Documentation ¶
Overview ¶
Package topdown provides low-level query evaluation support.
The topdown implementation is a modified version of the standard top-down evaluation algorithm used in Datalog. References and comprehensions are evaluated eagerly while all other terms are evaluated lazily.
Index ¶
- Constants
- func IsCancel(err error) bool
- func IsError(err error) bool
- func PrettyTrace(w io.Writer, trace []*Event)
- func RegisterBuiltinFunc(name string, f BuiltinFunc)
- func RegisterFunctionalBuiltin1(name string, fun FunctionalBuiltin1)
- func RegisterFunctionalBuiltin2(name string, fun FunctionalBuiltin2)
- func RegisterFunctionalBuiltin3(name string, fun FunctionalBuiltin3)
- func RegisterFunctionalBuiltinVoid1(name string, fun FunctionalBuiltinVoid1)
- type BufferTracer
- type BuiltinContext
- type BuiltinEmpty
- type BuiltinFunc
- type Cancel
- type Error
- type Event
- type FunctionalBuiltin1
- type FunctionalBuiltin2
- type FunctionalBuiltin3
- type FunctionalBuiltinVoid1
- type Op
- type Query
- func (q *Query) Iter(ctx context.Context, iter func(QueryResult) error) error
- func (q *Query) PartialRun(ctx context.Context) (partials []ast.Body, support []*ast.Module, err error)
- func (q *Query) Run(ctx context.Context) (QueryResultSet, error)
- func (q *Query) WithCancel(cancel Cancel) *Query
- func (q *Query) WithCompiler(compiler *ast.Compiler) *Query
- func (q *Query) WithInput(input *ast.Term) *Query
- func (q *Query) WithMetrics(metrics metrics.Metrics) *Query
- func (q *Query) WithPartialNamespace(ns string) *Query
- func (q *Query) WithStore(store storage.Store) *Query
- func (q *Query) WithTracer(tracer Tracer) *Query
- func (q *Query) WithTransaction(txn storage.Transaction) *Query
- func (q *Query) WithUnknowns(terms []*ast.Term) *Query
- type QueryResult
- type QueryResultSet
- type Tracer
Examples ¶
Constants ¶
const ( // InternalErr represents an unknown evaluation error. InternalErr string = "eval_internal_error" // CancelErr indicates the evaluation process was cancelled. CancelErr string = "eval_cancel_error" // ConflictErr indicates a conflict was encountered during evaluation. For // instance, a conflict occurs if a rule produces multiple, differing values // for the same key in an object. Conflict errors indicate the policy does // not account for the data loaded into the policy engine. ConflictErr string = "eval_conflict_error" // TypeErr indicates evaluation stopped because an expression was applied to // a value of an inappropriate type. TypeErr string = "eval_type_error" )
Variables ¶
This section is empty.
Functions ¶
func PrettyTrace ¶ added in v0.2.0
PrettyTrace pretty prints the trace to the writer.
func RegisterBuiltinFunc ¶
func RegisterBuiltinFunc(name string, f BuiltinFunc)
RegisterBuiltinFunc adds a new built-in function to the evaluation engine.
func RegisterFunctionalBuiltin1 ¶ added in v0.4.0
func RegisterFunctionalBuiltin1(name string, fun FunctionalBuiltin1)
RegisterFunctionalBuiltin1 adds a new built-in function to the evaluation engine.
