[ML] Make logger test more robust to slow VMs (elastic#101)

The non-ASCII logging test could hang indefinitely if the VM it
was running on stalled for 200ms before reconfiguring the logger
to use the test pipe.

This change makes it more robust.  A stall of up to 5 seconds is
now tolerated, but with checks every 50ms to avoid slowing the
test down.

Author: David Roberts

lib/core/unittest/CLoggerTest.cc

@@ -150,11 +150,18 @@ void CLoggerTest::testNonAsciiJsonLogging() {
 
     std::ostringstream loggedData;
     std::thread reader([&loggedData] {
-        // wait a bit so that pipe has been created
-        ml::core::CSleep::sleep(200);
-        std::ifstream strm(TEST_PIPE_NAME);
-        std::copy(std::istreambuf_iterator<char>(strm), std::istreambuf_iterator<char>(),
-                  std::ostreambuf_iterator<char>(loggedData));
+        for (std::size_t attempt = 1; attempt <= 100; ++attempt) {
+            // wait a bit so that pipe has been created
+            ml::core::CSleep::sleep(50);
+            std::ifstream strm(TEST_PIPE_NAME);
+            if (strm.is_open()) {
+                std::copy(std::istreambuf_iterator<char>(strm),
+                          std::istreambuf_iterator<char>(),
+                          std::ostreambuf_iterator<char>(loggedData));
+                return;
+            }
+        }
+        CPPUNIT_FAIL("Failed to connect to logging pipe within a reasonable time");
     });
 
     ml::core::CLogger& logger = ml::core::CLogger::instance();