[BENCH] Fix a couple bugs in stress tester

3 files changed, 70 insertions, 82 deletions

diff --git a/src/bench/connection.c b/src/bench/connection.c
index 129ae3c..2e40019 100644
--- a/src/bench/connection.c
+++ b/src/bench/connection.c
@@ -18,23 +18,10 @@ static const size_t SHORTBUF = 100;
 #define SOCKET_KEEPALIVE_TIMEOUT (3)
 #define MAX_ALTS (8)
 #define MAX_HOSTS_PER_ADDRESS (2)
-// If a server wasn't reachable this many times, we slowly start skipping it on measurements
-static const int FAIL_BACKOFF_START_COUNT = 8;
 #define RTT_COUNT (4)
 
 /* Module variables */
-
-// Init guard
-static bool connectionInitDone = false;
-static bool keepRunning = true;
-
-static struct {
-	int sockFd;
-	pthread_mutex_t sendMutex;
-	dnbd3_signal_t* panicSignal;
-	dnbd3_host_t currentServer;
-	uint64_t startupTime;
-} connection;
+static char trash[4096];
 
 // Known alt servers
 typedef struct _alt_server {
@@ -54,13 +41,13 @@ bool connection_init_n_times(
 		const char *lowerImage,
 		const uint16_t rid,
 		int ntimes,
-		BenchCounters* counters,
-		bool closeSockets
+		BenchCounters* counters
 		) {
 	for (int run_i = 0; run_i < ntimes; ++run_i) {
 		counters->attempts++;
 
-		printf(".");
+		putchar('.');
+		fflush(stdout);
 		int sock = -1;
 		char host[SHORTBUF];
 		serialized_buffer_t buffer;
@@ -68,66 +55,70 @@ bool connection_init_n_times(
 		char *remoteName;
 		uint64_t remoteSize;
 
-		if ( !connectionInitDone && keepRunning ) {
-			dnbd3_host_t tempHosts[MAX_HOSTS_PER_ADDRESS];
-			const char *current, *end;
-			int altIndex = 0;
-			memset( altservers, 0, sizeof altservers );
-			connection.sockFd = -1;
-			current = hosts;
-			do {
-				// Get next host from string
-				while ( *current == ' ' ) current++;
-				end = strchr( current, ' ' );
-				size_t len = (end == NULL ? SHORTBUF : (size_t)( end - current ) + 1);
-				if ( len > SHORTBUF ) len = SHORTBUF;
-				snprintf( host, len, "%s", current );
-				int newHosts = sock_resolveToDnbd3Host( host, tempHosts, MAX_HOSTS_PER_ADDRESS );
-				for ( int i = 0; i < newHosts; ++i ) {
-					if ( altIndex >= MAX_ALTS )
-						break;
-					altservers[altIndex].host = tempHosts[i];
-					altIndex += 1;
-				}
-				current = end + 1;
-			} while ( end != NULL && altIndex < MAX_ALTS );
-			logadd( LOG_INFO, "Got %d servers from init call", altIndex );
-			// Connect
-			for ( int i = 0; i < altIndex; ++i ) {
-				if ( altservers[i].host.type == 0 )
-					continue;
-				// Try to connect
-				sock = sock_connect( &altservers[i].host, 500, SOCKET_KEEPALIVE_TIMEOUT * 1000 );
-				if ( sock == -1 ) {
-					counters->fails++;
-					logadd( LOG_ERROR, "Could not connect to host" );
-				} else if ( !dnbd3_select_image( sock, lowerImage, rid, 0 ) ) {
-					counters->fails++;
-					logadd( LOG_ERROR, "Could not send select image" );
-				} else if ( !dnbd3_select_image_reply( &buffer, sock, &remoteVersion, &remoteName, &remoteRid, &remoteSize ) ) {
-					counters->fails++;
-					logadd( LOG_ERROR, "Could not read select image reply (%d)", errno );
-				} else if ( rid != 0 && rid != remoteRid ) {
-					counters->fails++;
-					logadd( LOG_ERROR, "rid mismatch" );
-				} else {
-					counters->success++;
+		dnbd3_host_t tempHosts[MAX_HOSTS_PER_ADDRESS];
+		const char *current, *end;
+		int altIndex = 0;
+		memset( altservers, 0, sizeof altservers );
+		current = hosts;
+		do {
+			// Get next host from string
+			while ( *current == ' ' ) current++;
+			end = strchr( current, ' ' );
+			size_t len = (end == NULL ? SHORTBUF : (size_t)( end - current ) + 1);
+			if ( len > SHORTBUF ) len = SHORTBUF;
+			snprintf( host, len, "%s", current );
+			int newHosts = sock_resolveToDnbd3Host( host, tempHosts, MAX_HOSTS_PER_ADDRESS );
+			for ( int i = 0; i < newHosts; ++i ) {
+				if ( altIndex >= MAX_ALTS )
 					break;
-				}
-				// Failed
-				logadd( LOG_DEBUG1, "Server does not offer requested image... " );
-				if ( sock != -1 ) {
-					close( sock );
-					sock = -1;
-				}
+				altservers[altIndex].host = tempHosts[i];
+				altIndex += 1;
 			}
+			current = end + 1;
+		} while ( end != NULL && altIndex < MAX_ALTS );
+		// Connect
+		for ( int i = 0; i < altIndex; ++i ) {
+			if ( altservers[i].host.type == 0 )
+				continue;
+			// Try to connect
+			dnbd3_reply_t reply;
+			sock = sock_connect( &altservers[i].host, 500, SOCKET_KEEPALIVE_TIMEOUT * 1000 );
+			if ( sock == -1 ) {
+				counters->fails++;
+				logadd( LOG_ERROR, "Could not connect to host" );
+			} else if ( !dnbd3_select_image( sock, lowerImage, rid, 0 ) ) {
+				counters->fails++;
+				logadd( LOG_ERROR, "Could not send select image" );
+			} else if ( !dnbd3_select_image_reply( &buffer, sock, &remoteVersion, &remoteName, &remoteRid, &remoteSize ) ) {
+				counters->fails++;
+				logadd( LOG_ERROR, "Could not read select image reply (%d)", errno );
+			} else if ( rid != 0 && rid != remoteRid ) {
+				counters->fails++;
+				logadd( LOG_ERROR, "rid mismatch" );
+			} else if ( !dnbd3_get_block( sock, run_i * 4096, 4096, 0, 0 ) ) {
+				counters->fails++;
+				logadd( LOG_ERROR, "send: get block failed" );
+			} else if ( !dnbd3_get_reply( sock, &reply ) ) {
+				counters->fails++;
+				logadd( LOG_ERROR, "recv: get block header failed" );
+			} else if ( recv( sock, trash, sizeof(trash), 0 ) != sizeof(trash) ) {
+				counters->fails++;
+				logadd( LOG_ERROR, "recv: get block payload failed" );
+			} else {
+				counters->success++;
+				close( sock );
+				sock = -1;
+				continue;
+			}
+			// Failed
 			if ( sock != -1 ) {
-				// connectionInitDone = true;
-				if (closeSockets) {
-					close( sock );
-				}
+				close( sock );
+				sock = -1;
 			}
 		}
+		if ( sock != -1 ) {
+			close( sock );
+		}
 	}
 	return true;
 }
diff --git a/src/bench/connection.h b/src/bench/connection.h
index 9cb59ef..ff71e15 100644
--- a/src/bench/connection.h
+++ b/src/bench/connection.h
@@ -19,7 +19,7 @@ typedef struct _dnbd3_async {
 } dnbd3_async_t;
 
