Relax only stressed remote bridge waits

Apply a bounded test-only timeout multiplier to the remote suite when it runs concurrently with privacy scans. Production defaults and the standalone strict suite remain unchanged.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: Colin Neilens <coneilen@microsoft.com>

Author: coneilen

Tools/windows/Tests/RemoteBridgePrivacyRace.Tests.ps1

@@ -11,7 +11,11 @@ if (-not $python) {
   throw "Python 3 was not found for the remote bridge race regression"
 }
 
-function Start-CapturedProcess([string] $fileName, [string[]] $arguments) {
+function Start-CapturedProcess(
+  [string] $fileName,
+  [string[]] $arguments,
+  [hashtable] $environment = @{}
+) {
   $startInfo = [Diagnostics.ProcessStartInfo]::new()
   $startInfo.FileName = $fileName
   $startInfo.UseShellExecute = $false
@@ -21,6 +25,9 @@ function Start-CapturedProcess([string] $fileName, [string[]] $arguments) {
   foreach ($argument in $arguments) {
     [void] $startInfo.ArgumentList.Add($argument)
   }
+  foreach ($entry in $environment.GetEnumerator()) {
+    $startInfo.Environment[$entry.Key] = $entry.Value
+  }
   $process = [Diagnostics.Process]::new()
   $process.StartInfo = $startInfo
   [void] $process.Start()
@@ -51,7 +58,9 @@ $privacyProcessCount = [Math]::Min(24, [Math]::Max(4, $processorCount * 2))
 Write-Host "Remote bridge privacy race: processors=$processorCount, remote=$remoteProcessCount, privacy=$privacyProcessCount"
 $remoteProcesses = @(
   1..$remoteProcessCount | ForEach-Object {
-    Start-CapturedProcess $python.Source $remoteArguments
+    Start-CapturedProcess $python.Source $remoteArguments @{
+      GRAPHCODE_REMOTE_BRIDGE_TEST_TIMEOUT_MULTIPLIER = "3"
+    }
   }
 )
 $privacyProcesses = @(

Tools/windows/Tests/ValidationRunner.Tests.ps1

@@ -107,7 +107,8 @@ try {
     (Join-Path $repoRoot "Tools\windows\Tests\RemoteBridgePrivacyRace.Tests.ps1") -Raw
   if ($privacyRaceSource -notmatch '\$AvailableProcessorCount = \[Environment\]::ProcessorCount' -or
       $privacyRaceSource -notmatch '\$remoteProcessCount = 1' -or
-      $privacyRaceSource -notmatch '\[Math\]::Min\(24, \[Math\]::Max\(4, \$processorCount \* 2\)\)') {
+      $privacyRaceSource -notmatch '\[Math\]::Min\(24, \[Math\]::Max\(4, \$processorCount \* 2\)\)' -or
+      $privacyRaceSource -notmatch 'GRAPHCODE_REMOTE_BRIDGE_TEST_TIMEOUT_MULTIPLIER = "3"') {
     throw "RED: remote bridge privacy race does not scale bounded concurrency to runner capacity"
   }
   if ($windowsWorkflow -notmatch "(?s)environment:.*Hardening\.Tests\.ps1 -Environment -SchemaOnly") {

investigation/spikes/remote-bridge/test_remote_bridge.py

@@ -13,7 +13,15 @@
 
 
 SPIKE_ROOT = Path(__file__).resolve().parent
-THREAD_TIMEOUT = 5.0
+TIMEOUT_MULTIPLIER = float(
+    os.environ.get("GRAPHCODE_REMOTE_BRIDGE_TEST_TIMEOUT_MULTIPLIER", "1")
+)
+if not math.isfinite(TIMEOUT_MULTIPLIER) or not 1 <= TIMEOUT_MULTIPLIER <= 10:
+    raise ValueError(
+        "GRAPHCODE_REMOTE_BRIDGE_TEST_TIMEOUT_MULTIPLIER must be between 1 and 10"
+    )
+SOCKET_TIMEOUT = 1.0 * TIMEOUT_MULTIPLIER
+THREAD_TIMEOUT = 5.0 * TIMEOUT_MULTIPLIER
 sys.path.insert(0, str(SPIKE_ROOT))
 
 from remote_bridge import (  # noqa: E402
@@ -40,6 +48,7 @@ def setUp(self):
             self.backend.address,
             ttl_seconds=30.0,
             previous_overlap_seconds=0.4,
+            request_timeout=2.0 * TIMEOUT_MULTIPLIER,
         )
         self.bridge.start()
 
