mirror of
https://github.com/open-policy-agent/opa.git
synced 2026-08-12 19:32:48 -06:00
topdown: eliminate closure allocations in Set and virtual doc enumeration (#8242)
Replace closure allocations in evalTree.enumerate with method values for Set iteration and virtual document traversal. Set enumeration now uses Slice() instead of Iter(callback), and virtual doc enumeration uses enumerateNext helper instead of inline closures. Add BenchmarkEnumerateComprehensions to measure memory impact of closure optimizations in evalTree.enumerate with set/array comprehensions over large datasets. Signed-off-by: alex60217101990 <alex6021710@gmail.com>
This commit is contained in:
alex60217101990
@@ -0,0 +1,236 @@
|
||||
// Copyright 2026 The OPA Authors. All rights reserved.
|
||||
// Use of this source code is governed by an Apache2
|
||||
// license that can be found in the LICENSE file.
|
||||
|
||||
package topdown
|
||||
|
||||
import (
|
||||
"context"
|
||||
"fmt"
|
||||
"math/rand"
|
||||
"testing"
|
||||
|
||||
"github.com/open-policy-agent/opa/v1/ast"
|
||||
"github.com/open-policy-agent/opa/v1/storage"
|
||||
inmem "github.com/open-policy-agent/opa/v1/storage/inmem/test"
|
||||
)
|
||||
|
||||
// BenchmarkEnumerateComprehensions benchmarks policy evaluation with
|
||||
// comprehensions over large datasets. This specifically targets the
|
||||
// enumerate optimization that eliminates closure allocations.
|
||||
func BenchmarkEnumerateComprehensions(b *testing.B) {
|
||||
sizes := []int{1000, 5000, 10000}
|
||||
|
||||
for _, size := range sizes {
|
||||
b.Run(fmt.Sprintf("size_%d", size), func(b *testing.B) {
|
||||
ctx := context.Background()
|
||||
|
||||
// Generate mock dataset with nested objects
|
||||
data := generateNestedDataset(size)
|
||||
store := inmem.NewFromObject(data)
|
||||
|
||||
// Policy with multiple comprehensions that exercise enumerate
|
||||
module := `package test
|
||||
|
||||
import rego.v1
|
||||
|
||||
# Set comprehension over users
|
||||
active_users contains user.id if {
|
||||
some user in data.users
|
||||
user.profile.active == true
|
||||
}
|
||||
|
||||
# Array comprehension with nested access
|
||||
premium_users := [user |
|
||||
some user in data.users
|
||||
user.profile.settings.subscription.tier == "premium"
|
||||
]
|
||||
|
||||
# Object comprehension with filtering
|
||||
users_by_age contains age_group if {
|
||||
age_group := "20-30"
|
||||
some u in data.users
|
||||
u.profile.age >= 20
|
||||
u.profile.age < 30
|
||||
}
|
||||
|
||||
users_by_age contains age_group if {
|
||||
age_group := "30-40"
|
||||
some u in data.users
|
||||
u.profile.age >= 30
|
||||
u.profile.age < 40
|
||||
}
|
||||
|
||||
# Nested comprehension
|
||||
high_value_users contains user.email if {
|
||||
some user in data.users
|
||||
user.profile.active == true
|
||||
count([p | some p in user.permissions; p.level > 5]) > 0
|
||||
}
|
||||
|
||||
# Random access pattern
|
||||
user_lookup[id] := user if {
|
||||
some user in data.users
|
||||
id := user.id
|
||||
}
|
||||
`
|
||||
|
||||
compiler := ast.MustCompileModules(map[string]string{
|
||||
"test.rego": module,
|
||||
})
|
||||
|
||||
// Query that exercises all comprehensions
|
||||
query := ast.MustParseBody(`
|
||||
data.test.active_users
|
||||
data.test.premium_users
|
||||
data.test.users_by_age
|
||||
data.test.high_value_users
|
||||
`)
|
||||
|
||||
b.ReportAllocs()
|
||||
b.ResetTimer()
|
||||
|
||||
for b.Loop() {
|
||||
err := storage.Txn(ctx, store, storage.TransactionParams{}, func(txn storage.Transaction) error {
|
||||
q := NewQuery(query).
|
||||
WithCompiler(compiler).
|
||||
WithStore(store).
