client: pass back server side error to client exit value
[fio.git] / client.c
index 713194f9284e4c86a06738a7e99e35eefbce6301..fbeac35bfb30dd46635edfab70a6adb851b81c0e 100644 (file)
--- a/client.c
+++ b/client.c
@@ -28,6 +28,7 @@ struct client_eta {
 struct fio_client {
        struct flist_head list;
        struct flist_head hash_list;
+       struct flist_head arg_list;
        struct sockaddr_in addr;
        struct sockaddr_un addr_un;
        char *hostname;
@@ -40,10 +41,15 @@ struct fio_client {
 
        int skip_newline;
        int is_sock;
+       int disk_stats_shown;
+       unsigned int jobs;
+       int error;
 
        struct flist_head eta_list;
        struct client_eta *eta_in_flight;
 
+       struct flist_head cmd_list;
+
        uint16_t argc;
        char **argv;
 };
@@ -54,13 +60,21 @@ enum {
        Client_created          = 0,
        Client_connected        = 1,
        Client_started          = 2,
-       Client_stopped          = 3,
-       Client_exited           = 4,
+       Client_running          = 3,
+       Client_stopped          = 4,
+       Client_exited           = 5,
 };
 
 static FLIST_HEAD(client_list);
 static FLIST_HEAD(eta_list);
 
+static FLIST_HEAD(arg_list);
+
+static struct thread_stat client_ts;
+static struct group_run_stats client_gs;
+static int sum_stat_clients;
+static int sum_stat_nr;
+
 #define FIO_CLIENT_HASH_BITS   7
 #define FIO_CLIENT_HASH_SZ     (1 << FIO_CLIENT_HASH_BITS)
 #define FIO_CLIENT_HASH_MASK   (FIO_CLIENT_HASH_SZ - 1)
@@ -127,6 +141,7 @@ static void remove_client(struct fio_client *client)
 
        free(client);
        nr_clients--;
+       sum_stat_clients--;
 }
 
 static void __fio_client_add_cmd_option(struct fio_client *client,
@@ -143,23 +158,49 @@ static void __fio_client_add_cmd_option(struct fio_client *client,
 void fio_client_add_cmd_option(void *cookie, const char *opt)
 {
        struct fio_client *client = cookie;
+       struct flist_head *entry;
 
        if (!client || !opt)
                return;
 
        __fio_client_add_cmd_option(client, opt);
+
+       /*
+        * Duplicate arguments to shared client group
+        */
+       flist_for_each(entry, &arg_list) {
+               client = flist_entry(entry, struct fio_client, arg_list);
+
+               __fio_client_add_cmd_option(client, opt);
+       }
 }
 
 int fio_client_add(const char *hostname, void **cookie)
 {
+       struct fio_client *existing = *cookie;
        struct fio_client *client;
 
+       if (existing) {
+               /*
+                * We always add our "exec" name as the option, hence 1
+                * means empty.
+                */
+               if (existing->argc == 1)
+                       flist_add_tail(&existing->arg_list, &arg_list);
+               else {
+                       while (!flist_empty(&arg_list))
+                               flist_del_init(arg_list.next);
+               }
+       }
+
        client = malloc(sizeof(*client));
        memset(client, 0, sizeof(*client));
 
        INIT_FLIST_HEAD(&client->list);
        INIT_FLIST_HEAD(&client->hash_list);
+       INIT_FLIST_HEAD(&client->arg_list);
        INIT_FLIST_HEAD(&client->eta_list);
+       INIT_FLIST_HEAD(&client->cmd_list);
 
        if (fio_server_parse_string(hostname, &client->hostname,
                                        &client->is_sock, &client->port,
@@ -238,6 +279,8 @@ static int fio_client_connect(struct fio_client *client)
        else
                fd = fio_client_connect_ip(client);
 
+       dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
+
        if (fd < 0)
                return 1;
 