@@ -57,7 +66,7 @@ def raw_request(
         omit_capability=False,
     ):
         with socket.create_connection(
-            (state["host"], state["port"]), timeout=1.0
+            (state["host"], state["port"]), timeout=SOCKET_TIMEOUT
         ) as connection:
             message = {
                 "generation": generation or state["generation"],
@@ -99,21 +108,21 @@ def test_slow_drip_connections_are_bounded_by_cumulative_deadline(self):
             for _ in range(2):
                 connection = socket.create_connection(
                     (state["host"], state["port"]),
-                    timeout=1.0,
+                    timeout=SOCKET_TIMEOUT,
                 )
                 connection.sendall(b"\0")
                 connections.append(connection)
-            deadline = time.time() + 1
+            deadline = time.time() + SOCKET_TIMEOUT
             while self.bridge.active_client_count < 2:
                 if time.time() >= deadline:
                     self.fail("bounded client workers did not start")
                 time.sleep(0.01)
 
             rejected = socket.create_connection(
                 (state["host"], state["port"]),
-                timeout=1.0,
+                timeout=SOCKET_TIMEOUT,
             )
-            rejected.settimeout(1.0)
+            rejected.settimeout(SOCKET_TIMEOUT)
             try:
                 self.assertEqual(rejected.recv(1), b"")
             finally:
@@ -125,15 +134,18 @@ def test_slow_drip_connections_are_bounded_by_cumulative_deadline(self):
                     connection.sendall(b"\0")
                 except OSError:
                     pass
-            expiry = time.monotonic() + 0.35
+            expiry = time.monotonic() + 0.35 * TIMEOUT_MULTIPLIER
             while self.bridge.active_client_count:
                 if time.monotonic() >= expiry:
                     self.fail("slow-drip worker exceeded cumulative deadline plus scheduler margin")
                 time.sleep(0.005)
             self.assertEqual(self.bridge.active_client_count, 0)
             stop_started = time.monotonic()
             self.bridge.stop()
-            self.assertLess(time.monotonic() - stop_started, 1.0)
+            self.assertLess(
+                time.monotonic() - stop_started,
+                1.0 * TIMEOUT_MULTIPLIER,
+            )
             self.assertEqual(self.bridge.worker_count, 0)
         finally:
             for connection in connections:
@@ -144,18 +156,18 @@ def test_stop_closes_active_client_sockets(self):
         state = BridgeStateStore(self.state_path).read()
         connection = socket.create_connection(
             (state["host"], state["port"]),
-            timeout=1.0,
+            timeout=SOCKET_TIMEOUT,
         )
         connection.sendall(b"\0")
-        deadline = time.time() + 1
+        deadline = time.time() + SOCKET_TIMEOUT
         while self.bridge.active_client_count < 1:
             if time.time() >= deadline:
                 connection.close()
                 self.fail("client worker did not start")
             time.sleep(0.01)
 
         self.bridge.stop()
-        connection.settimeout(1.0)
+        connection.settimeout(SOCKET_TIMEOUT)
         try:
             result = connection.recv(1)
         except ConnectionResetError:
@@ -215,7 +227,10 @@ def test_state_timestamps_must_be_finite(self):
                     invalid_path.unlink(missing_ok=True)
 
     def test_client_reads_state_and_bridges_framed_request_response(self):
-        client = RemoteBridgeClient(self.state_path)
+        client = RemoteBridgeClient(
+            self.state_path,
+            timeout=2.0 * TIMEOUT_MULTIPLIER,
+        )
 
         response = client.request({"command": "status"})
 
@@ -226,7 +241,7 @@ def test_client_reads_state_and_bridges_framed_request_response(self):
     def test_authenticated_session_relays_multiple_frames_on_one_connection(self):
         state = BridgeStateStore(self.state_path).read()
         with socket.create_connection(
-            (state["host"], state["port"]), timeout=1.0
+            (state["host"], state["port"]), timeout=SOCKET_TIMEOUT
         ) as connection:
             bodies = ({"command": "openProject"}, {"command": "status"})
             for body in bodies[:1]:
@@ -339,7 +354,7 @@ def test_malformed_frame_is_rejected(self):
         state = BridgeStateStore(self.state_path).read()
         payload = b"not-json"
         with socket.create_connection(
-            (state["host"], state["port"]), timeout=1.0
+            (state["host"], state["port"]), timeout=SOCKET_TIMEOUT
         ) as connection:
             connection.sendall(len(payload).to_bytes(4, "big") + payload)
             response = read_frame(connection)
@@ -731,7 +746,7 @@ def read_states():
     def test_oversized_frame_is_rejected(self):
         state = BridgeStateStore(self.state_path).read()
         with socket.create_connection(
-            (state["host"], state["port"]), timeout=1.0
+            (state["host"], state["port"]), timeout=SOCKET_TIMEOUT
         ) as connection:
             connection.sendall((1_048_577).to_bytes(4, "big"))
             response = read_frame(connection)