f50b29abae
v2-ci / unit-tests (push) Failing after 1h12m10s
Second P1 slice of docs/plans/2026-11-09-out-of-process-script-runner-design.md; still opt-in behind Scripting:Runner=OutOfProcess (InProcess unchanged). - Compiled-PE cache (design §7, §14 #8): Scripting:Worker:PeCacheDir holds <dir>/pe/<id>.pe entries authenticated with HMAC-SHA256 under a per-node random key (<dir>/pe-cache.key, created by the host, 0600/0700 on Unix), handed to the worker in Hello. Entry id = SHA-256(compile fingerprint + kind + source); the fingerprint covers the Roslyn version, the Core.Scripting build (analyzer, wrapper, stack guard), the sandbox reference set and the runtime. Edited, truncated, other-key or renamed entries are rejected, deleted and recompiled; the worker never loads unverified bytes. Pruned by recency at startup (30 d / 50 000). ScriptEvaluator splits into CompileToImage + FromImage. Measured on 40 scripts, restart to every first result: macOS 820 ms no cache / 840 ms empty / 85 ms warm; windev (Windows) 9.5 s / 9.4 s / 1.0 s. - Quarantine race: WorkerConnection runs exit attribution synchronously BEFORE completing the evaluations/loads it held (and before Exited), and the shard passes it straight to the runner — a kill strike is in place before the killed key's slot is released. - Supervision: a dead worker restarts by itself after its backoff; Unavailable outcomes carry ScriptRunOutcome.Recovery (-> VirtualTagEvalResult.RetryWhen) and the VT actor / Calculation driver re-evaluate once against current inputs when it completes. Isolation worker stops after Scripting:Worker:IsolationIdleTimeout (10 min). Keys compiling when a worker dies are suspects too (isolation finds a poison compile and quarantines it). Startup orphan sweep by launch marker: kills only a process whose marker's host is gone and whose pid + start time + image match. - Crash-loop circuit (Scripting:Worker:CircuitOpenAfter, 2 min): runner Unavailable, one Error, evaluations answered at once while shards keep retrying at the cap. - Observability (§10): otopcua.script.runner.state, worker.up/inflight/queue_depth/heap_bytes (shard), ipc.roundtrip + eval.duration (kind), compile{cache}, isolation.suspects, exits reason=idle. `script-runner` health check on the ready tier, at worst Degraded (§14 #12: degraded but still ready). Left on #780: self-contained worker publish, installer (account, pipe ACL, firewall), docker-dev live drill, scripted-alarm seam, the §4.5 per-seam batch submit, the §8 perf harness. Tests: SupervisionTests, PeCacheStoreTests/PeCacheWorkerTests (real workers), ScriptRunnerHealthCheckTests, VT + Calc recovery seam tests; 14 mutations each turned their test red. Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
293 lines
13 KiB
C#
293 lines
13 KiB
C#
using System.Security.Cryptography;
|
|
using System.Text;
|
|
|
|
namespace ZB.MOM.WW.OtOpcUa.Core.Scripting.Protocol;
|
|
|
|
/// <summary>What a <see cref="ScriptPeCacheStore.TryRead"/> found.</summary>
|
|
public enum PeCacheLookup
|
|
{
|
|
/// <summary>No entry for the id.</summary>
|
|
Miss = 0,
|
|
|
|
/// <summary>An entry whose HMAC verified; its PE bytes are returned.</summary>
|
|
Hit = 1,
|
|
|
|
/// <summary>
|
|
/// An entry that exists but did not verify — truncated, edited, written under another key, or renamed from
|
|
/// another id. It has been deleted; the caller recompiles.
|
|
/// </summary>
|
|
Rejected = 2,
|
|
}
|
|
|
|
/// <summary>
|
|
/// The node-local compiled-PE cache (lmxopcua#780 design §7, owner decision §14 #8): content-addressed files under
|
|
/// <c>Scripting:Worker:PeCacheDir</c>, each authenticated with HMAC-SHA256 under a per-node random key, so a worker
|
|
/// restart loads its scripts' PE images instead of re-entering Roslyn — and never loads bytes it did not write.
