Excessive time spent in ast.(*parser).parseExpr

[open-policy-agent]

Excessive time spent in ast.(*parser).parseExpr

[patrick-east]

A quick look at where the OPA eval happens (I think? Correct me if I'm wrong on this): https://github.com/forseti-security/config-validator/blob/258b9eb9e8e6d6e91da8edbcab771d19ece3fad8/pkg/gcv/cf/cf.go#L214-L222 it appears that the query is passed in as a string for each evaluation. If the actual eval is fast it wouldn't be super surprising to see that a large chunk of time is spent parsing the query on each of the benchmark loop iterations. I see the topdown.eval entries only taking up a few % on that screenshot so it seems likely that this is the case, especially if the iteration count was large for the benchmark.

More recently in OPA we have added a set of new API's on the rego package that prepares the rego.Rego object as a "prepared query" and can then be re-used for evaluations that eliminate the need to do the query prep (and a number of other things, depending on the options passed in). You can then use it with different inputs for evaluations and save a bunch of time, especially on repeated queries where only input is changing (which it looks like in this case here). Check out https://pkg.go.dev/github.com/open-policy-agent/[email protected]/rego?tab=doc#example-Rego.PrepareForEval

Again, just from a quick look. I might have totally missed some other spot that is doing additional parsing.

[patrick-east]

@maxsmythe, as discussed, to parse the query ahead of time you'd use https://godoc.org/github.com/open-policy-agent/opa/ast#ParseBody and then use this option to pass in the parsed query https://godoc.org/github.com/open-policy-agent/opa/rego#ParsedQuery

[willbeason]

This has been resolved with #202 and #206 . We only parse Template rego once, and we don't recompile already existing Templates.