distributed/protocol.c

   1 /* The functions implementing the master-slave protocol of the
   2  * distributed engine are grouped here. They are independent
   3  * of the gtp protocol. See the comments at the top of distributed.c
   4  * for a general introduction to the distributed engine. */
   5
   6 /* The receive queue is an array of pointers to binary buffers.
   7  * These pointers are invalidated in one of two ways when a buffer
   8  * is recycled: (1) the queue age is increased when the queue is
   9  * emptied at a new move, (2) the pointer itself is set to NULL
  10  * immmediately, and stays so until at least the next queue age
  11  * increment. */
  12
  13 #include <assert.h>
  14 #include <stdio.h>
  15 #include <pthread.h>
  16 #include <ctype.h>
  17
  18 #define DEBUG
  19
  20 #include "random.h"
  21 #include "timeinfo.h"
  22 #include "playout.h"
  23 #include "network.h"
  24 #include "debug.h"
  25 #include "distributed/distributed.h"
  26 #include "distributed/protocol.h"
  27
  28 /* All gtp commands for current game separated by \n */
  29 static char gtp_cmds[CMDS_SIZE];
  30
  31 /* Latest gtp command sent to slaves. */
  32 static char *gtp_cmd = NULL;
  33
  34 /* Slaves send gtp_cmd when cmd_count changes. */
  35 static int cmd_count = 0;
  36
  37 /* Remember at most 10 gtp ids per move: kgs-rules, boardsize, clear_board,
  38  * time_settings, komi, handicap, genmoves, play pass, play pass, final_status_list */
  39 #define MAX_CMDS_PER_MOVE 10
  40
  41 /* History of gtp commands sent for current game, indexed by move. */
  42 static struct cmd_history {
  43         int gtp_id;
  44         char *next_cmd;
  45 } history[MAX_GAMELEN][MAX_CMDS_PER_MOVE];
  46
  47 /* Number of active slave machines working for this master. */
  48 int active_slaves = 0;
  49
  50 /* Number of replies to last gtp command already received. */
  51 int reply_count = 0;
  52
  53 /* All replies to latest gtp command are in gtp_replies[0..reply_count-1]. */
  54 char **gtp_replies;
  55
  56
  57 struct buf_state **receive_queue;
  58 int queue_length = 0;
  59 int queue_age = 0;
  60 static int queue_max_length;
  61
  62 /* Mutex protecting all variables above. receive_queue may be
  63  * read without the lock but is only written with lock held. */
  64 static pthread_mutex_t slave_lock = PTHREAD_MUTEX_INITIALIZER;
  65
  66 /* Condition signaled when a new gtp command is available. */
  67 static pthread_cond_t cmd_cond = PTHREAD_COND_INITIALIZER;
  68
  69 /* Condition signaled when reply_count increases. */
  70 static pthread_cond_t reply_cond = PTHREAD_COND_INITIALIZER;
  71
  72 /* Mutex protecting stderr. Must not be held at same time as slave_lock. */
  73 static pthread_mutex_t log_lock = PTHREAD_MUTEX_INITIALIZER;
  74
  75 /* Absolute time when this program was started.
  76  * For debugging only. */
  77 static double start_time;
  78
  79 /* Default slave state. */
  80 struct slave_state default_sstate;
  81
  82
  83 /* Get exclusive access to the threads and commands state. */
  84 void
  85 protocol_lock(void)
  86 {
  87         pthread_mutex_lock(&slave_lock);
  88 }
  89
  90 /* Release exclusive access to the threads and commands state. */
  91 void
  92 protocol_unlock(void)
  93 {
  94         pthread_mutex_unlock(&slave_lock);
  95 }
  96
  97 /* Write the time, client address, prefix, and string s to stderr atomically.
  98  * s should end with a \n */
  99 void
 100 logline(struct in_addr *client, char *prefix, char *s)
 101 {
 102         double now = time_now();
 103
 104         char addr[INET_ADDRSTRLEN];
 105         if (client) {
 106                 inet_ntop(AF_INET, client, addr, sizeof(addr));
 107         } else {
 108                 addr[0] = '\0';
 109         }
 110         pthread_mutex_lock(&log_lock);
 111         fprintf(stderr, "%s%15s %9.3f: %s", prefix, addr, now - start_time, s);
 112         pthread_mutex_unlock(&log_lock);
 113 }
 114
 115 /* Thread opening a connection on the given socket and copying input
 116  * from there to stderr. */
 117 static void *
 118 proxy_thread(void *arg)
 119 {
 120         int proxy_sock = (long)arg;
 121         assert(proxy_sock >= 0);
 122         for (;;) {
 123                 struct in_addr client;
 124                 int conn = open_server_connection(proxy_sock, &client);
 125                 FILE *f = fdopen(conn, "r");
 126                 char buf[BSIZE];
 127                 while (fgets(buf, BSIZE, f)) {
 128                         logline(&client, "< ", buf);
 129                 }
 130                 fclose(f);
 131         }
 132 }
 133
 134 /* Get a reply to one gtp command. Return the gtp command id,
 135  * or -1 if error. reply must have at least CMDS_SIZE bytes.
