Skip to content

Remove legacy code; serialize every sketch as ASAPv1 only #23

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
zzylol wants to merge 2 commits into
base: feat/cmsheap-topk-demo
Choose a base branch
from
Open

Remove legacy code; serialize every sketch as ASAPv1 only #23

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
88d696c
Remove file-based demo and legacy surfaces
zzylol Oct 12, 2026
974b425
Serialize every sketch as ASAPv1; let every sketch type transmit
zzylol Oct 12, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 1 addition & 4 deletions .dockerignore
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,4 @@
.cache
.git
benchmark-results
asap-precompute-rs/target/*
!asap-precompute-rs/target/release/
asap-precompute-rs/target/release/*
!asap-precompute-rs/target/release/asap-otap-demo
asap-precompute-rs/target
57 changes: 20 additions & 37 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,8 @@ ASAP's [OTAP Dataflow](https://github.com/open-telemetry/otel-arrow/tree/main/ru
processor.

It ports ASAP's edge **precompute + sketching** runtime (windowed
DDSketch / KLL / HLL / CountSketch / CountMin-Sketch aggregation) onto
DDSketch / KLL / HLL / CountSketch / Count-Min / Count-Min-with-heap
aggregation) onto
OTAP Dataflow's Arrow-native streaming model as a single unified
`asap_sketches` processor.

Expand All @@ -17,8 +18,8 @@ overlay step:

| Feature | What it adds | Extra dependency |
|---|---|---|
| *(default, no features)* | Host-neutral precompute runtime — the five sketch types, windowing, overflow policy. No Arrow, no OTAP, no Tokio. | none |
| `otap` | The standalone Arrow codec (`encode_batch`/`decode_batch`, `OtapMetricRecords`). | `arrow-array`, `arrow-schema` |
| *(default, no features)* | Host-neutral precompute runtime — the sketch types, windowing, overflow policy. No Arrow, no OTAP, no Tokio. | none |
| `otap` | Plugin config (`otap::config`) and the `sketch.*` carrier keys. | `arrow-array`, `arrow-schema` |
| `otap-engine` | The real OTAP node: `AsapSketchesProcessor` (a genuine `local::Processor<OtapPdata>`, `linkme`-registered) and its direct `SketchEnvelope <-> OtapPdata` binding (`otap::codec`). | the real `otel-arrow-dfe-*` crates, via a plain git dependency pinned to a specific `open-telemetry/otel-arrow` commit — see `Cargo.toml` |

`otap-engine`'s git dependency is what used to require staging this
Expand All @@ -44,11 +45,10 @@ same way it already does for the `asap_sketchlib` git dependency —
`SketchEnvelope <-> OtapPdata` binding: implements OTAP's own
`MetricsView` trait family straight over `&[SketchEnvelope]`
(encode) and decodes a real `OtapPdata` straight into
`Vec<Observation>` (decode). No intermediate `RecordBatch` or
`OtapMetricRecords` two-batch hop — those exist as a general,
real-OTAP-free carrier format for other potential Strategy-B
adapters (Telegraf, Vector), but this repo only ever ships to OTAP,
so the OTAP-facing path skips straight to the real type.
`Vec<Observation>` (decode). A sketch record's `sketch.envelope` is
the sketch's full state as the self-describing ASAPv1 envelope from
`asap_sketchlib`; the codec rejects bytes that do not decode as the
declared `sketch.type`. There are no deltas.

**There is exactly one transport: the pipeline.** `AsapSketchesProcessor`
sends only via `effect_handler.send_message_with_source_node` — an
Expand All @@ -70,38 +70,28 @@ pipeline", for the full rationale and the real schema this produces.
`cargo fmt --check`, against `open-telemetry/otel-arrow` commit
`3e85c346` (2026-08-24, pinned in `Cargo.toml`).