@@ -257,7 +300,7 @@ void fio_clients_terminate(void)
        flist_for_each(entry, &client_list) {
                client = flist_entry(entry, struct fio_client, list);
 
-               fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0);
+               fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
        }
 }
 
@@ -286,8 +329,7 @@ static void probe_client(struct fio_client *client)
 {
        dprint(FD_NET, "client: send probe\n");
 
-       fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0);
-       handle_client(client);
+       fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 }
 
 static int send_client_cmd_line(struct fio_client *client)
@@ -295,17 +337,22 @@ static int send_client_cmd_line(struct fio_client *client)
        struct cmd_single_line_pdu *cslp;
        struct cmd_line_pdu *clp;
        unsigned long offset;
+       unsigned int *lens;
        void *pdu;
        size_t mem;
        int i, ret;
 
        dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 
+       lens = malloc(client->argc * sizeof(unsigned int));
+
        /*
         * Find out how much mem we need
         */
-       for (i = 0, mem = 0; i < client->argc; i++)
-               mem += strlen(client->argv[i]) + 1;
+       for (i = 0, mem = 0; i < client->argc; i++) {
+               lens[i] = strlen(client->argv[i]) + 1;
+               mem += lens[i];
+       }
 
        /*
         * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
@@ -317,7 +364,7 @@ static int send_client_cmd_line(struct fio_client *client)
        offset = sizeof(*clp);
 
        for (i = 0; i < client->argc; i++) {
-               uint16_t arg_len = strlen(client->argv[i]) + 1;
+               uint16_t arg_len = lens[i];
 
                cslp = pdu + offset;
                strcpy((char *) cslp->text, client->argv[i]);
@@ -325,6 +372,7 @@ static int send_client_cmd_line(struct fio_client *client)
                offset += sizeof(*cslp) + arg_len;
        }
 
+       free(lens);
        clp->lines = cpu_to_le16(client->argc);
        ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
        free(pdu);
@@ -533,6 +581,20 @@ static void handle_ts(struct fio_net_cmd *cmd)
        convert_gs(&p->rs, &p->rs);
 
        show_thread_status(&p->ts, &p->rs);
+
+       if (sum_stat_clients == 1)
+               return;
+
+       sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
+       sum_group_stats(&client_gs, &p->rs);
+
+       client_ts.members++;
+       client_ts.groupid = p->ts.groupid;
+
+       if (++sum_stat_nr == sum_stat_clients) {
+               strcpy(client_ts.name, "All clients");
+               show_thread_status(&client_ts, &client_gs);
+       }
 }
 
 static void handle_gs(struct fio_net_cmd *cmd)
@@ -543,6 +605,54 @@ static void handle_gs(struct fio_net_cmd *cmd)
        show_group_stats(gs);
 }
 
+static void convert_agg(struct disk_util_agg *agg)
+{
+       int i;
+
+       for (i = 0; i < 2; i++) {
+               agg->ios[i]     = le32_to_cpu(agg->ios[i]);
+               agg->merges[i]  = le32_to_cpu(agg->merges[i]);
+               agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
+               agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
+       }
+
+       agg->io_ticks           = le32_to_cpu(agg->io_ticks);
+       agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
+       agg->slavecount         = le32_to_cpu(agg->slavecount);
+       agg->max_util.u.f       = __le64_to_cpu(fio_uint64_to_double(agg->max_util.u.i));
+}
+
+static void convert_dus(struct disk_util_stat *dus)
+{
+       int i;
+
+       for (i = 0; i < 2; i++) {
+               dus->ios[i]     = le32_to_cpu(dus->ios[i]);
+               dus->merges[i]  = le32_to_cpu(dus->merges[i]);
+               dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
+               dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
+       }
+
+       dus->io_ticks           = le32_to_cpu(dus->io_ticks);
+       dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
+       dus->msec               = le64_to_cpu(dus->msec);
+}
+
+static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
+{
+       struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
+
+       convert_dus(&du->dus);
+       convert_agg(&du->agg);
+
+       if (!client->disk_stats_shown) {
+               client->disk_stats_shown = 1;
+               log_info("\nDisk stats (read/write):\n");
+       }
+
+       print_disk_util(&du->dus, &du->agg, terse_output);
+}
+
 static void convert_jobs_eta(struct jobs_eta *je)
 {
        int i;
@@ -597,13 +707,40 @@ static void dec_jobs_eta(struct client_eta *eta)
        }
 }
 
