Merge pull request #3635 from ctiller/be-useful

Add a more useful log message when failing to connect

src/core/iomgr/tcp_client_posix.c

@@ -141,7 +141,8 @@ static void on_writable(grpc_exec_ctx *exec_ctx, void *acp, int success) {
       err = getsockopt(fd->fd, SOL_SOCKET, SO_ERROR, &so_error, &so_error_size);
     } while (err < 0 && errno == EINTR);
     if (err < 0) {
-      gpr_log(GPR_ERROR, "getsockopt(ERROR): %s", strerror(errno));
+      gpr_log(GPR_ERROR, "failed to connect to '%s': getsockopt(ERROR): %s",
+              ac->addr_str, strerror(errno));
       goto finish;
     } else if (so_error != 0) {
       if (so_error == ENOBUFS) {
@@ -166,10 +167,14 @@ static void on_writable(grpc_exec_ctx *exec_ctx, void *acp, int success) {
       } else {
         switch (so_error) {
           case ECONNREFUSED:
-            gpr_log(GPR_ERROR, "socket error: connection refused");
+            gpr_log(
+                GPR_ERROR,
+                "failed to connect to '%s': socket error: connection refused",
+                ac->addr_str);
             break;
           default:
-            gpr_log(GPR_ERROR, "socket error: %d", so_error);
+            gpr_log(GPR_ERROR, "failed to connect to '%s': socket error: %d",
+                    ac->addr_str, so_error);
             break;
         }
         goto finish;
@@ -181,7 +186,8 @@ static void on_writable(grpc_exec_ctx *exec_ctx, void *acp, int success) {
       goto finish;
     }
   } else {
-    gpr_log(GPR_ERROR, "on_writable failed during connect");
+    gpr_log(GPR_ERROR, "failed to connect to '%s': timeout occurred",
+            ac->addr_str);
     goto finish;
   }
 

test/core/util/port_posix.c

@@ -213,15 +213,25 @@ static void got_port_from_server(grpc_exec_ctx *exec_ctx, void *arg,
   size_t i;
   int port = 0;
   portreq *pr = arg;
+  int failed = 0;
 
-  if (!response || response->status != 200) {
+  if (!response) {
+    failed = 1;
+    gpr_log(GPR_DEBUG,
+            "failed port pick from server: retrying [response=NULL]");
+  } else if (response->status != 200) {
+    failed = 1;
+    gpr_log(GPR_DEBUG, "failed port pick from server: status=%d",
+            response->status);
+  }
+
+  if (failed) {
     grpc_httpcli_request req;
     memset(&req, 0, sizeof(req));
     GPR_ASSERT(pr->retries < 10);
     pr->retries++;
     req.host = pr->server;
     req.path = "/get";
-    gpr_log(GPR_DEBUG, "failed port pick from server: retrying");
     sleep(1);
     grpc_httpcli_get(exec_ctx, pr->ctx, &pr->pollset, &req,
                      GRPC_TIMEOUT_SECONDS_TO_DEADLINE(10), got_port_from_server,

tools/run_tests/run_tests.py

@@ -715,17 +715,21 @@ def _start_port_server(port_server_port):
   try:
     version = urllib2.urlopen('http://localhost:%d/version' % port_server_port,
                               timeout=1).read()
+    print 'detected port server running'
     running = True
   except Exception:
+    print 'failed to detect port server: %s' % sys.exc_info()[0]
     running = False
   if running:
     with open('tools/run_tests/port_server.py') as f:
       current_version = hashlib.sha1(f.read()).hexdigest()
       running = (version == current_version)
       if not running:
+        print 'port_server version mismatch: killing the old one'
         urllib2.urlopen('http://localhost:%d/quit' % port_server_port).read()
         time.sleep(1)
   if not running:
+    print 'starting port_server'
     port_log = open('portlog.txt', 'w')
     port_server = subprocess.Popen(
         ['python2.7', 'tools/run_tests/port_server.py', '-p', '%d' % port_server_port],
@@ -737,7 +741,7 @@ def _start_port_server(port_server_port):
       if waits > 10:
         port_server.kill()
       if port_server.poll() is not None:
-        print "port_server failed to start"
+        print 'port_server failed to start'
         port_log = open('portlog.txt', 'r').read()
         print port_log
         sys.exit(1)
@@ -746,11 +750,11 @@ def _start_port_server(port_server_port):
                         timeout=1).read()
         break
       except socket.timeout:
-        print "waiting for port_server"
+        print 'waiting for port_server: timeout'
         time.sleep(0.5)
         waits += 1
       except urllib2.URLError:
-        print "waiting for port_server"
+        print 'waiting for port_server: urlerror'
         time.sleep(0.5)
         waits += 1
       except: