Add per-execution memory budgets to RVM

CHANGELOG.md

@@ -6,6 +6,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+
+- *(rvm)* add opt-in per-execution memory budgets for run-to-completion evaluation, including typed Rust and binding errors ([#792](https://github.com/microsoft/regorus/pull/792))
+
 ## [0.11.0](https://github.com/microsoft/regorus/compare/regorus-v0.10.1...regorus-v0.11.0) - 2026-07-21
 
 ### Added
@@ -634,4 +638,3 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - LICENSE committed
 - CODE_OF_CONDUCT.md committed
 - Initial commit
-

benches/rvm_benchmark.rs

@@ -38,6 +38,8 @@
 
 use std::hint::black_box;
 use std::num::NonZeroU32;
+#[cfg(all(feature = "allocator-memory-limits", not(miri)))]
+use std::num::NonZeroU64;
 use std::path::Path;
 use std::sync::Arc;
 use std::time::Duration;
@@ -50,14 +52,16 @@ use regorus::languages::rego::compiler::Compiler;
 use regorus::rvm::program::Program;
 use regorus::rvm::vm::{ExecutionMode, RegoVM};
 use regorus::utils::limits::ExecutionTimerConfig;
+#[cfg(all(feature = "allocator-memory-limits", not(miri)))]
+use regorus::MemoryBudgetConfig;
 use regorus::{Engine, Rc, Value};
 
 // ---------------------------------------------------------------------------
 // Limit constants – generous ceilings that still exercise the limit-checking
 // hot path (memory_check, execution_timer_tick, instruction-limit compare).
 // ---------------------------------------------------------------------------
 
-#[cfg(feature = "allocator-memory-limits")]
+#[cfg(all(feature = "allocator-memory-limits", not(miri)))]
 const MEMORY_LIMIT_BYTES: u64 = 256 * 1024 * 1024;
 const TIME_LIMIT: Duration = Duration::from_secs(30);
 const TIMER_CHECK_INTERVAL: NonZeroU32 = NonZeroU32::new(16).unwrap();
@@ -68,28 +72,39 @@ struct EvalConfig {
     name: &'static str,
     mode: ExecutionMode,
     limits: bool,
+    memory_budget: bool,
 }
 
-const EVAL_CONFIGS: [EvalConfig; 4] = [
+const EVAL_CONFIGS: [EvalConfig; 5] = [
     EvalConfig {
         name: "regular_no_limits",
         mode: ExecutionMode::RunToCompletion,
         limits: false,
+        memory_budget: false,
+    },
+    EvalConfig {
+        name: "regular_memory_budget",
+        mode: ExecutionMode::RunToCompletion,
+        limits: false,
+        memory_budget: true,
     },
     EvalConfig {
         name: "regular_with_limits",
         mode: ExecutionMode::RunToCompletion,
         limits: true,
+        memory_budget: true,
     },
     EvalConfig {
         name: "suspendable_no_limits",
         mode: ExecutionMode::Suspendable,
         limits: false,
+        memory_budget: false,
     },
     EvalConfig {
         name: "suspendable_with_limits",
         mode: ExecutionMode::Suspendable,
         limits: true,
+        memory_budget: false,
     },
 ];
 
@@ -358,22 +373,30 @@ fn compile_all_programs() -> Vec<BenchmarkProgram> {
 // Limit helpers
 // ---------------------------------------------------------------------------
 
