[TT-9864] Optimize the creation usage of AST documents

[buraksezer]

PR for TT-9864.

After reading a lot of code, trying different solutions and measuring them, I decided to go with the simplest solution.

I used sync.Pool for reusing ast.Document instances and added a Cleanup method to Request struct to recycle ast.Document structs if it's possible.

The library caches the plans for queries but before caching or deciding to use a plan from the cache, it creates a new ast.Document, parses the query, normalizes and validates it against the schema. If a plan is cached, resetting and putting ast.Document to the pool can lead to subtle bugs because the cached plan still uses it for the incoming queries.

I added a boolean field to Request struct to mark points out the request will be processed by a plan from the cache. If so, we don't need the ast.Document in the Request struct anymore and we can recycle it. Cleanup method does this job. I plan to use Cleanup method from Tyk GW. I tried to run the CI pipeline with a dummy PR and it worked. See this: https://github.com/TykTechnologies/tyk/pull/6345/files#

Because Cleanup method is only called from Tyk GW, it should not has any side effects on the tests in the library.

ast.NewDocument method has been called from different places in the library but recycling ast.Document in GraphQL data source and federation code leads to bugs and breaks the tests.

[sonarcloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
No data about Coverage
0.0% Duplication on New Code

See analysis details on SonarCloud

[buraksezer]

Latest benchmark results can be found here:

# Pooling ast.Document instances

  graphql git:(feat/TT-9864/Optimize-the-creation-usage-of-AST-documents) ✗ go test -bench=BenchmarkExecutionEngineV2_Execute_AstDocumentPool -run='^$' -v -memprofile memprofile.out -count=10 .
Alloc = 1 MiB	TotalAlloc = 1 MiB	Sys = 9 MiB	NumGC = 0
goos: darwin
goarch: arm64
pkg: github.com/TykTechnologies/graphql-go-tools/pkg/graphql
BenchmarkExecutionEngineV2_Execute_AstDocumentPool
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   79106	     13289 ns/op	   38904 B/op	     118 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   91957	     13402 ns/op	   38852 B/op	     118 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   93105	     12736 ns/op	   38819 B/op	     118 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   93937	     12798 ns/op	   38781 B/op	     118 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   87145	     13045 ns/op	   38806 B/op	     118 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   91832	     12914 ns/op	   38773 B/op	     118 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   94764	     12832 ns/op	   38797 B/op	     118 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   84831	     13630 ns/op	   38942 B/op	     118 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   95127	     12783 ns/op	   38821 B/op	     118 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   94060	     13051 ns/op	   38846 B/op	     118 allocs/op
PASS
Alloc = 2 MiB	TotalAlloc = 37317 MiB	Sys = 19 MiB	NumGC = 18325
ok  	github.com/TykTechnologies/graphql-go-tools/pkg/graphql	13.616s

# Without pooling
➜  graphql git:(master) ✗ go test -bench=BenchmarkExecutionEngineV2_Execute_AstDocumentPool -run='^$' -v -memprofile memprofile.out -count=10 .
Alloc = 1 MiB	TotalAlloc = 1 MiB	Sys = 9 MiB	NumGC = 0
goos: darwin
goarch: arm64
pkg: github.com/TykTechnologies/graphql-go-tools/pkg/graphql
BenchmarkExecutionEngineV2_Execute_AstDocumentPool
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   31033	     38686 ns/op	  176441 B/op	     165 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   31431	     38183 ns/op	  176437 B/op	     165 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   31693	     38475 ns/op	  176469 B/op	     165 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   31638	     38381 ns/op	  176481 B/op	     165 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   31202	     38643 ns/op	  176496 B/op	     165 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   31650	     37853 ns/op	  176421 B/op	     165 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   30669	     38295 ns/op	  176468 B/op	     165 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   32060	     39800 ns/op	  176517 B/op	     165 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   30771	     38303 ns/op	  176439 B/op	     165 allocs/op
BenchmarkExecutionEngineV2_Execute_AstDocumentPool-8   	   31297	     38701 ns/op	  176467 B/op	     165 allocs/op
PASS
Alloc = 2 MiB	TotalAlloc = 69767 MiB	Sys = 51 MiB	NumGC = 36602
ok  	github.com/TykTechnologies/graphql-go-tools/pkg/graphql	16.436s

# benchstat output

➜  ~ benchstat old.txt new.txt
goos: darwin
goarch: arm64
pkg: github.com/TykTechnologies/graphql-go-tools/pkg/graphql
                                            │   old.txt   │               new.txt               │
                                            │   sec/op    │   sec/op     vs base                │
ExecutionEngineV2_Execute_AstDocumentPool-8   38.43µ ± 1%   12.98µ ± 3%  -66.22% (p=0.000 n=10)

                                            │    old.txt    │               new.txt                │
                                            │     B/op      │     B/op      vs base                │
ExecutionEngineV2_Execute_AstDocumentPool-8   172.33Ki ± 0%   37.91Ki ± 0%  -78.00% (p=0.000 n=10)

                                            │  old.txt   │              new.txt               │
                                            │ allocs/op  │ allocs/op   vs base                │
ExecutionEngineV2_Execute_AstDocumentPool-8   165.0 ± 0%   118.0 ± 0%  -28.48% (p=0.000 n=10)

[kofoworola]

Seen the benchmark result, nice!