[Telemetry] fix run_benchmark crash on CrOS

telemetry/telemetry/internal/backends/chrome/chrome_browser_backend.py

Index: telemetry/telemetry/internal/backends/chrome/chrome_browser_backend.py
diff --git a/telemetry/telemetry/internal/backends/chrome/chrome_browser_backend.py b/telemetry/telemetry/internal/backends/chrome/chrome_browser_backend.py
index f5b02fb12b6163871e6036e532abd7b115dd8024..908239d798e35054745a91fcab9a802e15298df9 100644
--- a/telemetry/telemetry/internal/backends/chrome/chrome_browser_backend.py
+++ b/telemetry/telemetry/internal/backends/chrome/chrome_browser_backend.py
@@ -146,6 +146,12 @@ class ChromeBrowserBackend(browser_backend.BrowserBackend):
       remote_devtools_port: The remote devtools port, if
           any. Otherwise assumed to be the same as self._port.
     """
+    if self._devtools_client:
+      # In case we are launching a second browser instance (as is done by
+      # the CrOS backend), ensure that the old devtools_client is closed,
+      # otherwise re-creating it will fail.
+      self._devtools_client.Close()
+      self._devtools_client = None
     try:
       timeout = self.browser_options.browser_startup_timeout
       py_utils.WaitFor(self.HasBrowserFinishedLaunching, timeout=timeout)