Repository navigation
Capabilities removed with the old backend that are not replaced yet (#818) #823
Description
Problem, by example
On backend main, a SQL query sent to the ClickHouse listener is accelerated by asap_clickhouse_query_engine, OTel metrics are ingested through asap_otel_proto, and stored sketches persist to S3 through the Gorilla object store. After the stack #818 → #822 lands, none of these exist: the backend executes ASAPPlanner's selected plan for PromQL over Prometheus Remote Write only, with state in memory.
This issue records every capability #818 deletes that has no replacement yet, so it can be added back on the new architecture when needed. Replaced capabilities are listed in #818's deletion table.
Deferred capabilities
| Capability | Deleted from | When it comes back |
|---|---|---|
| SQL / ClickHouse query acceleration and exact ClickHouse plans | data_plane/src/query_engines/asap_clickhouse_query_engine, control_plane/src/clickhouse.rs, control_plane/src/query_plan/clickhouse_exact.rs |
After the PromQL path; the Planner already plans SQL |
| SQL evaluation tools | tools/clickhouse-benefit, tools/o11y-sql-main-eval |
With SQL support |
| OTel metric ingestion (incl. the vendored sketch-metric proto) | data_plane/src/drivers/ingest/otel.rs, crates/asap_otel_proto |
When a deployment ingests OTel |
| VictoriaMetrics query adapter and forwarding | data_plane/src/drivers/query/adapters/victoriametrics_http.rs, .../fallback/victoriametrics.rs |
When a VictoriaMetrics deployment is targeted |
| Persistent sketch storage (S3 / Gorilla object store) and recovery on restart | data_plane/src/storage_engines/sketch_db/persistence |
Out of scope now (state is memory only) |
| Backfill of stored summaries from ClickHouse or Prometheus history | data_plane/src/storage_engines/sketch_db/backfill |
With persistence |
| Late-sample revisions of built panes | data_plane/src/precompute_engine/revisions.rs |
If dropped late samples matter in practice (they are counted) |
| Multi-worker precompute (series partitioning, sub-DAG scheduler) | data_plane/src/precompute_engine/{partitioning,series_router,subdag_scheduler,worker}.rs |
When one engine thread is the bottleneck (the Planner's DAG type is not Send today) |
| Update sampling (sampling coordinator, epochs, allocation) | data_plane/src/update_sampling |
Not planned; reopen if needed |
| OpAMP remote control, runtime monitoring and online re-planning | control_plane/src/opamp, control_plane/src/emit/monitor.rs, control_plane/src/physical/online.rs, control_plane/src/runtime_samples.rs, control_plane/src/metrics_exposer.rs |
Re-planning is a non-goal now; runtime feedback is ASAPPlanner #250 |
| Container images and compose files | control_plane/Dockerfile, data_plane/Dockerfile, data_plane/*.yml.j2, demos/asapquery |
With the benefit experiment harness |
| Google / Alibaba cluster-trace Prometheus exporter | tools/cluster-data-exporter |
If trace replay is used in experiments (the first experiment uses a generator) |
| Python package | pyproject.toml and its sources |
Not planned |
Scope
Tracking only. Each row becomes its own issue or PR when it is picked up, built on the installed PhysicalASAPDAG, never on the deleted code.
Activity