-/// Apply or remove production-style limits based on a boolean flag.
-fn configure_limits(vm: &mut RegoVM, limits: bool) {
-    if limits {
-        #[cfg(feature = "allocator-memory-limits")]
+/// Apply the limits selected for one benchmark configuration.
+fn configure_limits(vm: &mut RegoVM, config: EvalConfig) {
+    #[cfg(any(miri, not(feature = "allocator-memory-limits")))]
+    let _ = config.memory_budget;
+
+    if config.limits {
+        #[cfg(all(feature = "allocator-memory-limits", not(miri)))]
         regorus::set_global_memory_limit(Some(MEMORY_LIMIT_BYTES));
         vm.set_execution_timer_config(Some(ExecutionTimerConfig {
             limit: TIME_LIMIT,
             check_interval: TIMER_CHECK_INTERVAL,
         }));
         vm.set_max_instructions(INSTRUCTION_LIMIT);
     } else {
-        #[cfg(feature = "allocator-memory-limits")]
+        #[cfg(all(feature = "allocator-memory-limits", not(miri)))]
         regorus::set_global_memory_limit(None);
         vm.set_execution_timer_config(None);
         vm.set_max_instructions(usize::MAX);
     }
+
+    #[cfg(all(feature = "allocator-memory-limits", not(miri)))]
+    vm.set_memory_budget_config(config.memory_budget.then(|| MemoryBudgetConfig {
+        limit: NonZeroU64::new(MEMORY_LIMIT_BYTES).expect("non-zero memory budget"),
+    }));
 }
 
 // ---------------------------------------------------------------------------
@@ -408,7 +431,7 @@ fn bench_cold(c: &mut Criterion) {
                             vm.set_data(black_box(d.clone())).unwrap();
                         }
                         vm.set_input(black_box(input.clone()));
-                        configure_limits(&mut vm, config.limits);
+                        configure_limits(&mut vm, config);
                         black_box(vm.execute().unwrap())
                     })
                 });
@@ -445,7 +468,7 @@ fn bench_hot(c: &mut Criterion) {
                 if let Some(ref d) = data {
                     vm.set_data(d.clone()).unwrap();
                 }
-                configure_limits(&mut vm, config.limits);
+                configure_limits(&mut vm, config);
 
                 // Warm up: fill register window pools, caches, etc.
                 vm.set_input(inputs[0].clone());

bindings/csharp/API.md

@@ -81,6 +81,45 @@ The Regorus C# bindings provide a modern, thread-safe API for compiling and eval
 - **Thread Safety**: All operations are thread-safe without external synchronization
 - **Registry Management**: Centralized management of targets and schemas
 - **Policy Introspection**: Rich metadata about compiled policies
+- **RVM Memory Budgets**: Optional per-execution live-memory limits for run-to-completion evaluation
+
+## RVM Memory Budgets
+
+`Rvm.SetMemoryBudgetConfig` configures a fresh non-zero budget for each run-to-completion execution. `Rvm.ClearMemoryBudgetConfig` restores unlimited execution.
+
+```csharp
+public readonly struct MemoryBudgetConfig
+{
+    public MemoryBudgetConfig(ulong limitBytes);
+    public ulong LimitBytes { get; }
+}
+
+public sealed class Rvm : IDisposable
+{
+    public void SetMemoryBudgetConfig(MemoryBudgetConfig config);
+    public void ClearMemoryBudgetConfig();
+}
+```
+
+Ordinary `Execute` and `ExecuteEntryPoint` calls use the existing VM data and start a fresh budget for execution. Program compilation, program loading, and prior `SetDataJson`, `SetInputJson`, and `SetContextJson` calls occur before and outside the budget; use this path for static or preloaded data.
+
+```csharp
+using var vm = new Rvm();
+vm.SetMemoryBudgetConfig(new MemoryBudgetConfig(16UL * 1024 * 1024));
+
+try
+{
+    var result = vm.Execute();
+}
+catch (RegorusMemoryBudgetExceededException)
+{
+    // The execution exceeded its configured budget.
+}
+```
+
+Native result JSON serialization and C-string allocation are included before an `Execute` or `ExecuteEntryPoint` call returns. Managed UTF-8 decoding and the managed C# `string` allocation after the native call returns are excluded.
+
+Memory budgets require a native library built with allocator memory tracking and are supported only for run-to-completion execution. `RegorusMemoryBudgetExceededException` is thrown when a budget is exceeded. `RegorusMemoryBudgetUnsupportedException` is thrown if a configured budget is used to start or resume suspendable execution. Enforcement is cooperative, so one instruction can overshoot before the next checkpoint. Same-thread baseline ratcheting can make the effective limit stricter after unrelated frees are observed; those frees are never credited back. Failed terminal execution clears retained state, and a reused VM gets a fresh budget. Public multi-call begin/end scopes are intentionally absent because allocator counters are thread-local and abandoned or cross-thread scopes would be unsafe.
 