 136  * The ascii reply ends with an empty line; if the first line
 137  * contains "@size", a binary reply of size bytes follows the
 138  * empty line. @size is not standard gtp, it is only used
 139  * internally by Pachi for the genmoves command; it must be the
 140  * last parameter on the line.
 141  * *bin_size is the maximum size upon entry, actual size on return.
 142  * slave_lock is not held on either entry or exit of this function. */
 143 static int
 144 get_reply(FILE *f, struct in_addr client, char *reply, void *bin_reply, int *bin_size)
 145 {
 146         int reply_id = -1;
 147         *reply = '\0';
 148         if (!fgets(reply, CMDS_SIZE, f)) return -1;
 149
 150         /* Check for binary reply. */
 151         char *s = strchr(reply, '@');
 152         int size = 0;
 153         if (s) size = atoi(s+1);
 154         assert(size <= *bin_size);
 155         *bin_size = size;
 156
 157         if (DEBUGV(s, 2))
 158                 logline(&client, "<<", reply);
 159         if ((*reply == '=' || *reply == '?') && isdigit(reply[1]))
 160                 reply_id = atoi(reply+1);
 161
 162         /* Read the rest of the ascii reply */
 163         char *line = reply + strlen(reply);
 164         while (fgets(line, reply + CMDS_SIZE - line, f) && *line != '\n') {
 165                 if (DEBUGL(3))
 166                         logline(&client, "<<", line);
 167                 line += strlen(line);
 168         }
 169         if (*line != '\n') return -1;
 170
 171         /* Read the binary reply if any. */
 172         double start = time_now();
 173         int len;
 174         while (size && (len = fread(bin_reply, 1, size, f)) > 0) {
 175                 bin_reply = (char *)bin_reply + len;
 176                 size -= len;
 177         }
 178         if (*bin_size && DEBUGVV(2)) {
 179                 char buf[1024];
 180                 snprintf(buf, sizeof(buf), "read reply %d bytes in %.4fms\n", *bin_size,
 181                          (time_now() - start)*1000);
 182                 logline(&client, "= ", buf);
 183         }
 184         return size ? -1 : reply_id;
 185 }
 186
 187 /* Send the gtp command to_send and get a reply from the slave machine.
 188  * Write the reply in buf which must have at least CMDS_SIZE bytes.
 189  * If *bin_size > 0, send bin_buf after the gtp command.
 190  * Return any binary reply in bin_buf and set its size in bin_size.
 191  * bin_buf is private to the slave and need not be copied.
 192  * Return the gtp command id, or -1 if error.
 193  * slave_lock is held on both entry and exit of this function. */
 194 static int
 195 send_command(char *to_send, void *bin_buf, int *bin_size,
 196              FILE *f, struct slave_state *sstate, char *buf)
 197 {
 198         assert(to_send && gtp_cmd && bin_buf && bin_size);
 199         strncpy(buf, to_send, CMDS_SIZE);
 200         bool resend = to_send != gtp_cmd;
 201
 202         pthread_mutex_unlock(&slave_lock);
 203
 204         if (DEBUGL(1) && resend)
 205                 logline(&sstate->client, "? ",
 206                         to_send == gtp_cmds ? "resend all\n" : "partial resend\n");
 207         fputs(buf, f);
 208
 209         double start = time_now();
 210         if (*bin_size)
 211                 fwrite(bin_buf, 1, *bin_size, f);
 212         fflush(f);
 213
 214         if (DEBUGV(strchr(buf, '@'), 2)) {
 215                 double ms = (time_now() - start) * 1000.0;
 216                 if (!DEBUGL(3)) {
 217                         char *s = strchr(buf, '\n');
 218                         if (s) s[1] = '\0';
 219                 }
 220                 logline(&sstate->client, ">>", buf);
 221                 if (*bin_size) {
 222                         char b[1024];
 223                         snprintf(b, sizeof(b),
 224                                  "sent args %d bytes in %.4fms\n", *bin_size, ms);
 225                         logline(&sstate->client, "= ", b);
 226                 }
 227         }
 228
 229         /* Reuse the buffers for the reply. */
 230         *bin_size = sstate->max_buf_size;
 231         int reply_id = get_reply(f, sstate->client, buf, bin_buf, bin_size);
 232
 233         pthread_mutex_lock(&slave_lock);
 234         return reply_id;
 235 }
 236
 237 /* Return the command sent after that with the given gtp id,
 238  * or gtp_cmds if the id wasn't used in this game. If a play command
 239  * has overwritten a genmoves command, return the play command.
