Skip to content

Add shuffler disk residency for received chunks - #1196

Draft
nirandaperera wants to merge 8 commits into
rapidsai:mainfrom
nirandaperera:disk-spilling-shuffler
Draft

Add shuffler disk residency for received chunks#1196
nirandaperera wants to merge 8 commits into
rapidsai:mainfrom
nirandaperera:disk-spilling-shuffler

Conversation

@nirandaperera

Copy link
Copy Markdown
Contributor

Received shuffle chunks could only live in device or host memory, so host pressure had no further spill target. Chunks now keep an exclusive in-memory or disk payload, spill device data to host or disk, and restore from disk on extract.

  • Spill device chunks to host when available, otherwise to disk; restore on extract (device, then host, with host overbooking as a last resort).
  • Keep disk-resident chunks off the send path (is_ready() is false; ChunksToSend rejects them).
  • Cover chunk spill/restore, file cleanup, and shuffler insert/extract disk paths in tests.

Depends on #1195
Closes #1185
Related #1170

Signed-off-by: niranda perera <niranda.perera@gmail.com>
@copy-pr-bot

copy-pr-bot Bot commented Sep 3, 2026

Copy link
Copy Markdown

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

- Track pageable-host allocations through the generalized resource adaptor.
- Add an absolute, unbounded-by-default host memory limit.
- Reconcile pageable and pinned-host budgets and reject invalid pinned limits.
- Extend statistics, Python bindings, documentation, and tests.

Signed-off-by: niranda perera <niranda.perera@gmail.com>
- Spill device chunks to host or disk and restore them on extract.
- Keep disk payloads off the send path and reject send-ready disk chunks.

Signed-off-by: niranda perera <niranda.perera@gmail.com>
@nirandaperera nirandaperera added breaking Introduces a breaking change improvement Improves an existing functionality non-breaking Introduces a non-breaking change and removed breaking Introduces a breaking change labels Sep 3, 2026
@nirandaperera nirandaperera self-assigned this Sep 3, 2026
@nirandaperera

Copy link
Copy Markdown
Contributor Author

/ok to test

Signed-off-by: niranda perera <niranda.perera@gmail.com>
Signed-off-by: niranda perera <niranda.perera@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

improvement Improves an existing functionality non-breaking Introduces a non-breaking change

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Use disk buffer in the shuffler path

1 participant