client/server: pass back job options to client
[fio.git] / client.c
index 2c10c037602353da6e52541d0eed6874b0921176..366db71615306a98b6303b2f468682444b71f815 100644 (file)
--- a/client.c
+++ b/client.c
@@ -59,9 +59,12 @@ int sum_stat_clients;
 
 static int sum_stat_nr;
 static struct json_object *root = NULL;
+static struct json_object *job_opt_object = NULL;
 static struct json_array *clients_array = NULL;
 static struct json_array *du_array = NULL;
 
+static int error_clients;
+
 #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)
@@ -115,10 +118,23 @@ static int read_data(int fd, void *data, size_t size)
 
 static void fio_client_json_init(void)
 {
-       if (output_format != FIO_OUTPUT_JSON)
+       char time_buf[32];
+       time_t time_p;
+
+       if (!(output_format & FIO_OUTPUT_JSON))
                return;
+
+       time(&time_p);
+       os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf));
+       time_buf[strlen(time_buf) - 1] = '\0';
+
        root = json_create_object();
        json_object_add_value_string(root, "fio version", fio_version_string);
+       json_object_add_value_int(root, "timestamp", time_p);
+       json_object_add_value_string(root, "time", time_buf);
+
+       job_opt_object = json_create_object();
+       json_object_add_value_object(root, "global options", job_opt_object);
        clients_array = json_create_array();
        json_object_add_value_array(root, "client_stats", clients_array);
        du_array = json_create_array();
@@ -127,9 +143,11 @@ static void fio_client_json_init(void)
 
 static void fio_client_json_fini(void)
 {
-       if (output_format != FIO_OUTPUT_JSON)
+       if (!(output_format & FIO_OUTPUT_JSON))
                return;
-       json_print_object(root);
+
+       log_info("\n");
+       json_print_object(root, NULL);
        log_info("\n");
        json_free_object(root);
        root = NULL;
@@ -172,13 +190,29 @@ void fio_put_client(struct fio_client *client)
        }
        if (client->files)
                free(client->files);
+       if (client->opt_lists)
+               free(client->opt_lists);
 
        if (!client->did_stat)
                sum_stat_clients--;
 
+       if (client->error)
+               error_clients++;
+
        free(client);
 }
 
+static int fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
+{
+       if (!--eta->pending) {
+               eta_fn(&eta->eta);
+               free(eta);
+               return 0;
+       }
+
+       return 1;
+}
+
 static void remove_client(struct fio_client *client)
 {
        assert(client->refs);
@@ -380,6 +414,7 @@ static const char *server_name(struct fio_client *client, char *buf,
 static void probe_client(struct fio_client *client)
 {
        struct cmd_client_probe_pdu pdu;
+       const char *sname;
        uint64_t tag;
        char buf[64];
 
@@ -391,7 +426,9 @@ static void probe_client(struct fio_client *client)
        pdu.flags = 0;
 #endif
 
-       strcpy((char *) pdu.server, server_name(client, buf, sizeof(buf)));
+       sname = server_name(client, buf, sizeof(buf));
+       memset(pdu.server, 0, sizeof(pdu.server));
+       strncpy((char *) pdu.server, sname, sizeof(pdu.server) - 1);
 
        fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
 }
@@ -515,11 +552,6 @@ static void sig_int(int sig)
        fio_clients_terminate();
 }
 
-static void sig_show_status(int sig)
-{
-       show_running_run_stats();
-}
-
 static void client_signal_handler(void)
 {
        struct sigaction act;
@@ -883,6 +915,10 @@ static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
        dst->latency_target     = le64_to_cpu(src->latency_target);
        dst->latency_window     = le64_to_cpu(src->latency_window);
        dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
+
+       dst->nr_block_infos     = le64_to_cpu(src->nr_block_infos);
+       for (i = 0; i < dst->nr_block_infos; i++)
+               dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
 }
 
 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
@@ -916,9 +952,13 @@ static void json_object_add_client_info(struct json_object *obj,
 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 {
        struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
+       struct flist_head *opt_list = NULL;
        struct json_object *tsobj;
 
