|
| 1 | +# Try asap-planner on a SQL Workload |
| 2 | + |
| 3 | +Check whether your SQL workload is amenable to ASAPQuery acceleration by running just the |
| 4 | +planner. |
| 5 | + |
| 6 | +## 1. Build |
| 7 | + |
| 8 | +```bash |
| 9 | +cargo build --release -p asap_planner |
| 10 | +``` |
| 11 | + |
| 12 | +This only compiles `asap-planner-rs` and its small internal deps — not the full query engine. |
| 13 | + |
| 14 | +## 2. Write a workload config |
| 15 | + |
| 16 | +```yaml |
| 17 | +# sql_workload.yaml |
| 18 | +tables: |
| 19 | + - name: metrics_table |
| 20 | + time_column: time |
| 21 | + value_columns: [cpu_usage] |
| 22 | + metadata_columns: [hostname, datacenter, region] |
| 23 | +query_groups: |
| 24 | + - id: 1 |
| 25 | + repetition_delay: 300 # seconds between repeats of this query |
| 26 | + controller_options: |
| 27 | + accuracy_sla: 0.95 |
| 28 | + latency_sla: 100.0 |
| 29 | + queries: |
| 30 | + - >- |
| 31 | + SELECT avg(cpu_usage) FROM metrics_table |
| 32 | + WHERE time BETWEEN DATEADD(s, -300, NOW()) AND NOW() |
| 33 | + GROUP BY datacenter |
| 34 | +aggregate_cleanup: |
| 35 | + policy: read_based |
| 36 | +``` |
| 37 | +
|
| 38 | +List each recurring query your application actually runs, with its real repeat interval. |
| 39 | +
|
| 40 | +### Column fields |
| 41 | +
|
| 42 | +For each table: |
| 43 | +- `time_column` — the column your queries filter/bucket on (e.g. in a `WHERE time BETWEEN ...` clause). |
| 44 | +- `value_columns` — the numeric columns being aggregated (e.g. the column passed to `avg()`, `sum()`, etc.). |
| 45 | +- `metadata_columns` — every other column your queries `GROUP BY` or filter on (dimensions like `hostname`, `datacenter`, `region`). List all of them, not just the ones used in these queries' `GROUP BY`. The planner uses the full set to figure out which dimensions to roll up vs. keep grouped. |
| 46 | + |
| 47 | +### `controller_options` |
| 48 | + |
| 49 | +`accuracy_sla` and `latency_sla` are required by the config schema but not currently used by |
| 50 | +the planner's decision logic — any numeric values are fine (e.g. the placeholders in the |
| 51 | +example above). |
| 52 | + |
| 53 | +### Choosing `repetition_delay` and `--data-ingestion-interval` |
| 54 | + |
| 55 | +- `--data-ingestion-interval` is how often new rows actually land in the table (your ingestion |
| 56 | + cadence) — e.g. `15` if a new batch/row arrives every 15 seconds. |
| 57 | +- `repetition_delay` is how often this specific query actually re-runs in your application — |
| 58 | + e.g. `300` for a dashboard panel that refreshes every 5 minutes. |
| 59 | +- **Constraint:** `repetition_delay` must be an exact multiple of `--data-ingestion-interval`, |
| 60 | + or the planner errors out. |
| 61 | + |
| 62 | +## 3. Run the planner |
| 63 | + |
| 64 | +```bash |
| 65 | +asap-planner --query-language sql \ |
| 66 | + --input_config sql_workload.yaml \ |
| 67 | + --output_dir ./out \ |
| 68 | + --data-ingestion-interval 15 \ |
| 69 | + --streaming_engine precompute \ |
| 70 | + -v |
| 71 | +``` |
| 72 | + |
| 73 | +- `--data-ingestion-interval` is the expected data ingestion cadence in seconds (required for SQL mode). |
| 74 | +- `--streaming_engine` just needs a valid value (`precompute`, `arroyo`, or `flink`) — none are actually started. |
| 75 | +- `-v` logs which queries were skipped and why. |
| 76 | + |
| 77 | +## 4. Read the result |
| 78 | + |
| 79 | +The planner writes `streaming_config.yaml` and `inference_config.yaml` to `./out`. |
| 80 | + |
| 81 | +- Queries that show up there as aggregations are ones ASAPQuery can accelerate. |
| 82 | +- Queries silently skipped (visible with `-v`) are not currently supported — e.g. unsupported |
| 83 | + SQL shapes or queries with no inferable repeat pattern. |
| 84 | + |
| 85 | +If most of your workload appears in `streaming_config.yaml`, ASAPQuery is likely a good fit. |
0 commit comments