|
||||
WithTransaction(txn)
|
||||
|
||||
_, err := q.Run(ctx)
|
||||
return err
|
||||
})
|
||||
|
||||
if err != nil {
|
||||
b.Fatal(err)
|
||||
}
|
||||
}
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
// generateNestedDataset creates a dataset with nested objects of varying depth
|
||||
func generateNestedDataset(size int) map[string]any {
|
||||
users := make([]any, size)
|
||||
rng := rand.New(rand.NewSource(42)) // Fixed seed for reproducibility
|
||||
|
||||
tiers := []string{"free", "basic", "premium", "enterprise"}
|
||||
departments := []string{"engineering", "sales", "marketing", "support", "hr"}
|
||||
|
||||
for i := range size {
|
||||
// Random nested object with 3-5 levels of nesting
|
||||
permissions := make([]any, rng.Intn(10)+1)
|
||||
for j := 0; j < len(permissions); j++ {
|
||||
permissions[j] = map[string]any{
|
||||
"name": fmt.Sprintf("perm_%d", j),
|
||||
"level": rng.Intn(10),
|
||||
"scope": map[string]any{
|
||||
"resource": fmt.Sprintf("res_%d", rng.Intn(100)),
|
||||
"actions": []string{"read", "write", "delete"}[rng.Intn(3)],
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
users[i] = map[string]any{
|
||||
"id": fmt.Sprintf("user_%d", i),
|
||||
"name": fmt.Sprintf("User %d", i),
|
||||
"email": fmt.Sprintf("user%d@example.com", i),
|
||||
"profile": map[string]any{
|
||||
"active": rng.Float64() > 0.3, // 70% active
|
||||
"age": 20 + rng.Intn(40), // Age 20-59
|
||||
"settings": map[string]any{
|
||||
"subscription": map[string]any{
|
||||
"tier": tiers[rng.Intn(len(tiers))],
|
||||
"start_date": "2024-01-01",
|
||||
"features": map[string]any{
|
||||
"api_access": rng.Float64() > 0.5,
|
||||
"custom_domain": rng.Float64() > 0.7,
|
||||
"priority_support": map[string]any{
|
||||
"enabled": rng.Float64() > 0.8,
|
||||
"level": rng.Intn(5) + 1,
|
||||
},
|
||||
},
|
||||
},
|
||||
"notifications": map[string]any{
|
||||
"email": rng.Float64() > 0.4,
|
||||
"sms": rng.Float64() > 0.8,
|
||||
},
|
||||
},
|
||||
"department": departments[rng.Intn(len(departments))],
|
||||
},
|
||||
"permissions": permissions,
|
||||
"metadata": map[string]any{
|
||||
"created_at": "2024-01-01T00:00:00Z",
|
||||
"updated_at": "2024-01-15T00:00:00Z",
|
||||
"tags": map[string]any{
|
||||
"region": []string{"us-west", "us-east", "eu-central"}[rng.Intn(3)],
|
||||
"environment": []string{"prod", "staging", "dev"}[rng.Intn(3)],
|
||||
"cost_center": fmt.Sprintf("CC%04d", rng.Intn(1000)),
|
||||
},
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
return map[string]any{
|
||||
"users": users,
|
||||
}
|
||||
}
|
||||
|
||||
// BenchmarkEnumerateRandomAccess benchmarks random access patterns
|
||||
// that exercise virtual document enumeration
|
||||
func BenchmarkEnumerateRandomAccess(b *testing.B) {
|
||||
ctx := context.Background()
|
||||
|
||||
data := generateNestedDataset(10000)
|
||||
store := inmem.NewFromObject(data)
|
||||
|
||||
module := `package test
|
||||
|
||||
import rego.v1
|
||||
|
||||
# Virtual document with random access
|
||||
user_by_id[id] := user if {
|
||||
some user in data.users
|
||||
id := user.id
|
||||
}
|
||||
|
||||
# Nested virtual document access
|
||||
premium_by_dept[dept] := users if {
|
||||
dept := data.users[_].profile.department
|
||||
users := [u |
|
||||
some u in data.users
|
||||
u.profile.department == dept
|
||||
u.profile.settings.subscription.tier == "premium"
|
||||
]
|
||||
}
|
||||
`
|
||||
|
||||
compiler := ast.MustCompileModules(map[string]string{
|
||||
"test.rego": module,
|
||||
})
|
||||
|
||||
// Access random users
|
||||
query := ast.MustParseBody(`
|
||||
data.test.user_by_id["user_1234"]
|
||||
data.test.user_by_id["user_5678"]
|
||||
data.test.premium_by_dept.engineering
|
||||
`)
|
||||
|
||||
b.ReportAllocs()
|
||||
|
||||
for b.Loop() {
|
||||
err := storage.Txn(ctx, store, storage.TransactionParams{}, func(txn storage.Transaction) error {
|
||||
q := NewQuery(query).
|
||||
WithCompiler(compiler).
