client: dec sum_stat_clients if one a client is disconnected
[fio.git] / client.c
index 874f0aecd3b80ccd1b8bed646be783410a4e99e0..31b4862775eb159775d2fa784a28a9032cc600f6 100644 (file)
--- a/client.c
+++ b/client.c
 #include "flist.h"
 #include "hash.h"
 
+struct client_eta {
+       struct jobs_eta eta;
+       unsigned int pending;
+};
+
 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;
@@ -32,13 +38,21 @@ struct fio_client {
        char *name;
 
        int state;
+
        int skip_newline;
        int is_sock;
 
+       struct flist_head eta_list;
+       struct client_eta *eta_in_flight;
+
+       struct flist_head cmd_list;
+
        uint16_t argc;
        char **argv;
 };
 
+static struct timeval eta_tv;
+
 enum {
        Client_created          = 0,
        Client_connected        = 1,
@@ -48,6 +62,14 @@ enum {
 };
 
 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)
@@ -55,6 +77,7 @@ static FLIST_HEAD(client_list);
 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
 
 static int handle_client(struct fio_client *client);
+static void dec_jobs_eta(struct client_eta *eta);
 
 static void fio_client_add_hash(struct fio_client *client)
 {
@@ -101,6 +124,11 @@ static void remove_client(struct fio_client *client)
 
        fio_client_remove_hash(client);
 
+       if (!flist_empty(&client->eta_list)) {
+               flist_del_init(&client->eta_list);
+               dec_jobs_eta(client->eta_in_flight);
+       }
+
        free(client->hostname);
        if (client->argv)
                free(client->argv);
@@ -109,6 +137,7 @@ 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,
@@ -125,22 +154,49 @@ 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,
@@ -219,6 +275,8 @@ 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;
 
@@ -238,7 +296,7 @@ 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);
        }
 }
 
@@ -267,8 +325,7 @@ 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)
@@ -276,17 +333,22 @@ 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
@@ -298,7 +360,7 @@ 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]);
@@ -306,8 +368,9 @@ 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);
+       ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
        free(pdu);
        return ret;
 }
@@ -390,7 +453,7 @@ static int fio_client_send_ini(struct fio_client *client, const char *filename)
                return 1;
        }
 
-       ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size);
+       ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size, 0);
        free(buf);
        close(fd);
        return ret;
@@ -514,6 +577,20 @@ 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)
@@ -524,9 +601,8 @@ static void handle_gs(struct fio_net_cmd *cmd)
        show_group_stats(gs);
 }
 
-static void handle_eta(struct fio_net_cmd *cmd)
+static void convert_jobs_eta(struct jobs_eta *je)
 {
-       struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
        int i;
 
        je->nr_running          = le32_to_cpu(je->nr_running);
@@ -543,10 +619,81 @@ static void handle_eta(struct fio_net_cmd *cmd)
                je->iops[i]     = le32_to_cpu(je->iops[i]);
        }
 
-       je->elapsed_sec         = le32_to_cpu(je->nr_running);
+       je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
        je->eta_sec             = le64_to_cpu(je->eta_sec);
+}
+
+static void sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
+{
+       int i;
+
+       dst->nr_running         += je->nr_running;
+       dst->nr_ramp            += je->nr_ramp;
+       dst->nr_pending         += je->nr_pending;
+       dst->files_open         += je->files_open;
+       dst->m_rate             += je->m_rate;
+       dst->t_rate             += je->t_rate;
+       dst->m_iops             += je->m_iops;
+       dst->t_iops             += je->t_iops;
+
+       for (i = 0; i < 2; i++) {
+               dst->rate[i]    += je->rate[i];
+               dst->iops[i]    += je->iops[i];
+       }
 
-       display_thread_status(je);
+       dst->elapsed_sec        += je->elapsed_sec;
+
+       if (je->eta_sec > dst->eta_sec)
+               dst->eta_sec = je->eta_sec;
+}
+
+static void dec_jobs_eta(struct client_eta *eta)
+{
+       if (!--eta->pending) {
+               display_thread_status(&eta->eta);
+               free(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 == (uint64_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;
+
+       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);
+       sum_jobs_eta(&eta->eta, je);
+       dec_jobs_eta(eta);
 }
 
 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
@@ -580,8 +727,8 @@ 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:
@@ -612,10 +759,12 @@ static int handle_client(struct fio_client *client)
                free(cmd);
                break;
        case FIO_NET_CMD_ETA:
-               handle_eta(cmd);
+               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;
@@ -628,7 +777,7 @@ static int handle_client(struct fio_client *client)
                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;
        }
@@ -636,6 +785,89 @@ static int handle_client(struct fio_client *client)
        return 1;
 }
 
+static void request_client_etas(void)
+{
+       struct fio_client *client;
+       struct flist_head *entry;
+       struct client_eta *eta;
+       int skipped = 0;
+
+       dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
+
+       eta = malloc(sizeof(*eta));
+       memset(&eta->eta, 0, sizeof(eta->eta));
+       eta->pending = nr_clients;
+
+       flist_for_each(entry, &client_list) {
+               client = flist_entry(entry, struct fio_client, list);
+
+               if (!flist_empty(&client->eta_list)) {
+                       skipped++;
+                       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, &client->cmd_list);
+       }
+
+       while (skipped--)
+               dec_jobs_eta(eta);
+
+       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;
@@ -643,8 +875,14 @@ int fio_handle_clients(void)
        struct pollfd *pfds;
        int i, ret = 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) {
@@ -658,6 +896,17 @@ int fio_handle_clients(void)
                assert(i == nr_clients);
 
                do {
+                       struct timeval tv;
+
+                       gettimeofday(&tv, NULL);
+                       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);
                        if (ret < 0) {
                                if (errno == EINTR)