maven and official pip/UPack publishers have separate pipeline definitions and parameter contracts. ## Prompting Intent Complete, refresh and commit the coordinated release automation PRs for safe future releases. Resolve current-head findings with reproducible evidence without promising bug-free behavior or bypassing approvals and CI gates. ## Linked Sources - Release automation PR: microsoft#2628 - Dependency finding: microsoft#2628 (comment) - Separate-contract finding: microsoft#2628 (comment) - Operator contract: scripts/release/README.md - Review evidence: reviews/pr-2628/pr-2628-attempt-3-review-1-gpt-6-astra.md ## Rationale The artifact job is omitted at compile time when publishArtifacts is false. Omitting the dependent release job under the same condition preserves the guard and avoids constructing an invalid Azure graph. Valid release gating and ordinary artifact publication are unchanged. A real no-run preview reproduced the original failure; all four combinations now expand without missing dependencies, and the invalid request stops at the actual admission guard. The other finding confused two independently verified pipeline contracts, so changing the correctly declared publisher parameters would be a regression. This is release/build infrastructure covered by the PR's area/build label. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 879c5a1c-4efa-4620-93ea-f8f0052112f1
## Summary Update the two current release-tooling links to guide v3 and identify v2 as historical reference. Use the exact repository path rather than inventing a wiki page ID before the new page is published. Spell out the existing Maven coordinate in two newly merged tutorial references so the fail-closed version bumper recognizes them without broader matching. ## Prompting Intent The engineer requested a new release guide version while preserving the older guide. Keep current automation entry points directed to the new procedure so operators do not accidentally follow restored historical instructions. Preserve release automation across the required target refresh, including version references introduced by newly merged documentation. ## Linked Sources - Release automation PR: microsoft#2628 - Current command reference: scripts/release/README.md - Rollout reference: .github/skills/synapseml-release/references/recovery-and-rollout.md - Documentation review disposition: reviews/pr-2628/pr-2628-attempt-3-review-1-gpt-6-astra.md - Newly merged tutorial: microsoft#2701 ## Rationale Changing only the wiki navigation would leave the tooling's direct links pointing at historical v2. Updating both references keeps the current procedure consistent with the new version while preserving old permalinks. The link is explicitly gated on the guide change merging. No runtime or release behavior changes. The PR already carries the area/build label. The target refresh introduced two bare version references that the existing regression suite correctly rejected. Using the tutorial's already documented Maven coordinate fixes those references without changing version values or weakening the scanner's context requirement. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 879c5a1c-4efa-4620-93ea-f8f0052112f1
## Summary Adapt the incoming E2E pipeline contract to the release PR's conditional checkout. Assert full history and tags for release requests, preserve the ordinary two-commit checkout, and locate the single named impact step without depending on the release admission guard's position. Record the bounded rebase review and current validation evidence without replacing prior reports. ## Prompting Intent The engineer asked to restart the publishing automation work, beginning by updating PR microsoft#2628 with the latest master branch and rerunning CI. Preserve the existing release changes and master fixes without publishing version 1.2.0, creating release tags, or bypassing approvals. ## Linked Sources - Release automation PR: microsoft#2628 - Integrated master: microsoft@681bd96 - Incoming E2E impact selection: microsoft#2736 - Repository rules: AGENTS.md - Pipeline contract: tools/ci/tests/test_e2e_impact.py - Rebase review: reviews/pr-2628/pr-2628-attempt-3-review-1-gpt-6-astra.md ## Rationale Keep both upstream contracts rather than weakening the release guard or the E2E selector. The incoming assertion failed with KeyError for fetchDepth because the checkout now has explicit release and ordinary branches. Checking both branches and the unique impact step preserves the intended coverage without changing runtime behavior. Release and CI-helper contracts passed 959 cases and 63 subtests, with one explicit opt-in SBT check skipped. Native Windows version-bump/history coverage passed 228 cases. Black 22.3.0 accepted the changed test. Existing release review findings and full CI remain separate readiness gates. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 879c5a1c-4efa-4620-93ea-f8f0052112f1
|
/azp run |
|
Azure Pipelines: Successfully started running 1 pipeline(s). |
There was a problem hiding this comment.
