Skip to content

Partitioned processing#15

Draft
sharkinsspatial wants to merge 22 commits into
mainfrom
partitioned_processing
Draft

Partitioned processing#15
sharkinsspatial wants to merge 22 commits into
mainfrom
partitioned_processing

Conversation

@sharkinsspatial

Copy link
Copy Markdown
Member

No description provided.

sharkinsspatial and others added 22 commits June 17, 2026 15:17
Design for a local pytest spike that de-risks the core fork/merge
mechanics from issue #12 before building Step Functions infrastructure.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
All forks must descend from the single session that performs the merge,
so the coordinator owns fork creation and the merge. Removes the
independently-created-fork hypothesis and fallback framing.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Scratch verification while preparing the plan showed merge silently
last-writer-wins on overlapping chunk keys (conflict detection is a
rebase-against-branch-tip concern, not a sibling-fork-merge one).
Fix assertion #4 to a characterization test, record confirmed findings,
and capture the verified zarr array-creation mechanism.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The pre-commit mypy hook runs in an isolated env without icechunk, so
commit() resolves to Any there and warn_return_any flags a bare return.
Add a comment so the cast is not mistakenly removed.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Also record the storage finding in the spec: backfill requires durable
(filesystem/S3) storage because a pickled ForkSession cannot resolve its
base snapshot from in-memory storage.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Address A-Task 2 code-review minors: document the not-idempotent
constraint in the Protocol docstring and assert chunk/dtype so a geometry
regression is caught before the Task 3 round-trip.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Address A-Task 3 code-review minor: create_fork now uses keyword-only
branch, matching merge_and_commit and promote, before sub-project B
consumers are written against the API.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Address final-review minors for sub-project A: the region_for Protocol
docstring now signals that the file_key scheme is implementation-defined,
and process_backfill_file explains why it catches broad exceptions.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant