First, install glp. We use this to manage dependency pinning for this
project. After you get the dependencies using glp sync, usage is similar to the Go tool itself.
# Build the server
glp build -o bin/server server
# Run all the tests:
glp test ./...
- Use arrays for storing results for low-cardinality group-bys, and hashmaps for high-cardinality group-bys (assuming there's a large performance difference between arrays and hashmaps in the low-cardinality case, which is the common case).
- Expose metrics via HTTP routes so that summary metrics of GumshoeDB's data set are easy to inspect.
The benchmark suite is a critical tool for evaluating different implementation strategies.
To run:
glp test -run=NONE -bench=. gumshoe synthetic
The synthetic suite benchmarks small, narrow techniques and represents the upper-bound of performance. It provides a clean, isolated view on how fast a technique is.
The core benchmarks test the core GumshoeDB code paths. The core code paths should be comparable in speed to the ideal benchmarks -- ideally within 20%.
- Iterating over two-dimensional slices is twice as slow as contiguous slices/arrays, because of pointer indirection.
- Scan speed scales linearly with the bit-width of the row
The query API JSON format is inspired by Druid's.