Copilot review overview
🟡 Changes recommended
Unresolved release publication scoping and dependency issues, plus workflow-test coverage gaps, block approval.
Get a fresh assessment by requesting another Copilot review.
Review effort: Lite
Findings: 3
Open (8)
Maven-only plans trigger unauthorized non-Maven publication tasks · New Required staging arguments break the Spark 4.1 release pipeline Reject stale versioned files during ESRP staging Optional skipped dependencies can prevent release publication · New Review artifact uses a session-workspace path · New Review artifact exposes an absolute local session path · New Review artifact contains an absolute machine-local path · New Review artifact contains checkout-specific commands · New
| ${{ if eq(parameters.publishRelease, true) }}: | ||
| dependsOn: [BuildAndCacheSbt, Style, UnitTests, PythonTests, RTests, DatabricksCPUE2E, DatabricksGPUE2E, FabricE2E, WebsiteSamplesTests, BuildDocker] |
| - job: Publish | ||
| dependsOn: BuildAndCacheSbt | ||
| ${{ if eq(parameters.publishRelease, true) }}: | ||
| dependsOn: [BuildAndCacheSbt, Style, UnitTests, PythonTests, RTests, DatabricksCPUE2E, DatabricksGPUE2E, FabricE2E, WebsiteSamplesTests, BuildDocker] |
| - **Theme**: Architecture & Patterns | ||
| - **Mode**: sequential | ||
| - **Model**: gemini-3.8-flash | ||
| - **Artifact**: direct-pr-2628\reviews\pr-2628\pr-2628-attempt-2-review-2-gemini-3.8-flash.md |
| $root='C:\Users\singhrana\.copilot\session-state\16e6d9b2-ce73-41f9-9d38-9386edc5c48d\files\direct-pr-2628' | ||
| git --no-pager -C $root diff 8c7143875c843c649a817cf3e8ba9c7bee23689c -- build.sbt project\build.scala pipeline.yaml .github\workflows\pr-validation.yml |
| $root='C:\Users\singhrana\.copilot\session-state\16e6d9b2-ce73-41f9-9d38-9386edc5c48d\files\direct-pr-2628' | ||
| git --no-pager -C $root status --short | ||
| git --no-pager -C $root rev-parse HEAD |
| $root = 'C:\Users\singhrana\.copilot\session-state\16e6d9b2-ce73-41f9-9d38-9386edc5c48d\files\direct-pr-2628' | ||
| git --no-pager -C $root rev-parse HEAD | ||
| git --no-pager -C $root status --porcelain=v1 --untracked-files=all |



What changed
Make repeated OSS releases and independent Internal releases plan-driven and
resumable, with reviewed sources, explicit publication approval, durable
recovery records and immutable package coordinates.
September 22 refresh
Head:
3f7d9fcf376212b721fad69c3b8291e2c5cf142aIntegrated master:
681bd96990c421de3b91d2b1bf8f8f470764199dRebased all 24 original commits and preserved their complete messages. The
conflict resolutions retain master's current review-artifact guidance and
two-commit checkout for E2E impact detection, alongside the release PR's full
history/tag checkout and publication admission guards.
The incoming E2E pipeline contract initially failed with
KeyError: 'fetchDepth'.A test-only follow-up now checks both checkout modes and the unique named impact
step. No release implementation or release workflow changed during this refresh.
Fresh local evidence:
SYNAPSEML_TEST_RELEASE_SBT=1;no fresh SBT runtime result is claimed.
to the tested bytes. Incoming master changes outside the original PR paths
were preserved.
The bounded direct integration review is appended to
the existing follow-up report.
Earlier multi-model reviews and Azure preview results remain historical evidence,
not current-head CI or review approval.
Remaining gates
Fresh current-head Azure merge validation and automated review are being requested
after this update. Their absence or an old-head result is not a pass. See the
current Checks and review threads for results.
These existing findings still require disposition:
use an OSS-only plan for
master,spark4.0,spark4.1and a separate Internalplan for
master,spark4.1; the combined default still includes Internal Spark 4.0.Human last-push approval and applicable companion-pipeline gates remain.
No release tags, production publication, approvals or rollout changes were
performed. This PR is not being declared merge-ready by the refresh.
Same-coordinate Maven retry remains unsupported.
Documentation and companion changes