bazelbuild · brentleyjones · Apr 4, 2025
@@ -14,7 +14,7 @@ def _get_template_substitutions(
         create_xcresult_bundle,
         device_type,
         os_version,
-        simulator_creator,
+        simulator_manager_start,
         random,
         xcodebuild_args,
         command_line_args,
@@ -27,12 +27,12 @@ def _get_template_substitutions(
         post_action_binary,
         post_action_determines_exit_code):
     substitutions = {
-        "device_type": device_type,
+        "url_encoded_device_type": device_type.replace(" ", "%20"),
         "os_version": os_version,
         "create_xcresult_bundle": create_xcresult_bundle,
         "xcodebuild_args": xcodebuild_args,
         "command_line_args": command_line_args,
-        "simulator_creator.py": simulator_creator,
+        "simulator_manager_start": simulator_manager_start,
         # "ordered" isn't a special string, but anything besides "random" for this field runs in order
         "test_order": "random" if random else "ordered",
         "xctestrun_template": xctestrun_template,
@@ -69,7 +69,7 @@ def _ios_xctestrun_runner_impl(ctx):
     runfiles = ctx.runfiles(files = [
         ctx.file._xctestrun_template,
         ctx.file._xctrunner_entitlements_template,
-    ]).merge(ctx.attr._simulator_creator[DefaultInfo].default_runfiles)
+    ]).merge(ctx.attr._simulator_manager_start[DefaultInfo].default_runfiles)
 
     default_action_binary = "/usr/bin/true"
 
@@ -93,7 +93,7 @@ def _ios_xctestrun_runner_impl(ctx):
             create_xcresult_bundle = "true" if ctx.attr.create_xcresult_bundle else "false",
             device_type = device_type,
             os_version = os_version,
-            simulator_creator = ctx.executable._simulator_creator.short_path,
+            simulator_manager_start = ctx.executable._simulator_manager_start.short_path,
             random = ctx.attr.random,
             xcodebuild_args = " ".join(ctx.attr.xcodebuild_args) if ctx.attr.xcodebuild_args else "",
             command_line_args = " ".join(ctx.attr.command_line_args) if ctx.attr.command_line_args else "",
@@ -204,9 +204,9 @@ A binary to run following test execution. Runs after testing but before test res
 When true, the exit code of the test run will be set to the exit code of the post action. This is useful for tests that need to fail the test run based on their own criteria.
 """,
         ),
-        "_simulator_creator": attr.label(
+        "_simulator_manager_start": attr.label(
             default = Label(
-                "//apple/testing/default_runner:simulator_creator",
+                "//apple/testing/simulator_manager:start",
             ),
             executable = True,
             cfg = "exec",

@@ -23,17 +23,13 @@ if [[ -n "${CREATE_XCRESULT_BUNDLE:-}" ]]; then
 fi
 
 custom_xcodebuild_args=(%(xcodebuild_args)s)
-simulator_name=""
 device_id=""
 command_line_args=(%(command_line_args)s)
 attachment_lifetime="%(attachment_lifetime)s"
 destination_timeout="%(destination_timeout)s"
 while [[ $# -gt 0 ]]; do
   arg="$1"
   case $arg in
-    --simulator_name=*)
-      simulator_name="${arg##*=}"
-      ;;
     --xcodebuild_args=*)
       xcodebuild_arg="${arg#--xcodebuild_args=}" # Strip "--xcodebuild_args=" prefix
       custom_xcodebuild_args+=("$xcodebuild_arg")
@@ -391,24 +387,71 @@ fi
 
 readonly profraw="$test_tmp_dir/coverage.profraw"
 
-simulator_creator_args=(
-  "%(os_version)s" \
-  "%(device_type)s" \
-  --name "$simulator_name"
-)
+simulator_id=""
+if [[ "$build_for_device" == false ]]; then
+  simulator_manager_command() {
+    local _http_method="$1"
+    local _command_path="$2"
+
+    # Retry up to 10 times with a 1 second delay between each attempt, to allow
+    # for the server to be down during upgrades
+    for i in {1..10}; do
+      set +e
+      local _simulator_response
+      _simulator_response=$(
+        curl \
+          "http:/-/$_command_path" \
+          --request "$_http_method" \
+          --silent \
+          --fail-with-body \
+          --unix-socket "/tmp/simulator_manager.sock"
+      )
+      local _curl_exit_code=$?
+      set -e
+
+      if [[ $_curl_exit_code -eq 0 ]]; then
+        echo "$_simulator_response"
+        return 0
+      fi
 