+static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
+{
+       struct fio_net_int_cmd *icmd = NULL;
+       struct flist_head *entry;
+
+       flist_for_each(entry, &client->cmd_list) {
+               icmd = flist_entry(entry, struct fio_net_int_cmd, list);
+
+               if (cmd->tag == (uintptr_t) icmd)
+                       break;
+
+               icmd = NULL;
+       }
+
+       if (!icmd) {
+               log_err("fio: client: unable to find matching tag\n");
+               return;
+       }
+
+       flist_del(&icmd->list);
+       cmd->tag = icmd->saved_tag;
+       free(icmd);
+}
+
 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 {
        struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
-       struct client_eta *eta = (struct client_eta *) cmd->tag;
+       struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 
        dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 
+       assert(client->eta_in_flight == eta);
+
+       client->eta_in_flight = NULL;
        flist_del_init(&client->eta_list);
 
        convert_jobs_eta(je);
@@ -615,6 +752,7 @@ static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 {
        struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
        const char *os, *arch;
+       char bit[16];
 
        os = fio_get_os_string(probe->os);
        if (!os)
@@ -624,14 +762,35 @@ static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
        if (!arch)
                os = "unknown";
 
-       log_info("hostname=%s, be=%u, os=%s, arch=%s, fio=%u.%u.%u\n",
-               probe->hostname, probe->bigendian, os, arch, probe->fio_major,
-               probe->fio_minor, probe->fio_patch);
+       sprintf(bit, "%d-bit", probe->bpp * 8);
+
+       log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
+               probe->hostname, probe->bigendian, bit, os, arch,
+               probe->fio_major, probe->fio_minor, probe->fio_patch);
 
        if (!client->name)
                client->name = strdup((char *) probe->hostname);
 }
 
+static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
+{
+       struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
+
+       client->state = Client_started;
+       client->jobs = le32_to_cpu(pdu->jobs);
+}
+
+static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
+{
+       struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
+
+       client->state = Client_stopped;
+       client->error = le32_to_cpu(pdu->error);
+
+       if (client->error)
+               log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
+}
+
 static int handle_client(struct fio_client *client)
 {
        struct fio_net_cmd *cmd;
@@ -642,8 +801,8 @@ static int handle_client(struct fio_client *client)
        if (!cmd)
                return 0;
 
-       dprint(FD_NET, "client: got cmd op %d from %s\n",
-                                       cmd->opcode, client->hostname);
+       dprint(FD_NET, "client: got cmd op %s from %s\n",
+                               fio_server_op(cmd->opcode), client->hostname);
 