|
|
/// </summary>
|
|
/// <remarks>
|
|
/// <list type="bullet">
|
|
/// <item><b>Layout.</b> <c><dir>/pe-cache.key</c> (32 random bytes, created once by the host) and
|
|
/// <c><dir>/pe/<id>.pe</c>, one file per entry: an 8-byte magic, the 32-byte tag, then the PE.</item>
|
|
/// <item><b>Identity.</b> <see cref="EntryId"/> is SHA-256 over the compile fingerprint (Roslyn version, the
|
|
/// sandbox and analyzer build, the reference set, the runtime), the context kind and the source. A compiler or
|
|
/// analyzer change therefore misses every old entry; it can never be served an image a newer sandbox rule
|
|
/// would have refused.</item>
|
|
/// <item><b>Authentication.</b> The tag covers the magic, the id and the PE, so an entry copied to another id
|
|
/// is rejected as well as an edited or truncated one. Loading unauthenticated PE bytes would be code execution
|
|
/// outside the analyzer; a rejected entry is deleted and the source recompiled.</item>
|
|
/// <item><b>Key.</b> Readable only by the host account: on Linux/macOS the directory is 0700 and the key file
|
|
/// 0600. On Windows the file inherits the directory's ACL, which must be the installer's (lmxopcua#780 installer
|
|
/// work). The worker receives the key in <c>Hello</c>, never on disk or in its environment.</item>
|
|
/// <item><b>Writes</b> go to a unique temp file and are renamed into place, so a concurrent reader (another
|
|
/// shard) sees the old entry, the new one, or none — never a torn file.</item>
|
|
/// </list>
|
|
/// </remarks>
|
|
public sealed class ScriptPeCacheStore
|
|
{
|
|
/// <summary>The key file's name under the cache directory.</summary>
|
|
public const string KeyFileName = "pe-cache.key";
|
|
|
|
/// <summary>The entries' subdirectory.</summary>
|
|
public const string EntriesDirectoryName = "pe";
|
|
|
|
/// <summary>The HMAC key length.</summary>
|
|
public const int KeyLength = 32;
|
|
|
|
/// <summary>An entry above this size is not read (a 4 MiB source compiles to far less).</summary>
|
|
public const int MaxEntryBytes = 32 * 1024 * 1024;
|
|
|
|
private const string EntryExtension = ".pe";
|
|
private static readonly byte[] Magic = "OTPEC\u0001\0\0"u8.ToArray();
|
|
private const int TagLength = 32;
|
|
|
|
private readonly byte[] _key;
|
|
|
|
/// <summary>Opens the store over <paramref name="directory"/> with <paramref name="key"/>.</summary>
|
|
/// <param name="directory">The cache directory (absolute).</param>
|
|
/// <param name="key">The node's HMAC key (<see cref="KeyLength"/> bytes).</param>
|
|
public ScriptPeCacheStore(string directory, byte[] key)
|
|
{
|
|
ArgumentException.ThrowIfNullOrWhiteSpace(directory);
|
|
ArgumentNullException.ThrowIfNull(key);
|
|
if (key.Length != KeyLength) throw new ArgumentException($"the PE cache key must be {KeyLength} bytes", nameof(key));
|
|
Directory = directory;
|
|
EntriesDirectory = Path.Combine(directory, EntriesDirectoryName);
|
|
_key = (byte[])key.Clone();
|
|
}
|
|
|
|
/// <summary>The cache directory.</summary>
|
|
public string Directory { get; }
|
|
|
|
/// <summary>The entries' directory.</summary>
|
|
public string EntriesDirectory { get; }
|
|
|
|
/// <summary>
|
|
/// The host half: returns the node's key, creating the directory and the key on first use. An existing key of
|
|
/// the wrong length is not replaced (it may be a sign of tampering); the call throws and the caller runs without
|
|
/// a cache.
|
|
/// </summary>
|
|
/// <param name="directory">The cache directory (absolute).</param>
|
|
/// <returns>The key.</returns>
|
|
/// <exception cref="IOException">The directory or key cannot be created or read, or the key is malformed.</exception>
|
|
public static byte[] LoadOrCreateKey(string directory)
|
|
{
|
|
ArgumentException.ThrowIfNullOrWhiteSpace(directory);
|
|
CreatePrivateDirectory(directory);
|
|
CreatePrivateDirectory(Path.Combine(directory, EntriesDirectoryName));
|
|
var keyPath = Path.Combine(directory, KeyFileName);
|
|
if (!File.Exists(keyPath))
|
|
{
|
|
// Written to a temp name and moved without overwrite: two hosts racing on first start agree on one key.