Example ¶
package main import ( "context" "fmt" "strings" "github.com/open-policy-agent/opa/ast" "github.com/open-policy-agent/opa/topdown" "github.com/open-policy-agent/opa/topdown/builtins" "github.com/open-policy-agent/opa/types" ) func main() { // Rego includes a number of built-in functions ("built-ins") for performing // standard operations like string manipulation, regular expression // matching, and computing aggregates. // // This test shows how to add a new built-in to Rego and OPA. // Initialize context for the example. Normally the caller would obtain the // context from an input parameter or instantiate their own. ctx := context.Background() // The ast package contains a registry that enumerates the built-ins // included in Rego. When adding a new built-in, you must update the // registry to include your built-in. Otherwise, the compiler will complain // when it encounters your built-in. builtin := &ast.Builtin{ Name: "mybuiltins.upper", Decl: types.NewFunction( types.Args(types.S), types.S, ), } ast.RegisterBuiltin(builtin) // This is the implementation of the built-in that will be called during // query evaluation. builtinImpl := func(a ast.Value) (ast.Value, error) { str, err := builtins.StringOperand(a, 1) if err != nil { return nil, err } if str.Equal(ast.String("magic")) { // topdown.BuiltinEmpty indicates to the evaluation engine that the // expression is false/not defined. return nil, topdown.BuiltinEmpty{} } return ast.String(strings.ToUpper(string(str))), nil } // See documentation for registering functions that take different numbers // of arguments. topdown.RegisterFunctionalBuiltin1(builtin.Name, builtinImpl) // At this point, the new built-in has been registered and can be used in // queries. Our custom built-in converts strings to upper case but is not // defined for the input "magic". compiler := ast.NewCompiler() query, err := compiler.QueryCompiler().Compile(ast.MustParseBody(`mybuiltins.upper("custom", x); not mybuiltins.upper("magic", "MAGIC")`)) if err != nil { // Handle error. } // Evaluate the query. q := topdown.NewQuery(query).WithCompiler(compiler) q.Iter(ctx, func(qr topdown.QueryResult) error { fmt.Println("x:", qr[ast.Var("x")]) return nil }) // If you add a new built-in function to OPA, you should: // // 1. Update the Language Reference: http://www.openpolicyagent.org/docs/language-reference.html. // 2. Add an integration test to the topdown package. }
Output: x: "CUSTOM"
func RegisterFunctionalBuiltin2 ¶ added in v0.4.0
func RegisterFunctionalBuiltin2(name string, fun FunctionalBuiltin2)
RegisterFunctionalBuiltin2 adds a new built-in function to the evaluation engine.
func RegisterFunctionalBuiltin3 ¶ added in v0.4.0
func RegisterFunctionalBuiltin3(name string, fun FunctionalBuiltin3)
RegisterFunctionalBuiltin3 adds a new built-in function to the evaluation engine.
func RegisterFunctionalBuiltinVoid1 ¶ added in v0.5.0
func RegisterFunctionalBuiltinVoid1(name string, fun FunctionalBuiltinVoid1)
RegisterFunctionalBuiltinVoid1 adds a new built-in function to the evaluation engine.
Types ¶
type BufferTracer ¶ added in v0.2.0
type BufferTracer []*Event
BufferTracer implements the Tracer interface by simply buffering all events received.
func NewBufferTracer ¶ added in v0.2.0
func NewBufferTracer() *BufferTracer
NewBufferTracer returns a new BufferTracer.
func (*BufferTracer) Enabled ¶ added in v0.2.0
func (b *BufferTracer) Enabled() bool
Enabled always returns true if the BufferTracer is instantiated.
func (*BufferTracer) Trace ¶ added in v0.2.0
func (b *BufferTracer) Trace(evt *Event)
Trace adds the event to the buffer.
type BuiltinContext ¶ added in v0.5.11
BuiltinContext contains context from the evaluator that may be used by built-in functions.
type BuiltinEmpty ¶ added in v0.4.0
type BuiltinEmpty struct{}
BuiltinEmpty is used to signal that the built-in function evaluated, but the result is undefined so evaluation should not continue.
func (BuiltinEmpty) Error ¶ added in v0.4.0
func (BuiltinEmpty) Error() string
type BuiltinFunc ¶
BuiltinFunc defines a generic interface for built-in functions.
type Cancel ¶ added in v0.5.1
type Cancel interface { Cancel() Cancelled() bool }
Cancel defines the interface for cancelling topdown queries. Cancel operations are thread-safe and idempotent.
type Error ¶
type Error struct { Code string `json:"code"` Message string `json:"message"` Location *ast.Location `json:"location,omitempty"` }
Error is the error type returned by the Eval and Query functions when an evaluation error occurs.
type Event ¶ added in v0.2.0
type Event struct { Op Op // Identifies type of event. Node interface{} // Contains AST node relevant to the event. QueryID uint64 // Identifies the query this event belongs to. ParentID uint64 // Identifies the parent query this event belongs to. Locals *ast.ValueMap // Contains local variable bindings from the query context. }
Event contains state associated with a tracing event.
type FunctionalBuiltin1 ¶ added in v0.4.0
FunctionalBuiltin1 defines an interface for simple functional built-ins.