 240  * slave_lock is held on both entry and exit of this function. */
 241 static char *
 242 next_command(int cmd_id)
 243 {
 244         if (cmd_id == -1) return gtp_cmds;
 245
 246         int last_id = atoi(gtp_cmd);
 247         int reply_move = move_number(cmd_id);
 248         if (reply_move > move_number(last_id)) return gtp_cmds;
 249
 250         int slot;
 251         for (slot = 0; slot < MAX_CMDS_PER_MOVE; slot++) {
 252                 if (cmd_id == history[reply_move][slot].gtp_id) break;
 253         }
 254         if (slot == MAX_CMDS_PER_MOVE) return gtp_cmds;
 255
 256         char *next = history[reply_move][slot].next_cmd;
 257         assert(next);
 258         return next;
 259 }
 260
 261 /* Allocate buffers for a slave thread. The state should have been
 262  * initialized already as a copy of the default slave state.
 263  * slave_lock is not held on either entry or exit of this function. */
 264 static void
 265 slave_state_alloc(struct slave_state *sstate)
 266 {
 267         for (int n = 0; n < BUFFERS_PER_SLAVE; n++) {
 268                 sstate->b[n].buf = malloc2(sstate->max_buf_size);
 269                 sstate->b[n].owner = sstate->thread_id;
 270         }
 271         if (sstate->alloc_hook) sstate->alloc_hook(sstate);
 272 }
 273
 274 /* Get a free binary buffer, first invalidating it in the receive
 275  * queue if necessary. In practice all buffers should be used
 276  * before they are invalidated, if BUFFERS_PER_SLAVE is large enough.
 277  * slave_lock is held on both entry and exit of this function. */
 278 static void *
 279 get_free_buf(struct slave_state *sstate)
 280 {
 281         int newest = (sstate->newest_buf + 1) & (BUFFERS_PER_SLAVE - 1);
 282         sstate->newest_buf = newest;
 283         void *buf = sstate->b[newest].buf;
 284
 285         if (DEBUGVV(7)) {
 286                 char b[1024];
 287                 snprintf(b, sizeof(b),
 288                          "get free %d index %d buf=%p age %d qlength %d\n", newest,
 289                          sstate->b[newest].queue_index, buf, queue_age, queue_length);
 290                 logline(&sstate->client, "? ", b);
 291         }
 292
 293         int index = sstate->b[newest].queue_index;
 294         if (index < 0) return buf;
 295
 296         /* Invalidate the buffer if the calling thread still owns its previous
 297          * entry in the receive queue. The entry may have been overwritten by
 298          * another thread, but only after a new move which invalidates the
 299          * entire receive queue. */
 300         if (receive_queue[index] && receive_queue[index]->owner == sstate->thread_id) {
 301                 receive_queue[index] = NULL;
 302         }
 303         sstate->b[newest].queue_index = -1;
 304         return buf;
 305 }
 306
 307 /* Insert a buffer in the receive queue. It should be the most
 308  * recent buffer allocated by the calling thread.
 309  * slave_lock is held on both entry and exit of this function. */
 310 static void
 311 insert_buf(struct slave_state *sstate, void *buf, int size)
 312 {
 313         assert(queue_length < queue_max_length);
 314
 315         int newest = sstate->newest_buf;
 316         assert(buf == sstate->b[newest].buf);
 317
 318         /* Update the buffer if necessary before making it
 319          * available to other threads. */
 320         if (sstate->insert_hook) sstate->insert_hook(buf, size);
 321
 322         if (DEBUGVV(7)) {
 323                 char b[1024];
 324                 snprintf(b, sizeof(b),
 325                          "insert newest %d age %d rq[%d]->%p owner %d\n",
 326                          newest, queue_age, queue_length, buf, sstate->thread_id);
 327                         logline(&sstate->client, "? ", b);
 328         }
 329         receive_queue[queue_length] = &sstate->b[newest];
 330         receive_queue[queue_length]->size = size;
 331         receive_queue[queue_length]->queue_index = queue_length;
 332         queue_length++;
 333 }
 334
 335 /* Clear the receive queue. The buffer pointers do not have to be cleared
 336  * here, this is done as each buffer is recycled.