        switch (cmd->opcode) {
        case FIO_NET_CMD_QUIT:
@@ -665,6 +824,10 @@ static int handle_client(struct fio_client *client)
                free(cmd);
                break;
                }
+       case FIO_NET_CMD_DU:
+               handle_du(client, cmd);
+               free(cmd);
+               break;
        case FIO_NET_CMD_TS:
                handle_ts(cmd);
                free(cmd);
@@ -674,23 +837,29 @@ static int handle_client(struct fio_client *client)
                free(cmd);
                break;
        case FIO_NET_CMD_ETA:
+               remove_reply_cmd(client, cmd);
                handle_eta(client, cmd);
                free(cmd);
                break;
        case FIO_NET_CMD_PROBE:
+               remove_reply_cmd(client, cmd);
                handle_probe(client, cmd);
                free(cmd);
                break;
+       case FIO_NET_CMD_RUN:
+               client->state = Client_running;
+               free(cmd);
+               break;
        case FIO_NET_CMD_START:
-               client->state = Client_started;
+               handle_start(client, cmd);
                free(cmd);
                break;
        case FIO_NET_CMD_STOP:
-               client->state = Client_stopped;
+               handle_stop(client, cmd);
                free(cmd);
                break;
        default:
-               log_err("fio: unknown client op: %d\n", cmd->opcode);
+               log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
                free(cmd);
                break;
        }
@@ -718,10 +887,14 @@ static void request_client_etas(void)
                        skipped++;
                        continue;
                }
+               if (client->state != Client_running)
+                       continue;
 
+               assert(!client->eta_in_flight);
                flist_add_tail(&client->eta_list, &eta_list);
+               client->eta_in_flight = eta;
                fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
-                                               (uint64_t) eta);
+                                       (uintptr_t) eta, &client->cmd_list);
        }
 
        while (skipped--)
@@ -730,17 +903,70 @@ static void request_client_etas(void)
        dprint(FD_NET, "client: requested eta tag %p\n", eta);
 }
 
+static int client_check_cmd_timeout(struct fio_client *client,
+                                   struct timeval *now)
+{
+       struct fio_net_int_cmd *cmd;
+       struct flist_head *entry, *tmp;
+       int ret = 0;
+
+       flist_for_each_safe(entry, tmp, &client->cmd_list) {
+               cmd = flist_entry(entry, struct fio_net_int_cmd, list);
+
+               if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
+                       continue;
+
+               log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
+                                               fio_server_op(cmd->cmd.opcode));
+               flist_del(&cmd->list);
+               free(cmd);
+               ret = 1;
+       }
+
+       return flist_empty(&client->cmd_list) && ret;
+}
+
+static int fio_client_timed_out(void)
+{
+       struct fio_client *client;
+       struct flist_head *entry, *tmp;
+       struct timeval tv;
+       int ret = 0;
+
+       gettimeofday(&tv, NULL);
+
+       flist_for_each_safe(entry, tmp, &client_list) {
+               client = flist_entry(entry, struct fio_client, list);
+
+               if (flist_empty(&client->cmd_list))
+                       continue;
+
+               if (!client_check_cmd_timeout(client, &tv))
+                       continue;
+
+               log_err("fio: client %s timed out\n", client->hostname);
+               remove_client(client);
+               ret = 1;
+       }
+
+       return ret;
+}
+
 int fio_handle_clients(void)
 {
        struct fio_client *client;
        struct flist_head *entry;
        struct pollfd *pfds;
-       int i, ret = 0;
+       int i, ret = 0, retval = 0;
 
        gettimeofday(&eta_tv, NULL);
 
        pfds = malloc(nr_clients * sizeof(struct pollfd));
 
+       sum_stat_clients = nr_clients;
+       init_thread_stat(&client_ts);
+       init_group_run_stat(&client_gs);
+
        while (!exit_backend && nr_clients) {
                i = 0;
                flist_for_each(entry, &client_list) {
@@ -760,6 +986,9 @@ int fio_handle_clients(void)
                        if (mtime_since(&eta_tv, &tv) >= 900) {
                                request_client_etas();
                                memcpy(&eta_tv, &tv, sizeof(tv));
+
+                               if (fio_client_timed_out())
+                                       break;
                        }
 
                        ret = poll(pfds, nr_clients, 100);
@@ -785,10 +1014,13 @@ int fio_handle_clients(void)
                                log_info("client: host=%s disconnected\n",
                                                client->hostname);
                                remove_client(client);
+                               retval = 1;
                        }
+                       if (client->error)
+                               retval = 1;
                }
        }
 
        free(pfds);
-       return 0;
+       return retval;
 }