|
||||
WithStore(store).
|
||||
WithTransaction(txn)
|
||||
|
||||
_, err := q.Run(ctx)
|
||||
return err
|
||||
})
|
||||
|
||||
if err != nil {
|
||||
b.Fatal(err)
|
||||
}
|
||||
}
|
||||
}
|
||||
+32
-16
@@ -2486,6 +2486,20 @@ func (e evalTree) next(iter unifyIterator, plugged *ast.Term) error {
|
||||
return cpy.eval(iter)
|
||||
}
|
||||
|
||||
// enumerateNext is a helper to avoid closure allocation in enumerate loops.
|
||||
// Method values don't allocate, unlike explicit closures.
|
||||
// Using a pointer to evalTree avoids copying the 96-byte structure.
|
||||
// Fields are ordered by size for optimal memory alignment (16 > 8 > 8 bytes).
|
||||
type enumerateNext struct {
|
||||
iter unifyIterator // 16 bytes (interface)
|
||||
e *evalTree // 8 bytes (pointer)
|
||||
key *ast.Term // 8 bytes (pointer)
|
||||
}
|
||||
|
||||
func (en *enumerateNext) call() error {
|
||||
return en.e.next(en.iter, en.key)
|
||||
}
|
||||
|
||||
func (e evalTree) enumerate(iter unifyIterator) error {
|
||||
|
||||
if e.e.inliningControl.Disabled(e.plugged[:e.pos], true) {
|
||||
@@ -2501,14 +2515,17 @@ func (e evalTree) enumerate(iter unifyIterator) error {
|
||||
dc.deferred = nil
|
||||
defer deecPool.Put(dc)
|
||||
|
||||
// Use method value to avoid closure allocation.
|
||||
// Create once and reuse for both doc and virtual doc enumeration.
|
||||
en := enumerateNext{iter: iter, e: &e, key: nil}
|
||||
|
||||
if doc != nil {
|
||||
switch doc := doc.(type) {
|
||||
case *ast.Array:
|
||||
for i := range doc.Len() {
|
||||
k := ast.InternedTerm(i)
|
||||
err := e.e.biunify(k, e.ref[e.pos], e.bindings, e.bindings, func() error {
|
||||
return e.next(iter, k)
|
||||
})
|
||||
en.key = k
|
||||
err := e.e.biunify(k, e.ref[e.pos], e.bindings, e.bindings, en.call)
|
||||
|
||||
if err := dc.handleErr(err); err != nil {
|
||||
return err
|
||||
@@ -2517,21 +2534,20 @@ func (e evalTree) enumerate(iter unifyIterator) error {
|
||||
case ast.Object:
|
||||
ki := doc.KeysIterator()
|
||||
for k, more := ki.Next(); more; k, more = ki.Next() {
|
||||
err := e.e.biunify(k, e.ref[e.pos], e.bindings, e.bindings, func() error {
|
||||
return e.next(iter, k)
|
||||
})
|
||||
en.key = k
|
||||
err := e.e.biunify(k, e.ref[e.pos], e.bindings, e.bindings, en.call)
|
||||
if err := dc.handleErr(err); err != nil {
|
||||
return err
|
||||
}
|
||||
}
|
||||
case ast.Set:
|
||||
if err := doc.Iter(func(elem *ast.Term) error {
|
||||
err := e.e.biunify(elem, e.ref[e.pos], e.bindings, e.bindings, func() error {
|
||||
return e.next(iter, elem)
|
||||
})
|
||||
return dc.handleErr(err)
|
||||
}); err != nil {
|
||||
return err
|
||||
// Use Slice() to avoid closure allocation in Iter()
|
||||
for _, elem := range doc.Slice() {
|
||||
en.key = elem
|
||||
err := e.e.biunify(elem, e.ref[e.pos], e.bindings, e.bindings, en.call)
|
||||
if err := dc.handleErr(err); err != nil {
|
||||
return err
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -2544,11 +2560,11 @@ func (e evalTree) enumerate(iter unifyIterator) error {
|
||||
return nil
|
||||
}
|
||||
|
||||
// Reuse the same enumerateNext for virtual documents
|
||||
for _, k := range e.node.Sorted {
|
||||
key := ast.NewTerm(k)
|
||||
if err := e.e.biunify(key, e.ref[e.pos], e.bindings, e.bindings, func() error {
|
||||
return e.next(iter, key)
|
||||
}); err != nil {
|
||||
en.key = key
|
||||
if err := e.e.biunify(key, e.ref[e.pos], e.bindings, e.bindings, en.call); err != nil {
|
||||
return err
|
||||
}
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user