## Configuration + standalone Arrow codec
## Configuration

| File | Responsibility |
|---|---|
| [`config.rs`](./asap-precompute-rs/src/otap/config.rs) | `PluginConfig` + `resolve()` — the 5-sketch `sketch_type` dispatch table (factory + observer + `SketchType`). |
| [`decode.rs`](./asap-precompute-rs/src/otap/decode.rs) / [`encode.rs`](./asap-precompute-rs/src/otap/encode.rs) | Standalone Arrow codec: `RecordBatch ↔ Vec<Observation>` / `[SketchEnvelope]`, keyed on well-known columns + Strategy-B `_asap_*` carrier keys ([`schema.rs`](./asap-precompute-rs/src/otap/schema.rs)). |
| [`records.rs`](./asap-precompute-rs/src/otap/records.rs) | `OtapMetricRecords` + `flatten`/`lift` — projects an `OtapArrowRecords`-shaped sibling-batch family (metrics + per-row attribute child batch joined by `parent_id`) ↔ flat `RecordBatch`. Real-OTAP-free; same scope note as above. |

All of this sits on the host-neutral runtime in the same crate
(`precompute`, `window`, `snapshot_cache`, `sketches/*`).
[`config.rs`](./asap-precompute-rs/src/otap/config.rs) holds `PluginConfig`
and `resolve()` — the `sketch_type` dispatch table (factory + observer +
`SketchType`). All of this sits on the host-neutral runtime in the same
crate (`precompute`, `window`, `sketches/*`).

## Build & test

For a presentation-ready walkthrough of the real YAML pipeline integration,
see the [live OTAP pipeline demo guide](./docs/demo-guide.md). The guide also
includes a [Prometheus + Grafana setup](./docs/demo-guide.md#write-kll-estimates-to-prometheus-and-view-them-in-grafana)
that writes the KLL p50/p99 estimates to Prometheus and displays them in a
provisioned Grafana dashboard. For a continuous KLL vs exact comparison with
every processor in its own pinned process, see the
[multi-process streaming KLL demo](./docs/kll-windows.md).
For a continuous KLL vs exact comparison with every processor in its own
pinned process, see the [multi-process streaming KLL demo](./docs/kll-windows.md).

```sh
cd asap-precompute-rs
cargo build --features otap-engine
cargo test --features otap-engine # runtime + codec + lifecycle + real pipeline tests
cargo clippy --all-targets --features otap-engine -- -D warnings
cargo test --features otap-engine,stream-benchmark # runtime + codec + real pipeline tests
cargo clippy --all-targets --features otap-engine,stream-benchmark -- -D warnings
cargo fmt --check
```

The default (no-feature) build excludes Arrow/Tokio and compiles the
row-oriented runtime only; `otap` turns on the Arrow codec + legacy
transport + plugin lifecycle; `otap-engine` additionally turns on the
row-oriented runtime only; `otap` turns on the plugin config;
`otap-engine` additionally turns on the
real OTAP node (and pulls in the real OTAP Dataflow crates — a heavier,
slower build: datafusion, tonic, and the rest of that dependency tree).

Expand All @@ -114,20 +104,13 @@ a specific commit in `Cargo.toml`.

## Status

Phases **B** (Arrow codec) and **C** (five-sketch runtime) are complete and
tested. Phase **D** (the `linkme` registration + the
The sketch runtime is complete and tested. Phase **D** (the `linkme` registration + the
real `OtapPdata` binding) is complete and build/lint/test-verified by
this repo's own build. `tests/otap_pipeline_e2e.rs` parses a genuine
OTAP pipeline YAML, builds it through `OTAP_PIPELINE_FACTORY`, runs the
resulting `RuntimePipeline`, injects real scalar `OtapPdata`, and verifies a
two-source, four-processor creation → fan-in merge → estimation flow at the
configured exporter.
Run the same live pipeline as a terminal demo with:

```sh
cd asap-precompute-rs
cargo run --bin asap-otap-demo --features otap-engine
```

The real OTAP adapter merges an upstream processor's self-describing sketch
directly from `OtapPdata`. The multi-process demo uses standard OTLP metrics
Expand Down
Loading