-reuse_simulator=%(reuse_simulator)s
-if [[ "$reuse_simulator" == true ]]; then
-  simulator_creator_args+=(--reuse-simulator)
-else
-  simulator_creator_args+=(--no-reuse-simulator)
-fi
+      # If the error was a connection issue (e.g., couldn't connect), then retry
+      #   6: couldn't resolve host
+      #   7: failed to connect
+      #   28: operation timeout
+      if [[
+        $_curl_exit_code -eq 6 ||
+        $_curl_exit_code -eq 7 ||
+        $_curl_exit_code -eq 28
+      ]]; then
+        echo >&2 "$(date '+[%H:%M:%S]') warning: simulator manager command" \
+          "\"$_command_path\" failed with exit code $_curl_exit_code:"
+        echo >&2 "$(date '+[%H:%M:%S]') $_simulator_response"
+        echo >&2 "$(date '+[%H:%M:%S]') retrying in 1 second"
+        sleep 1
+      else
+        echo >&2 "$(date '+[%H:%M:%S]') error: simulator manager command" \
+          "\"$_command_path\" failed:"
+        echo >&2 "$(date '+[%H:%M:%S]') $_simulator_response"
+        return $_curl_exit_code
+      fi
+    done
+  }
 
-simulator_id="unused"
-if [[ "$build_for_device" == false ]]; then
-  simulator_id="$("./%(simulator_creator.py)s" \
-    "${simulator_creator_args[@]}"
+  reuse_simulator=%(reuse_simulator)s
+  if [[ "$reuse_simulator" == true ]]; then
+    exclusive_simulator=0
+  else
+    exclusive_simulator=1
+  fi
+
+  "./%(simulator_manager_start)s"
+
+  echo "$(date '+[%H:%M:%S]') Attempting to lease simulator"
+  simulator_id="$(
+    simulator_manager_command POST "simulator/$$?exclusive=$exclusive_simulator&deviceType=%(url_encoded_device_type)s&os=iOS&version=%(os_version)s"
   )"
+
+  echo "$(date '+[%H:%M:%S]') ✅ Leased simulator $simulator_id"
 fi
 
 test_exit_code=0
@@ -585,9 +628,13 @@ if [[
   rm -r "$result_bundle_path"
 fi
 
-if [[ "$reuse_simulator" == false ]]; then
-  # Delete will shutdown down the simulator if it's still currently running.
-  xcrun simctl delete "$simulator_id"
+if [[ -n "$simulator_id" ]]; then
+  echo "$(date '+[%H:%M:%S]') Releasing simulator $simulator_id"
+  if response=$(simulator_manager_command DELETE "simulator/$$"); then
+    echo "$(date '+[%H:%M:%S]') ✅ Released simulator $simulator_id"
+  else
+    echo "$(date '+[%H:%M:%S]') ❌ Failed to release simulator $simulator_id: $response" >&2
+  fi
 fi
 
 profdata="$test_tmp_dir/$simulator_id/Coverage.profdata"

