feat: a read-optimized FactStore

[hertzcodes]

Hello, regarding the issue: #65
I have studied about FactStore and implemented a semi-columnar storage for better reads in big queries. Write still needs optimizations and I'm thinking of it as of now. You can run the benchmarking tests in benchmark_test.go in factstore package. @burakemir I'm not sure if you meant I shouldn't have used maps or not but this is the benchmark so far. I'd appreciate if you could guide me through.

goos: linux
goarch: amd64
pkg: github.com/google/mangle/factstore
cpu: Intel(R) Core(TM) i7-1065G7 CPU @ 1.30GHz
BenchmarkAdd/factstore.SimpleInMemoryStore-8             1798400               655.9 ns/op           136 B/op          6 allocs/op
BenchmarkAdd/factstore.IndexedInMemoryStore-8            1676612               655.9 ns/op           136 B/op          6 allocs/op
BenchmarkAdd/factstore.MultiIndexedInMemoryStore-8       1230781               935.8 ns/op           184 B/op          7 allocs/op
BenchmarkAdd/*factstore.MultiIndexedArrayInMemoryStore-8                 1000000              1575 ns/op             202 B/op          7 allocs/op
BenchmarkAdd/factstore.ConcurrentFactStore-8                             1847586               646.7 ns/op           136 B/op          6 allocs/op
BenchmarkAdd/*factstore.ColumnarStore-8                                  1000000              1191 ns/op             173 B/op          6 allocs/op
BenchmarkGetFacts/factstore.SimpleInMemoryStore-8                        1991113               604.7 ns/op            86 B/op          4 allocs/op
BenchmarkGetFacts/factstore.IndexedInMemoryStore-8                       3648225               330.2 ns/op            86 B/op          4 allocs/op
BenchmarkGetFacts/factstore.MultiIndexedInMemoryStore-8                  3327081               399.7 ns/op            86 B/op          4 allocs/op
BenchmarkGetFacts/*factstore.MultiIndexedArrayInMemoryStore-8            3412228               335.9 ns/op            86 B/op          4 allocs/op
BenchmarkGetFacts/factstore.ConcurrentFactStore-8                        2024418               624.5 ns/op            86 B/op          4 allocs/op
BenchmarkGetFacts/*factstore.ColumnarStore-8                             3007182               470.8 ns/op            86 B/op          4 allocs/op
BenchmarkGetFacts_BigQuery/factstore.SimpleInMemoryStore-8                     8         132573589 ns/op              96 B/op          2 allocs/op
BenchmarkGetFacts_BigQuery/factstore.IndexedInMemoryStore-8                   15          76971862 ns/op              96 B/op          2 allocs/op
BenchmarkGetFacts_BigQuery/factstore.MultiIndexedInMemoryStore-8              10         108997991 ns/op              96 B/op          2 allocs/op
BenchmarkGetFacts_BigQuery/*factstore.MultiIndexedArrayInMemoryStore-8                 8         146630297 ns/op              96 B/op          2 allocs/op
BenchmarkGetFacts_BigQuery/factstore.ConcurrentFactStore-8                             8         137718846 ns/op              96 B/op          2 allocs/op
BenchmarkGetFacts_BigQuery/*factstore.ColumnarStore-8                                 52          20060120 ns/op              96 B/op          2 allocs/op
BenchmarkMerge/SimpleInMemoryStore-8                                               22758             51623 ns/op           36278 B/op        227 allocs/op
BenchmarkMerge/ColumnarFactStore-8                                                  2590            484880 ns/op         1054459 B/op       1109 allocs/op
PASS
ok      github.com/google/mangle/factstore      70.689s

[burakemir]

Nice job on doing a new factstore implementation! I probably wouldn't call it columnar but a column-indexed list store or something, but the benchmark show that it can be faster in some scenarios and that seems useful.

[hertzcodes]

yeah I didn't know what to call it either tbh. what do you think? this got potential to be better? I have actually searched about columnar databases for memory. I tried apache arrow itself but it wasn't good at all (2500ns/op). I found another go library https://github.com/kelindar/column it's using faster maps but I did not make an implementation. you think it's worth a try?