Implement this interface if your built-in function takes one input and produces one output.
If an error occurs, the functional built-in should return a descriptive message. The message should not be prefixed with the built-in name as the framework takes care of this.
type FunctionalBuiltin2 ¶ added in v0.4.0
FunctionalBuiltin2 defines an interface for simple functional built-ins.
Implement this interface if your built-in function takes two inputs and produces one output.
If an error occurs, the functional built-in should return a descriptive message. The message should not be prefixed with the built-in name as the framework takes care of this.
type FunctionalBuiltin3 ¶ added in v0.4.0
FunctionalBuiltin3 defines an interface for simple functional built-ins.
Implement this interface if your built-in function takes three inputs and produces one output.
If an error occurs, the functional built-in should return a descriptive message. The message should not be prefixed with the built-in name as the framework takes care of this.
type FunctionalBuiltinVoid1 ¶ added in v0.5.0
FunctionalBuiltinVoid1 defines an interface for simple functional built-ins.
Implement this interface if your built-in function takes one input and produces no outputs.
If an error occurs, the functional built-in should return a descriptive message. The message should not be prefixed with the built-in name as the framework takes care of this.
type Op ¶ added in v0.2.0
type Op string
Op defines the types of tracing events.
const ( // EnterOp is emitted when a new query is about to be evaluated. EnterOp Op = "Enter" // ExitOp is emitted when a query has evaluated to true. ExitOp Op = "Exit" // EvalOp is emitted when an expression is about to be evaluated. EvalOp Op = "Eval" // RedoOp is emitted when an expression, rule, or query is being re-evaluated. RedoOp Op = "Redo" // SaveOp is emitted when an expression is saved instead of evaluated // during partial evaluation. SaveOp Op = "Save" // FailOp is emitted when an expression evaluates to false. FailOp Op = "Fail" )
type Query ¶
type Query struct {
// contains filtered or unexported fields
}
Query provides a configurable interface for performing query evaluation.
func (*Query) Iter ¶ added in v0.5.11
Iter executes the query and invokes the iter function with query results produced by evaluating the query.
Example ¶
package main import ( "bytes" "context" "encoding/json" "fmt" "github.com/open-policy-agent/opa/ast" "github.com/open-policy-agent/opa/storage/inmem" "github.com/open-policy-agent/opa/topdown" ) func main() { // Initialize context for the example. Normally the caller would obtain the // context from an input parameter or instantiate their own. ctx := context.Background() compiler := ast.NewCompiler() // Define a dummy query and some data that the query will execute against. query, err := compiler.QueryCompiler().Compile(ast.MustParseBody(`data.a[_] = x; x >= 2`)) if err != nil { // Handle error. } var data map[string]interface{} // OPA uses Go's standard JSON library but assumes that numbers have been // decoded as json.Number instead of float64. You MUST decode with UseNumber // enabled. decoder := json.NewDecoder(bytes.NewBufferString(`{"a": [1,2,3,4]}`)) decoder.UseNumber() if err := decoder.Decode(&data); err != nil { // Handle error. } // Instantiate the policy engine's storage layer. store := inmem.NewFromObject(data) // Create a new transaction. Transactions allow the policy engine to // evaluate the query over a consistent snapshot fo the storage layer. txn, err := store.NewTransaction(ctx) if err != nil { // Handle error. } defer store.Abort(ctx, txn) // Prepare the evaluation parameters. Evaluation executes against the policy // engine's storage. In this case, we seed the storage with a single array // of number. Other parameters such as the input, tracing configuration, // etc. can be set on the query object. q := topdown.NewQuery(query). WithCompiler(compiler). WithStore(store). WithTransaction(txn) result := []interface{}{} // Execute the query and provide a callback function to accumulate the results. err = q.Iter(ctx, func(qr topdown.QueryResult) error { // Each variable in the query will have an associated binding. x := qr[ast.Var("x")] // The bindings are ast.Value types so we will convert to a native Go value here. v, err := ast.JSON(x.Value) if err != nil { return err } result = append(result, v) return nil }) // Inspect the query result. fmt.Println("result:", result) fmt.Println("err:", err) }
Output: result: [2 3 4] err: <nil>
func (*Query) PartialRun ¶ added in v0.5.11
func (q *Query) PartialRun(ctx context.Context) (partials []ast.Body, support []*ast.Module, err error)
PartialRun executes partial evaluation on the query with respect to unknown values. Partial evaluation attempts to evaluate as much of the query as possible without requiring values for the unknowns set on the query. The result of partial evaluation is a new set of queries that can be evaluated once the unknown value is known. In addition to new queries, partial evaluation may produce additional support modules that should be used in conjunction with the partially evaluated queries.