@@ -0,0 +1,31 @@
+load("@rules_shell//shell:sh_binary.bzl", "sh_binary")
+
+# This is a macro in our repo, so this won't compile as is. But it's basically `macos_application` and `macos_unit_test` combined.
+macos_swift_tool(
+    name = "simulator_manager",
+    deps = [
+        "@swiftpkg_shellout//:ShellOut",
+        "@swiftpkg_swift_argument_parser//:ArgumentParser",
+        "@swiftpkg_swift_nio//:NIO",
+        "@swiftpkg_swift_nio//:NIOCore",
+        "@swiftpkg_swift_nio//:NIOHTTP1",
+        "@swiftpkg_swift_nio_extras//:NIOExtras",
+    ],
+)
+
+sh_binary(
+    name = "start",
+    srcs = ["start.sh"],
+    data = [
+        ":prepare_simulator",
+        ":simulator_manager_opt",
+        ":start_tunnel",
+    ],
+    visibility = ["//visibility:public"],
+    deps = ["@bazel_tools//tools/bash/runfiles"],
+)
+
+sh_binary(
+    name = "prepare_simulator",
+    srcs = ["prepare_simulator.sh"],
+)
@@ -0,0 +1,17 @@
+# `simulator_manager`
+
+The `simulator_manager` replaces the `simulator_creator.py` used in
+**rules_apple**'s `ios_xctestrun_runner.template.sh`.
+
+It manages simulator "leases". Tests can lease and then release a simulator of a
+given configuration (device type and os version), and can also request that it's
+an "exclusive" lease. The manager automatically releases a simulator if the
+requested process exits without releasing first.
+
+Exclusive leases mean that test has exclusive access to the simulator, which is
+needed for App Host and UI tests.
+
+Base simulators are created for a given configuration, leases are on clones of
+the base simulators. After a simulator has been released for 10 minutes the
+clone is deleted. This allows us to free up disk space on remote executors, but
+also allow reuse in a short period of time.
@@ -0,0 +1,80 @@
+import NIO
+import NIOHTTP1
+import os.log
+
+extension Logger {
+  static let accumulatedHTTP = simulatorManager(category: "accumulated_http")
+}
+
+struct FullHTTPRequest {
+  let head: HTTPRequestHead
+  var body: ByteBuffer
+}
+
+struct FullHTTPResponse {
+  let head: HTTPResponseHead
+  var body: ByteBuffer
+}
+
+final class AccumulatedHTTPHandler: ChannelInboundHandler, ChannelOutboundHandler {
+  typealias InboundIn = HTTPServerRequestPart
+  typealias InboundOut = FullHTTPRequest
+
+  typealias OutboundIn = FullHTTPResponse
+  typealias OutboundOut = HTTPServerResponsePart
+
+  private var requestHead: HTTPRequestHead?
+  private var bodyBuffer: ByteBuffer?
+
+  func channelRead(context: ChannelHandlerContext, data: NIOAny) {
+    let part = self.unwrapInboundIn(data)
+
+    switch part {
+    case .head(let head):
+      self.requestHead = head
+      self.bodyBuffer = context.channel.allocator.buffer(capacity: 0)
+
+    case .body(var chunk):
+      self.bodyBuffer?.writeBuffer(&chunk)
+
+    case .end:
+      if let head = requestHead, let body = bodyBuffer {
+        Logger.accumulatedHTTP.info(
+          """
+          ▶️ Received \(head.method.rawValue, privacy: .public) request for \
+          \(head.uri, privacy: .public)
+          """
+        )
+
+        let fullRequest = FullHTTPRequest(head: head, body: body)
+        context.fireChannelRead(self.wrapInboundOut(fullRequest))
+      }
+
+      self.requestHead = nil
+      self.bodyBuffer = nil
+    }
+  }
+
+  func write(context: ChannelHandlerContext, data: NIOAny, promise: EventLoopPromise<Void>?) {
+    let fullResponse = unwrapOutboundIn(data)
+
+    Logger.accumulatedHTTP.info(
+      "◀️ Sending \(fullResponse.head.status, privacy: .public) response"
+    )
+
+    context.write(wrapOutboundOut(.head(fullResponse.head)), promise: nil)
+
+    if fullResponse.body.readableBytes > 0 {
+      context.write(wrapOutboundOut(.body(.byteBuffer(fullResponse.body))), promise: nil)
+    }
+
+    context.write(wrapOutboundOut(.end(nil)), promise: promise)
+  }
+
+  func errorCaught(context: ChannelHandlerContext, error: Error) {
+    Logger.accumulatedHTTP.error(
+      "❌ \(error.localizedDescription, privacy: .public)"
+    )
+    context.close(promise: nil)
+  }
+}