-       tsobj = show_thread_status(&p->ts, &p->rs);
+       if (client->opt_lists && p->ts.thread_number <= client->jobs)
+               opt_list = &client->opt_lists[p->ts.thread_number - 1];
+
+       tsobj = show_thread_status(&p->ts, &p->rs, opt_list, NULL);
        client->did_stat = 1;
        if (tsobj) {
                json_object_add_client_info(tsobj, client);
@@ -928,7 +968,7 @@ static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
        if (sum_stat_clients <= 1)
                return;
 
-       sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
+       sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
        sum_group_stats(&client_gs, &p->rs);
 
        client_ts.members++;
@@ -938,7 +978,7 @@ static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 
        if (++sum_stat_nr == sum_stat_clients) {
                strcpy(client_ts.name, "All clients");
-               tsobj = show_thread_status(&client_ts, &client_gs);
+               tsobj = show_thread_status(&client_ts, &client_gs, NULL, NULL);
                if (tsobj) {
                        json_object_add_client_info(tsobj, client);
                        json_array_add_value_object(clients_array, tsobj);
@@ -950,7 +990,37 @@ static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
 {
        struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 
-       show_group_stats(gs);
+       if (output_format & FIO_OUTPUT_NORMAL)
+               show_group_stats(gs, NULL);
+}
+
+static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
+{
+       struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
+       struct print_option *p;
+
+       pdu->global = le16_to_cpu(pdu->global);
+       pdu->groupid = le16_to_cpu(pdu->groupid);
+
+       p = malloc(sizeof(*p));
+       p->name = strdup((char *) pdu->name);
+       if (pdu->value[0] != '\0')
+               p->value = strdup((char *) pdu->value);
+       else
+               p->value = NULL;
+
+       if (pdu->global) {
+               const char *pos = "";
+
+               if (p->value)
+                       pos = p->value;
+
+               json_object_add_value_string(job_opt_object, p->name, pos);
+       } else if (client->opt_lists) {
+               struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
+
+               flist_add_tail(&p->list, opt_list);
+       }
 }
 
 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
@@ -974,14 +1044,14 @@ 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->ios[i]     = le64_to_cpu(agg->ios[i]);
+               agg->merges[i]  = le64_to_cpu(agg->merges[i]);
                agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
-               agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
+               agg->ticks[i]   = le64_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->io_ticks           = le64_to_cpu(agg->io_ticks);
+       agg->time_in_queue      = le64_to_cpu(agg->time_in_queue);
        agg->slavecount         = le32_to_cpu(agg->slavecount);
        agg->max_util.u.f       = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
 }
@@ -991,14 +1061,14 @@ static void convert_dus(struct disk_util_stat *dus)
        int i;
 
        for (i = 0; i < 2; i++) {
-               dus->s.ios[i]           = le32_to_cpu(dus->s.ios[i]);
-               dus->s.merges[i]        = le32_to_cpu(dus->s.merges[i]);
+               dus->s.ios[i]           = le64_to_cpu(dus->s.ios[i]);
+               dus->s.merges[i]        = le64_to_cpu(dus->s.merges[i]);
                dus->s.sectors[i]       = le64_to_cpu(dus->s.sectors[i]);
-               dus->s.ticks[i]         = le32_to_cpu(dus->s.ticks[i]);
+               dus->s.ticks[i]         = le64_to_cpu(dus->s.ticks[i]);
        }
 
-       dus->s.io_ticks         = le32_to_cpu(dus->s.io_ticks);
-       dus->s.time_in_queue    = le32_to_cpu(dus->s.time_in_queue);
+       dus->s.io_ticks         = le64_to_cpu(dus->s.io_ticks);
+       dus->s.time_in_queue    = le64_to_cpu(dus->s.time_in_queue);
        dus->s.msec             = le64_to_cpu(dus->s.msec);
 }
 
@@ -1011,13 +1081,16 @@ static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
                log_info("\nDisk stats (read/write):\n");
        }
 
-       if (output_format == FIO_OUTPUT_JSON) {
+       if (output_format & FIO_OUTPUT_JSON) {
                struct json_object *duobj;
                json_array_add_disk_util(&du->dus, &du->agg, du_array);
                duobj = json_array_last_value_object(du_array);
                json_object_add_client_info(duobj, client);
-       } else
-               print_disk_util(&du->dus, &du->agg, output_format == FIO_OUTPUT_TERSE);
+       }
+       if (output_format & FIO_OUTPUT_TERSE)
+               print_disk_util(&du->dus, &du->agg, 1, NULL);
+       if (output_format & FIO_OUTPUT_NORMAL)
+               print_disk_util(&du->dus, &du->agg, 0, NULL);
 }
 
 static void convert_jobs_eta(struct jobs_eta *je)
@@ -1079,14 +1152,6 @@ void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
        strcpy((char *) dst->run_str, (char *) je->run_str);
 }
 