Example ¶
package main import ( "bytes" "context" "encoding/json" "fmt" "github.com/open-policy-agent/opa/ast" "github.com/open-policy-agent/opa/storage/inmem" "github.com/open-policy-agent/opa/topdown" ) func main() { // Initialize context for the example. Normally the caller would obtain the // context from an input parameter or instantiate their own. ctx := context.Background() var data map[string]interface{} decoder := json.NewDecoder(bytes.NewBufferString(`{ "roles": [ { "permissions": ["read_bucket"], "groups": ["dev", "test", "sre"] }, { "permissions": ["write_bucket", "delete_bucket"], "groups": ["sre"] } ] }`)) if err := decoder.Decode(&data); err != nil { // Handle error. } // Instantiate the policy engine's storage layer. store := inmem.NewFromObject(data) // Create a new transaction. Transactions allow the policy engine to // evaluate the query over a consistent snapshot fo the storage layer. txn, err := store.NewTransaction(ctx) if err != nil { // Handle error. } defer store.Abort(ctx, txn) // Define policy that searches for roles that match input request. If no // roles are found, allow is undefined and the caller will reject the // request. This is the user supplied policy that OPA will partially // evaluate. modules := map[string]*ast.Module{ "authz.rego": ast.MustParseModule(` package example default allow = false allow { role = data.roles[i] input.group = role.groups[j] input.permission = role.permissions[k] } `), } // Compile policy. compiler := ast.NewCompiler() if compiler.Compile(modules); compiler.Failed() { // Handle error. } // Construct query and mark the entire input document as partial. q := topdown.NewQuery(ast.MustParseBody("data.example.allow = true")). WithCompiler(compiler). WithUnknowns([]*ast.Term{ ast.InputRootDocument, }). WithStore(store). WithTransaction(txn) // Execute partial evaluation. partial, support, err := q.PartialRun(ctx) if err != nil { // Handle error. } // Show result of partially evaluating the policy. fmt.Println("# partial evaluation result:") for i := range partial { fmt.Println(partial[i]) } fmt.Println() fmt.Println("# partial evaluation support:") for i := range support { fmt.Println(support[i]) } // Construct a new policy to contain the result of partial evaluation. module := ast.MustParseModule("package partial") for i := range partial { rule := &ast.Rule{ Head: &ast.Head{ Name: ast.Var("allow"), Value: ast.BooleanTerm(true), }, Body: partial[i], Module: module, } module.Rules = append(module.Rules, rule) } // Compile the partially evaluated policy with the original policy. modules["partial"] = module for i, v := range support { modules[fmt.Sprintf("partial_support_%d", i)] = v } if compiler.Compile(modules); compiler.Failed() { // Handle error. } // Test different inputs against partially evaluated policy. inputs := []string{ `{"group": "dev", "permission": "read_bucket"}`, // allow `{"group": "dev", "permission": "write_bucket"}`, // deny `{"group": "sre", "permission": "write_bucket"}`, // allow } fmt.Println() fmt.Println("# evaluation results:") for i := range inputs { // Query partially evaluated policy. q = topdown.NewQuery(ast.MustParseBody("data.partial.allow = true")). WithCompiler(compiler). WithStore(store). WithTransaction(txn). WithInput(ast.MustParseTerm(inputs[i])) qrs, err := q.Run(ctx) if err != nil { // Handle error. } // Check if input is allowed. allowed := len(qrs) == 1 fmt.Printf("input %d allowed: %v\n", i+1, allowed) } }
Output: # partial evaluation result: data.partial.example.allow = true # partial evaluation support: package partial.example allow = true { "dev" = input.group; "read_bucket" = input.permission } allow = true { "test" = input.group; "read_bucket" = input.permission } allow = true { "sre" = input.group; "read_bucket" = input.permission } allow = true { "sre" = input.group; "write_bucket" = input.permission } allow = true { "sre" = input.group; "delete_bucket" = input.permission } default allow = false # evaluation results: input 1 allowed: true input 2 allowed: false input 3 allowed: true
func (*Query) Run ¶ added in v0.5.11
func (q *Query) Run(ctx context.Context) (QueryResultSet, error)
Run is a wrapper around Iter that accumulates query results and returns them in one shot.
