Benchmarks
Apache SpatialBench
Run on a single m7i.2xlarge (8 vCPU, 32 GB), the same hardware used by Apache SpatialBench. PyCanopy is measured with index_mode="auto".
PyCanopy is fastest on 11/24 testcases (there is some variance among benchmark runs).
SF1 (~6M trips)
Apache SpatialBench SF1 · lower is better · bars past the cap truncated with their value · TIMEOUT / ERROR annotated
| Query | PyCanopy | SedonaDB | DuckDB | GeoPandas |
|---|---|---|---|---|
| q1 | 1.05 | 0.54 | 0.76 | 12.92 |
| q2 | 2.01 | 1.70 | 2.00 | 14.94 |
| q3 | 0.80 | 0.58 | 1.61 | 13.63 |
| q4 | 4.99 | 4.60 | 3.82 | 18.35 |
| q5 | 1.18 | 3.92 | 1.85 | 49.75 |
| q6 | 4.66 | 3.12 | 4.24 | 19.74 |
| q7 | 1.35 | 1.14 | 1.30 | 136.98 |
| q8 | 0.71 | 0.73 | 1.21 | 15.72 |
| q9 | 0.15 | 0.11 | 0.26 | 0.38 |
| q10 | 5.92 | 6.52 | 203.77 | 43.20 |
| q11 | 6.91 | 10.70 | 360.37 | 47.39 |
| q12 | 3.48 | 10.15 | TIMEOUT | TIMEOUT |
SF10 (~60M trips)
Apache SpatialBench SF10 · lower is better · bars past the cap truncated with their value · TIMEOUT / ERROR annotated
| Query | PyCanopy | SedonaDB | DuckDB | GeoPandas |
|---|---|---|---|---|
| q1 | 5.33 | 2.29 | 4.78 | TIMEOUT |
| q2 | 4.28 | 5.84 | 5.08 | TIMEOUT |
| q3 | 4.67 | 3.65 | 4.75 | TIMEOUT |
| q4 | 8.52 | 5.88 | 6.14 | OOM |
| q5 | 11.86 | 23.85 | 12.32 | OOM |
| q6 | 10.12 | 11.22 | 8.78 | ERROR |
| q7 | 11.11 | 7.39 | 5.91 | ERROR |
| q8 | 5.71 | 5.29 | 6.85 | ERROR |
| q9 | 0.23 | 0.25 | 0.50 | ERROR |
| q10 | 29.49 | 38.29 | TIMEOUT | ERROR |
| q11 | 32.30 | 61.19 | TIMEOUT | ERROR |
| q12 | 38.59 | 117.50 | TIMEOUT | ERROR |
All times in seconds. Bold = fastest on that query. Every engine was measured by the PyCanopy harness against the pinned SpatialBench workload.