-void fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
-{
-       if (!--eta->pending) {
-               eta_fn(&eta->eta);
-               free(eta);
-       }
-}
-
 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 {
        struct fio_net_cmd_reply *reply = NULL;
@@ -1146,6 +1211,7 @@ static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 
        client->eta_in_flight = NULL;
        flist_del_init(&client->eta_list);
+       client->eta_timeouts = 0;
 
        if (client->ops->jobs_eta)
                client->ops->jobs_eta(client, je);
@@ -1187,6 +1253,17 @@ static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
        client->jobs = le32_to_cpu(pdu->jobs);
        client->nr_stat = le32_to_cpu(pdu->stat_outputs);
 
+       if (client->jobs) {
+               int i;
+
+               if (client->opt_lists)
+                       free(client->opt_lists);
+
+               client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
+               for (i = 0; i < client->jobs; i++)
+                       INIT_FLIST_HEAD(&client->opt_lists[i]);
+       }
+
        sum_stat_clients += client->nr_stat;
 }
 
@@ -1482,9 +1559,15 @@ int fio_handle_client(struct fio_client *client)
                break;
        case FIO_NET_CMD_VTRIGGER: {
                struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
-               char buf[64];
+               char buf[128];
+               int off = 0;
 
-               __verify_save_state(pdu, server_name(client, buf, sizeof(buf)));
+               if (aux_path) {
+                       strcpy(buf, aux_path);
+                       off = strlen(buf);
+               }
+
+               __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
                exec_trigger(trigger_cmd);
                break;
                }
@@ -1493,6 +1576,10 @@ int fio_handle_client(struct fio_client *client)
                send_file(client, pdu, cmd->tag);
                break;
                }
+       case FIO_NET_CMD_JOB_OPT: {
+               handle_job_opt(client, cmd);
+               break;
+       }
        default:
                log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
                break;
@@ -1561,12 +1648,42 @@ static void request_client_etas(struct client_ops *ops)
                                        (uintptr_t) eta, &client->cmd_list);
        }
 
-       while (skipped--)
-               fio_client_dec_jobs_eta(eta, ops->eta);
+       while (skipped--) {
+               if (!fio_client_dec_jobs_eta(eta, ops->eta))
+                       break;
+       }
 
        dprint(FD_NET, "client: requested eta tag %p\n", eta);
 }
 
+/*
+ * A single SEND_ETA timeout isn't fatal. Attempt to recover.
+ */
+static int handle_cmd_timeout(struct fio_client *client,
+                             struct fio_net_cmd_reply *reply)
+{
+       if (reply->opcode != FIO_NET_CMD_SEND_ETA)
+               return 1;
+
+       log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
+       flist_del(&reply->list);
+       free(reply);
+
+       flist_del_init(&client->eta_list);
+       if (client->eta_in_flight) {
+               fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
+               client->eta_in_flight = NULL;
+       }
+
+       /*
+        * If we fail 5 in a row, give up...
+        */
+       if (client->eta_timeouts++ > 5)
+               return 1;
+
+       return 0;
+}
+
 static int client_check_cmd_timeout(struct fio_client *client,
                                    struct timeval *now)
 {
@@ -1580,6 +1697,9 @@ static int client_check_cmd_timeout(struct fio_client *client,
                if (mtime_since(&reply->tv, now) < FIO_NET_CLIENT_TIMEOUT)
                        continue;
 
+               if (!handle_cmd_timeout(client, reply))
+                       continue;
+
                log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
                                                fio_server_op(reply->opcode));
                flist_del(&reply->list);
@@ -1613,6 +1733,7 @@ static int fio_check_clients_timed_out(void)
                else
                        log_err("fio: client %s timed out\n", client->hostname);
 
+               client->error = ETIMEDOUT;
                remove_client(client);
                ret = 1;
        }
@@ -1706,5 +1827,5 @@ int fio_handle_clients(struct client_ops *ops)
        fio_client_json_fini();
 
        free(pfds);
-       return retval;
+       return retval || error_clients;
 }