Repository navigation
doc(ingest): sidecars as views over the ingested rows, staged locally and uploaded - #526
Open
tanejagagan wants to merge 1 commit into
Open
tanejagagan wants to merge 1 commit into
tanejagagan wants to merge 1 commit into
Conversation
… and uploaded Design spec replacing the producer-side SidecarRows mechanism. A queue declares any number of sidecars, each a table and a SQL transformation over the batch's base rows (__base). With an uploader configured, the base is written to a local staging directory and its upload starts at once; each sidecar is derived from that staged Parquet and uploaded as it is written; one DuckLake transaction then registers every file. By default no uploader is configured, and parallel DuckDB COPYs write the base and every sidecar directly, as the base is written today. Includes the benchmark behind the decisions (tools/sidecar-bench), run on DuckDB 1.5.6 and 2.0.0 against MinIO with ~40 ms of injected latency, and the uploader's library, authentication and configuration. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Design spec,
doc/SIDECAR_VIEWS_SPEC.md, replacing the producer-sideSidecarRowsmechanism specified indoc/ATTRIBUTE_TYPE_REGISTRY_SPEC.md. Sidecars become configuration instead of producer code: each is a table plus a SQL transformation over the batch's base rows (__base), and a queue may declare any number of them.Write path
COPYs straight to the destination, as the base is written today. A queue with no sidecars behaves exactly as it does now.Decisions, all measured
s3+stson its JDK-based HTTP client, without the transfer manager or Netty. That's 8.2 MB in 32 jars, against 13.7 MB with Netty and 22.9 MB for the MinIO SDK.PROVIDER credential_chain) and the uploader (the AWS default chain) read one identity from the environment instead, e.g. the pod's IAM role via IRSA or EKS Pod Identity.Benchmark
tools/sidecar-benchis a standalone Maven project, not a module of the build. It writes one batch as a base table plus three sidecars, to MinIO behind a proxy adding about 40 ms per round trip. Median ms:COPYs (the default)The spec has all seven strategies. In every run, all strategies wrote the same rows per output, and each output was read back from S3 and counted.
On DuckDB 2.0.0, direct writes are faster for large batches but about 2× slower for small ones: its HTTP log shows two extra
HEADrequests per file. The staged path doesn't use DuckDB's S3 writer and is unaffected.The numbers come from one machine with synthetic data, so treat them as directional until they're repeated against a real bucket. That and the other open questions are listed at the end of the spec.
🤖 Generated with Claude Code