[Perf] Cache schema types

[nikpivkin]

What is the underlying problem you're trying to solve?

Type checking is very time-consuming with a large number of rules. I noticed that the schema is compiled for each rule , but the result is not cached.

Describe the ideal solution

typeChecker can cache schema types.

Additional Context

We use about 500 Rego files and a large JSON schema. If you don't mind, I can open a PR with a solution to this issue.

I did a comparison of the current version of OPA and with caching:

Before

     11.30 real        19.01 user         0.24 sys
           220233728  maximum resident set size
                   0  average shared memory size
                   0  average unshared data size
                   0  average unshared stack size
               12513  page reclaims
                1147  page faults
                   0  swaps
                   0  block input operations
                   0  block output operations
                   0  messages sent
                   0  messages received
                3164  signals received
                1425  voluntary context switches
               34594  involuntary context switches
        196114350840  instructions retired
         54559370628  cycles elapsed
           158878016  peak memory footprint

After

        0.69 real         0.98 user         0.08 sys
           207503360  maximum resident set size
                   0  average shared memory size
                   0  average unshared data size
                   0  average unshared stack size
               11718  page reclaims
                1163  page faults
                   0  swaps
                   0  block input operations
                   0  block output operations
                   0  messages sent
                   0  messages received
                 217  signals received
                  92  voluntary context switches
                2432  involuntary context switches
         10168081937  instructions retired
          3013213385  cycles elapsed
           146622784  peak memory footprint

[anderseknert]

Interesting! I haven't experienced this myself in Regal, where we pass the input schema when running tests. But that could be that it's simply a smaller schema, us using the same schema for everything using the subpackages scope... or simply that we only have ~200 .rego files compared to your 500.

If you don't mind, I can open a PR with a solution to this issue.

Of course! Thanks 😃

[nikpivkin]

The size of our scheme is about 200kb in compressed form.

[anderseknert]

Ooof, ours is 8kb uncompressed, so yes, that'll explain it.