v0.1.2

What's Changed

• [BREAKING] refactor(server): forward per-request bearer to FlowMesh by @timzsu in #25
• build(deps): Bump docker/build-push-action from 6.17.0 to 7.1.0 by @dependabot[bot] in #26
• build(deps): Bump docker/login-action from 3.5.0 to 4.1.0 by @dependabot[bot] in #27
• feat(cli,sdk): add api_key auth matching FlowMesh's init/deinit/config contract by @timzsu in #28
• fix: Relax lumid-hooks version. by @timzsu in #29

Full Changelog: v0.1.1...v0.1.2

Published image for v0.1.2

Image	Digest
ghcr.io/mlsys-io/lumilake_server:v0.1.2	sha256:4d5fc5fe35adb03811e1506af8bb425e93cad7733d34671e8d52fd68a8520238

Built from commit 3f327216475a452832e0ecd5880494b73a172f3d (multi-arch: linux/amd64, linux/arm64).

v0.1.1

Important performance fixes plus new debug toggles.

• Optimizer visibility and toggles — expose per-stage timing fields (optimization_seconds, selection_seconds, clustering_seconds), add LUMILAKE_DISABLE_DATA_PROFILE, and stop preview truncation. (#15)
• Deploy-time plugins and image purge — mount ${LUMILAKE_PLUGIN_DIR} into the server at /app/plugins and add lumilake deploy purge <image_tag>. (#15, #16)
• Multi-batch data profiling stability — fix batch-2 lookup, stabilize DSL ids across slices, propagate DP failures, drop stale FlowMesh keep-alive sockets, and resolve S3 buckets through S3_URL_PREFIX. (#22)
• HALO cold-cache regression closed — recalibrate vLLM model-switch cost, coalesce same-request slices into one parent workflow, and promote topological-sort to a first-class optimizer. (#14)
• Security — bump starlette to clear PYSEC-2026-161 and idna to clear GHSA-65pc-fj4g-8rjx. (#22, #23)

Full Changelog: v0.1.0...v0.1.1

Published image for v0.1.1

Image	Digest
ghcr.io/mlsys-io/lumilake_server:v0.1.1	sha256:426fe82253af94e3ebbabc96aba81e893f497ada481861e87d21e83c441494d8

Built from commit bc8bbf21b2481db7a70b696818dab3d620671e9c (multi-arch: linux/amd64, linux/arm64).

v0.1.0

Initial open-source release of Lumilake.

Includes

• Workflow parsing for YAML, n8n exports, and native graph payloads.
• HALO runtime graph optimization.
• FlowMesh-backed local execution through the deploy helper.
• CLI and Python SDK for deployment, job submission, workers, traces, and results.
• Direct PostgreSQL and S3-compatible data retrieval.
• lumid.data agent retrieval support.
• Shared hook integration through lumid-hooks.

Packages

• lumilake
• lumilake-sdk
• lumilake-cli
• lumilake-deploy
• lumilake-hook

The Lumilake server runtime is published as a container image:

• ghcr.io/mlsys-io/lumilake_server:v0.1.0

Published image for v0.1.0

Image	Digest
ghcr.io/mlsys-io/lumilake_server:v0.1.0	sha256:9cf2b0a246b3776b8228bdf0bb30c18311086e7b349b3c6801389e972189b3df

Built from commit e00da52979ea048ce15d5dcc6a7b35f98fe63fb4 (multi-arch: linux/amd64, linux/arm64).

Lumilake demo dataset v1

The reproducible demo dataset that scripts/dev/load_demo_data.py fetches by default (--tag demo-data-v1). Required by the three workflow pairs shipped under examples/templates/ — see docs/E2E_DEMO.md for the full reproduction recipe.

This dataset is versioned independently of the Lumilake code releases (vX.Y.Z) so a code patch doesn't re-upload 1 GB of unchanged data.

Assets

File	Size	Contents
lumilake-demo-pg.dump	33 MiB	pg_dump -Fc of the lumilake_demo schema
lumilake-demo-s3.tar.gz	614 MiB	news/{html,images} objects (1,512 files)

What's in the dataset

lumilake_demo Postgres schema — 100 symbols, 756 news rows, dates 2024-10-30 → 2025-05-02:

Table	Rows / Size	Used by
lumilake_demo.ohlc_10m	271 MB	trading-agent, agent-retrieval
lumilake_demo.financial_income_statement	6.1 MB	trading-agent, agent-retrieval
lumilake_demo.market_metrics	4.8 MB	trading-agent, agent-retrieval
lumilake_demo.insider_sentiment	1.1 MB	trading-agent, agent-retrieval
lumilake_demo.news_metadata	1.1 MB / 756 rows	all three workflows
lumilake_demo.instrument_profile	296 KB	trading-agent, agent-retrieval

S3 layout under <bucket>/example-data/news/:

• html/{id}.html — full article bodies (756 files, 26 MiB).
• images/{id}.png — featured images (756 files, 656 MiB).

{id} matches lumilake_demo.news_metadata.id; the Postgres and S3 halves are loaded as a unit.

Loading

From a Lumilake source checkout with the bundled data plane up:

docker compose -f scripts/dev/compose.data-plane.yml up -d
uv run python scripts/dev/load_demo_data.py

The loader restores the schema with pg_restore, uploads the news objects via the minio Python client, and short-circuits the download if ~/.cache/lumilake-demo/ already holds the assets.

License

Demo content; not for redistribution outside the Lumilake demo flow.