Example ¶
package main import ( "bytes" "context" "encoding/json" "fmt" "github.com/open-policy-agent/opa/ast" "github.com/open-policy-agent/opa/storage/inmem" "github.com/open-policy-agent/opa/topdown" ) func main() { // Initialize context for the example. Normally the caller would obtain the // context from an input parameter or instantiate their own. ctx := context.Background() compiler := ast.NewCompiler() // Define a dummy query and some data that the query will execute against. query, err := compiler.QueryCompiler().Compile(ast.MustParseBody(`data.a[_] = x; x >= 2`)) if err != nil { // Handle error. } var data map[string]interface{} // OPA uses Go's standard JSON library but assumes that numbers have been // decoded as json.Number instead of float64. You MUST decode with UseNumber // enabled. decoder := json.NewDecoder(bytes.NewBufferString(`{"a": [1,2,3,4]}`)) decoder.UseNumber() if err := decoder.Decode(&data); err != nil { // Handle error. } // Instantiate the policy engine's storage layer. store := inmem.NewFromObject(data) // Create a new transaction. Transactions allow the policy engine to // evaluate the query over a consistent snapshot fo the storage layer. txn, err := store.NewTransaction(ctx) if err != nil { // Handle error. } defer store.Abort(ctx, txn) // Prepare the evaluation parameters. Evaluation executes against the policy // engine's storage. In this case, we seed the storage with a single array // of number. Other parameters such as the input, tracing configuration, // etc. can be set on the query object. q := topdown.NewQuery(query). WithCompiler(compiler). WithStore(store). WithTransaction(txn) rs, err := q.Run(ctx) // Inspect the query result set. fmt.Println("len:", len(rs)) for i := range rs { fmt.Printf("rs[%d][\"x\"]: %v\n", i, rs[i]["x"]) } fmt.Println("err:", err) }
Output: len: 3 rs[0]["x"]: 2 rs[1]["x"]: 3 rs[2]["x"]: 4 err: <nil>
func (*Query) WithCancel ¶ added in v0.5.11
WithCancel sets the cancellation object to use for the query. Set this if you need to abort queries based on a deadline. This is optional.
func (*Query) WithCompiler ¶ added in v0.5.11
WithCompiler sets the compiler to use for the query.
func (*Query) WithInput ¶ added in v0.5.11
WithInput sets the input object to use for the query. References rooted at input will be evaluated against this value. This is optional.
func (*Query) WithMetrics ¶ added in v0.5.11
WithMetrics sets the metrics collection to add evaluation metrics to. This is optional.
func (*Query) WithPartialNamespace ¶ added in v0.6.0
WithPartialNamespace sets the namespace to use for supporting rules generated as part of the partial evaluation process. The ns value must be a valid package path component.
func (*Query) WithTracer ¶ added in v0.5.11
WithTracer sets the query tracer to use during evaluation. This is optional.
func (*Query) WithTransaction ¶ added in v0.5.11
func (q *Query) WithTransaction(txn storage.Transaction) *Query
WithTransaction sets the transaction to use for the query. All queries should be performed over a consistent snapshot of the storage layer.
type QueryResult ¶ added in v0.2.1
QueryResult represents a single result returned by a query. The result contains bindings for all variables that appear in the query.
type QueryResultSet ¶ added in v0.2.1
type QueryResultSet []QueryResult
QueryResultSet represents a collection of results returned by a query.