|
|
var temp = keyPath + "." + Convert.ToHexString(RandomNumberGenerator.GetBytes(8)) + ".tmp";
|
|
var options = new FileStreamOptions { Mode = FileMode.CreateNew, Access = FileAccess.Write, Share = FileShare.None };
|
|
if (!OperatingSystem.IsWindows()) options.UnixCreateMode = UnixFileMode.UserRead | UnixFileMode.UserWrite;
|
|
using (var stream = new FileStream(temp, options))
|
|
{
|
|
stream.Write(RandomNumberGenerator.GetBytes(KeyLength));
|
|
}
|
|
try
|
|
{
|
|
File.Move(temp, keyPath, overwrite: false);
|
|
}
|
|
catch (IOException) when (File.Exists(keyPath))
|
|
{
|
|
File.Delete(temp);
|
|
}
|
|
}
|
|
var key = File.ReadAllBytes(keyPath);
|
|
if (key.Length != KeyLength)
|
|
throw new IOException($"the PE cache key '{keyPath}' is {key.Length} bytes, not {KeyLength}; delete it to start a fresh cache");
|
|
return key;
|
|
}
|
|
|
|
/// <summary>The entry id for one compile: hex SHA-256 over the fingerprint, the kind and the source.</summary>
|
|
/// <param name="fingerprint">The compile fingerprint (<c>ScriptEvaluator.CacheFingerprint</c>).</param>
|
|
/// <param name="kind">The context kind (the wire enum value).</param>
|
|
/// <param name="source">The script source.</param>
|
|
/// <returns>A 64-character lowercase hex id.</returns>
|
|
public static string EntryId(ReadOnlySpan<byte> fingerprint, int kind, string source)
|
|
{
|
|
ArgumentNullException.ThrowIfNull(source);
|
|
using var hash = IncrementalHash.CreateHash(HashAlgorithmName.SHA256);
|
|
hash.AppendData(fingerprint);
|
|
hash.AppendData(BitConverter.GetBytes(kind));
|
|
hash.AppendData(Encoding.UTF8.GetBytes(source));
|
|
return Convert.ToHexString(hash.GetHashAndReset()).ToLowerInvariant();
|
|
}
|
|
|
|
/// <summary>The path of an entry.</summary>
|
|
/// <param name="entryId">The id.</param>
|
|
/// <returns>The file path.</returns>
|
|
public string EntryPath(string entryId)
|
|
{
|
|
if (!IsValidId(entryId)) throw new ArgumentException("not a PE cache entry id", nameof(entryId));
|
|
return Path.Combine(EntriesDirectory, entryId + EntryExtension);
|
|
}
|
|
|
|
/// <summary>Reads and authenticates an entry. A rejected entry is deleted.</summary>
|
|
/// <param name="entryId">The id.</param>
|
|
/// <param name="peImage">The PE bytes on a hit.</param>
|
|
/// <returns>Hit, miss, or rejected.</returns>
|
|
public PeCacheLookup TryRead(string entryId, out byte[]? peImage)
|
|
{
|
|
peImage = null;
|
|
var path = EntryPath(entryId);
|
|
byte[] bytes;
|
|
try
|
|
{
|
|
var info = new FileInfo(path);
|
|
if (!info.Exists) return PeCacheLookup.Miss;
|
|
if (info.Length > MaxEntryBytes)
|
|
{
|
|
TryDelete(path);
|
|
return PeCacheLookup.Rejected;
|
|
}
|
|
bytes = File.ReadAllBytes(path);
|
|
}
|
|
catch (FileNotFoundException)
|
|
{
|
|
return PeCacheLookup.Miss;
|
|
}
|
|
catch (DirectoryNotFoundException)
|
|
{
|
|
return PeCacheLookup.Miss;
|
|
}
|
|
|
|
if (bytes.Length <= Magic.Length + TagLength || !bytes.AsSpan(0, Magic.Length).SequenceEqual(Magic))
|
|
{
|
|
TryDelete(path);
|
|
return PeCacheLookup.Rejected;
|
|
}
|
|
var tag = bytes.AsSpan(Magic.Length, TagLength);
|
|
var pe = bytes.AsSpan(Magic.Length + TagLength);
|
|
if (!CryptographicOperations.FixedTimeEquals(tag, Tag(entryId, pe)))
|
|
{
|
|
TryDelete(path);
|
|
return PeCacheLookup.Rejected;
|
|
}
|
|
|
|
peImage = pe.ToArray();
|
|
// Recency for Prune: a hit counts as a use.