 ## Core Classes
 

bindings/csharp/README.md

@@ -105,6 +105,31 @@ var result = vm.Execute();
 Console.WriteLine($"allow: {result}");
 ```
 
+### Per-execution memory budget
+
+RVM run-to-completion evaluation can use an optional additional live-memory budget. Each ordinary `Execute` or `ExecuteEntryPoint` call starts with a fresh budget for execution; program compilation, program loading, and prior `SetDataJson`, `SetInputJson`, and `SetContextJson` calls occur before and outside that budget.
+
+```csharp
+using var vm = new Rvm();
+vm.LoadProgram(program);
+vm.SetDataJson(Data);
+vm.SetInputJson(Input);
+vm.SetMemoryBudgetConfig(new MemoryBudgetConfig(16 * 1024 * 1024));
+
+try
+{
+    var result = vm.Execute();
+}
+catch (RegorusMemoryBudgetExceededException ex)
+{
+    Console.WriteLine(ex.Message);
+}
+```
+
+The native execution budget includes result JSON serialization and C-string allocation before the native call returns. Managed UTF-8 decoding and C# `string` allocation after that return are not charged.
+
+The budget is cooperative and may overshoot between VM checks. Same-thread allocation-counter baseline ratcheting can make the effective limit stricter after unrelated frees are observed; those frees are not credited back. Budgets are not supported in suspendable execution mode. `ClearMemoryBudgetConfig` restores the previous unlimited per-execution behavior. Public multi-call begin/end scopes are intentionally absent because allocator counters are thread-local. Failed terminal execution clears retained state, and a reused VM starts a fresh budget. The process-wide limit exposed by `MemoryLimits` remains a separate safeguard.
+
 ## Azure RBAC Condition Evaluation
 
 Evaluate Azure RBAC condition expressions directly with a JSON evaluation context:

bindings/csharp/Regorus.Tests/RvmMemoryBudgetTests.cs

@@ -0,0 +1,120 @@
+// Copyright (c) Microsoft Corporation.
+// Licensed under the MIT License.
+
+using System;
+using System.Linq;
+using System.Text.Json;
+using Microsoft.VisualStudio.TestTools.UnitTesting;
+
+namespace Regorus.Tests;
+
+[TestClass]
+public sealed class RvmMemoryBudgetTests
+{
+    private const ulong TightMemoryBudgetBytes = 64 * 1024;
+
+    private const string Policy = """
+package limits.memory
+import rego.v1
+
+large_array := json.unmarshal(data.large_json)
+""";
+
+    private const string EntryPoint = "data.limits.memory.large_array";
+
+    private const string PreloadedResultPolicy = """
+package limits.memory
+
+large_string := data.large_string
+""";
+
+    private const string PreloadedResultEntryPoint = "data.limits.memory.large_string";
+
+    [TestMethod]
+    public void Memory_budget_must_be_non_zero()
+    {
+        Assert.ThrowsException<ArgumentOutOfRangeException>(() => new MemoryBudgetConfig(0));
+
+        using var vm = new Rvm();
+        Assert.ThrowsException<ArgumentOutOfRangeException>(() => vm.SetMemoryBudgetConfig(default));
+    }
+
+    [TestMethod]
+    public void Execute_exceeding_memory_budget_throws_typed_exception()
+    {
+        using var program = CreateProgram();
+        using var vm = CreateRvm(program);
+        vm.SetMemoryBudgetConfig(new MemoryBudgetConfig(TightMemoryBudgetBytes));
+
+        Assert.ThrowsException<RegorusMemoryBudgetExceededException>(() => vm.ExecuteEntryPoint(EntryPoint));
+    }
+
+    [TestMethod]
+    public void Clearing_memory_budget_restores_unlimited_execution()
+    {
+        using var program = CreateProgram();
+        using var vm = CreateRvm(program);
+        vm.SetMemoryBudgetConfig(new MemoryBudgetConfig(TightMemoryBudgetBytes));
+        Assert.ThrowsException<RegorusMemoryBudgetExceededException>(() => vm.ExecuteEntryPoint(EntryPoint));
+
+        vm.ClearMemoryBudgetConfig();
+
+        var result = vm.ExecuteEntryPoint(EntryPoint);
+        Assert.IsFalse(string.IsNullOrWhiteSpace(result));
+    }
+
+    [TestMethod]
+    public void Serialization_budget_failure_leaves_error_state()
+    {
+        var data = JsonSerializer.Serialize(new
+        {
+            large_string = new string('x', 2 * 1024 * 1024),
+        });
+        var modules = new[] { new PolicyModule("memory_budget.rego", PreloadedResultPolicy) };
+        using var program = Program.CompileFromModules(data, modules, new[] { PreloadedResultEntryPoint });
+        using var vm = new Rvm();
+        vm.LoadProgram(program);
+        vm.SetDataJson(data);
+        vm.SetMemoryBudgetConfig(new MemoryBudgetConfig(512 * 1024));
+
+        Assert.ThrowsException<RegorusMemoryBudgetExceededException>(
+            () => vm.ExecuteEntryPoint(PreloadedResultEntryPoint));
+
+        var state = vm.GetExecutionState();
+        Assert.IsNotNull(state);
+        StringAssert.Contains(state, "Error { error: MemoryBudgetExceeded");
+    }
+
+    [TestMethod]
+    public void Suspendable_execution_rejects_memory_budget()
+    {
+        using var vm = new Rvm();
+        vm.SetExecutionMode(ExecutionMode.Suspendable);
+        vm.SetMemoryBudgetConfig(new MemoryBudgetConfig(1024));
+
+        Assert.ThrowsException<RegorusMemoryBudgetUnsupportedException>(() => vm.Execute());
+    }
+
+    private static Program CreateProgram()
+    {
+        var modules = new[] { new PolicyModule("memory_budget.rego", Policy) };
+        return Program.CompileFromModules(CreateData(), modules, new[] { EntryPoint });
+    }
+
+    private static Rvm CreateRvm(Program program)
+    {
+        var vm = new Rvm();
+        vm.LoadProgram(program);
+        vm.SetDataJson(CreateData());
+        return vm;
+    }
+
+    private static string CreateData()
+    {
+        var values = Enumerable.Range(0, 200_000).ToArray();
+        return JsonSerializer.Serialize(new
+        {
+            large_json = JsonSerializer.Serialize(values),
+        });
+    }
+}

bindings/csharp/Regorus/MemoryBudgetConfig.cs

@@ -0,0 +1,46 @@
+// Copyright (c) Microsoft Corporation.
+// Licensed under the MIT License.
+
+using System;
+
+namespace Regorus
+{
+    /// <summary>
+    /// Configures the additional live-memory budget for one RVM execution.
+    /// </summary>
+    public readonly struct MemoryBudgetConfig
+    {
+        /// <summary>
+        /// Initializes a new instance of the <see cref="MemoryBudgetConfig"/> struct.
+        /// </summary>
+        /// <param name="limitBytes">Maximum additional live bytes allowed during one execution.</param>
+        /// <exception cref="ArgumentOutOfRangeException">Thrown when <paramref name="limitBytes"/> is zero.</exception>
+        public MemoryBudgetConfig(ulong limitBytes)
+        {
+            if (limitBytes == 0)
+            {
+                throw new ArgumentOutOfRangeException(nameof(limitBytes), "Memory budget must be non-zero.");
+            }
+
+            LimitBytes = limitBytes;
+        }
+
+        /// <summary>
+        /// Gets the maximum additional live bytes allowed during one execution.
+        /// </summary>
+        public ulong LimitBytes { get; }
+
+        internal Regorus.Internal.RegorusMemoryBudgetConfig ToNative()
+        {
+            if (LimitBytes == 0)
+            {
+                throw new ArgumentOutOfRangeException(nameof(LimitBytes), "Memory budget must be non-zero.");
+            }
+
+            return new Regorus.Internal.RegorusMemoryBudgetConfig
+            {
+                limit_bytes = LimitBytes,
+            };
+        }
+    }
+}