 
-bool connection_init_n_times(const char *hosts, const char *image, const uint16_t rid, int ntimes, BenchCounters* counters, bool closeSockets);
+bool connection_init_n_times(const char *hosts, const char *image, const uint16_t rid, int ntimes, BenchCounters* counters);
 
 bool connection_init(const char *hosts, const char *image, const uint16_t rid);
 
diff --git a/src/bench/main.c b/src/bench/main.c
index 2f32dbf..c86af81 100644
--- a/src/bench/main.c
+++ b/src/bench/main.c
@@ -31,8 +31,6 @@ static void printUsage(char *argv0, int exitCode)
 	printf( "   -n --runs       Number of connection attempts per thread\n" );
 	printf( "   -t --threads    number of threads\n" );
 	printf( "   -l --log        Write log to given location\n" );
-	printf( "   -d --debug      Don't fork and print debug output (fuse > stderr, dnbd3 > stdout)\n" );
-	// // fuse_main( 2, arg, &dnbd3_fuse_no_operations, NULL );
 	exit( exitCode );
 }
 
@@ -41,8 +39,8 @@ static const struct option longOpts[] = {
         { "host", required_argument, NULL, 'h' },
         { "image", required_argument, NULL, 'i' },
         { "nruns", optional_argument, NULL, 'n' },
-        { "threads", optional_argument, NULL, 't' },
-        { "help", optional_argument, NULL, 'H' },
+        { "threads", required_argument, NULL, 't' },
+        { "help", required_argument, NULL, 'H' },
         { "version", no_argument, NULL, 'v' },
         { 0, 0, 0, 0 }
 };
@@ -59,11 +57,10 @@ void* runBenchThread(void* t) {
 	BenchThreadData* data = t;
 	connection_init_n_times(
 			data->server_address,
-			data->server_address,
+			data->image_name,
 			0,
 			data->runs,
-			data->counter,
-			data->closeSockets);
+			data->counter);
 	printf("Thread #%d finished\n", data->threadNumber);
 	return NULL;
 }
@@ -85,10 +82,10 @@ int main(int argc, char *argv[])
 	while ( ( opt = getopt_long( argc, argv, optString, longOpts, &lidx ) ) != -1 ) {
 		switch ( opt ) {
 		case 'h':
-			server_address = optarg;
+			server_address = strdup(optarg);
 			break;
 		case 'i':
-			image_Name = optarg;
+			image_Name = strdup(optarg);
 			break;
 		case 'n':
 			n_runs = atoi(optarg);