|
|
try { File.SetLastWriteTimeUtc(path, DateTime.UtcNow); } catch (IOException) { } catch (UnauthorizedAccessException) { }
|
|
return PeCacheLookup.Hit;
|
|
}
|
|
|
|
/// <summary>Writes an entry (atomically replacing any existing one).</summary>
|
|
/// <param name="entryId">The id.</param>
|
|
/// <param name="peImage">The PE bytes.</param>
|
|
/// <returns>False when the write failed (the cache is best effort: a failed write costs a recompile later).</returns>
|
|
public bool TryWrite(string entryId, ReadOnlySpan<byte> peImage)
|
|
{
|
|
var path = EntryPath(entryId);
|
|
var temp = Path.Combine(EntriesDirectory, $".{entryId}.{Convert.ToHexString(RandomNumberGenerator.GetBytes(6))}.tmp");
|
|
try
|
|
{
|
|
System.IO.Directory.CreateDirectory(EntriesDirectory);
|
|
var options = new FileStreamOptions { Mode = FileMode.CreateNew, Access = FileAccess.Write, Share = FileShare.None };
|
|
if (!OperatingSystem.IsWindows()) options.UnixCreateMode = UnixFileMode.UserRead | UnixFileMode.UserWrite;
|
|
using (var stream = new FileStream(temp, options))
|
|
{
|
|
stream.Write(Magic);
|
|
stream.Write(Tag(entryId, peImage));
|
|
stream.Write(peImage);
|
|
}
|
|
File.Move(temp, path, overwrite: true);
|
|
return true;
|
|
}
|
|
catch (Exception ex) when (ex is IOException or UnauthorizedAccessException)
|
|
{
|
|
TryDelete(temp);
|
|
return false;
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Deletes entries not used for <paramref name="maxAge"/>, then the least recently used beyond
|
|
/// <paramref name="maxEntries"/>, plus stray temp files. The design prunes "entries not in the current
|
|
/// generation"; this runner has no generation object, so recency stands in for it.
|
|
/// </summary>
|
|
/// <param name="maxAge">Entries older than this go.</param>
|
|
/// <param name="maxEntries">At most this many entries stay.</param>
|
|
/// <returns>How many files were deleted.</returns>
|
|
public int Prune(TimeSpan maxAge, int maxEntries)
|
|
{
|
|
if (!System.IO.Directory.Exists(EntriesDirectory)) return 0;
|
|
var deleted = 0;
|
|
var now = DateTime.UtcNow;
|
|
var entries = new List<FileInfo>();
|
|
foreach (var file in new DirectoryInfo(EntriesDirectory).EnumerateFiles())
|
|
{
|
|
var stale = now - file.LastWriteTimeUtc;
|
|
if (file.Name.EndsWith(".tmp", StringComparison.Ordinal))
|
|
{
|
|
if (stale > TimeSpan.FromHours(1) && TryDelete(file.FullName)) deleted++;
|
|
continue;
|
|
}
|
|
if (!file.Name.EndsWith(EntryExtension, StringComparison.Ordinal)) continue;
|
|
if (stale > maxAge)
|
|
{
|
|
if (TryDelete(file.FullName)) deleted++;
|
|
continue;
|
|
}
|
|
entries.Add(file);
|
|
}
|
|
if (entries.Count > maxEntries)
|
|
{
|
|
foreach (var file in entries.OrderBy(f => f.LastWriteTimeUtc).Take(entries.Count - maxEntries))
|
|
if (TryDelete(file.FullName)) deleted++;
|
|
}
|
|
return deleted;
|
|
}
|
|
|
|
private byte[] Tag(string entryId, ReadOnlySpan<byte> pe)
|
|
{
|
|
using var hmac = IncrementalHash.CreateHMAC(HashAlgorithmName.SHA256, _key);
|
|
hmac.AppendData(Magic);
|
|
hmac.AppendData(Encoding.ASCII.GetBytes(entryId));
|
|
hmac.AppendData(pe);
|
|
return hmac.GetHashAndReset();
|
|
}
|
|
|
|
private static bool IsValidId(string id)
|
|
=> id is { Length: 64 } && id.All(static c => c is >= '0' and <= '9' or >= 'a' and <= 'f');
|
|
|
|
private static void CreatePrivateDirectory(string path)
|
|
{
|
|
if (OperatingSystem.IsWindows()) System.IO.Directory.CreateDirectory(path);
|
|
else System.IO.Directory.CreateDirectory(path, UnixFileMode.UserRead | UnixFileMode.UserWrite | UnixFileMode.UserExecute);
|
|
}
|
|
|
|
private static bool TryDelete(string path)
|
|
{
|
|
try
|
|
{
|
|
File.Delete(path);
|
|
return true;
|
|
}
|
|
catch (Exception ex) when (ex is IOException or UnauthorizedAccessException)
|
|
{
|
|
return false;
|
|
}
|
|
}
|
|
}
|