 337  * slave_lock is held on both entry and exit of this function. */
 338 void
 339 clear_receive_queue(void)
 340 {
 341         if (DEBUGL(3)) {
 342                 char buf[1024];
 343                 snprintf(buf, sizeof(buf), "clear queue, old length %d age %d\n",
 344                          queue_length, queue_age);
 345                 logline(NULL, "? ", buf);
 346         }
 347         queue_length = 0;
 348         queue_age++;
 349 }
 350
 351 /* Process the reply received from a slave machine.
 352  * Copy the ascii part to reply_buf and insert the binary part
 353  * (if any) in the receive queue.
 354  * Return false if ok, true if the slave is out of sync.
 355  * slave_lock is held on both entry and exit of this function. */
 356 static bool
 357 process_reply(int reply_id, char *reply, char *reply_buf,
 358               void *bin_reply, int bin_size, int *last_reply_id,
 359               int *reply_slot, struct slave_state *sstate)
 360 {
 361         /* Resend everything if slave returned an error. */
 362         if (*reply != '=') {
 363                 *last_reply_id = -1;
 364                 return true;
 365         }
 366         /* Make sure we are still in sync. cmd_count may have
 367          * changed but the reply is valid as long as cmd_id didn't
 368          * change (this only occurs for consecutive genmoves). */
 369         int cmd_id = atoi(gtp_cmd);
 370         if (reply_id != cmd_id) {
 371                 *last_reply_id = reply_id;
 372                 return true;
 373         }
 374
 375         strncpy(reply_buf, reply, CMDS_SIZE);
 376         if (reply_id != *last_reply_id)
 377                 *reply_slot = reply_count++;
 378         gtp_replies[*reply_slot] = reply_buf;
 379
 380         if (bin_size) insert_buf(sstate, bin_reply, bin_size);
 381
 382         pthread_cond_signal(&reply_cond);
 383         *last_reply_id = reply_id;
 384         return false;
 385 }
 386
 387 /* Get the binary arg for the given command, and update the command
 388  * if necessary. For now, only genmoves has a binary argument, and
 389  * we return the best stats increments from all other slaves.
 390  * Set *bin_size to 0 if the command doesn't take binary arguments,
 391  * but still return a buffer, to be used for the reply.
 392  * Return NULL if the binary arg is obsolete by the time we have
 393  * finished computing it, because a new command is available.
 394  * This version only gets the buffer for the reply, to be completed
 395  * in future commits.
 396  * slave_lock is held on both entry and exit of this function. */
 397 void *
 398 get_binary_arg(struct slave_state *sstate, char *cmd, int cmd_size, int *bin_size)
 399 {
 400         int cmd_id = atoi(gtp_cmd);
 401         void *buf = get_free_buf(sstate);
 402
 403         *bin_size = 0;
 404         char *s = strchr(cmd, '@');
 405         if (!s || !sstate->args_hook) return buf;
 406
 407         int size = sstate->args_hook(buf, sstate, cmd_id);
 408
 409         /* Check that the command is still valid. */
 410         if (atoi(gtp_cmd) != cmd_id) return NULL;
 411
 412         /* Set the correct binary size for this slave.
 413          * cmd may have been overwritten with new parameters. */
 414         *bin_size = size;
 415         s = strchr(cmd, '@');
 416         assert(s);
 417         snprintf(s, cmd + cmd_size - s, "@%d\n", size);
 418         return buf;
 419 }
 420
 421 /* Main loop of a slave thread.
 422  * Send the current command to the slave machine and wait for a reply.
 423  * Resend command history if the slave machine is out of sync.
 424  * Returns when the connection with the slave machine is cut.
 425  * slave_lock is held on both entry and exit of this function. */
 426 static void
 427 slave_loop(FILE *f, char *reply_buf, struct slave_state *sstate, bool resend)
 428 {
 429         char *to_send;
 430         int last_cmd_count = 0;
 431         int last_reply_id = -1;
 432         int reply_slot = -1;
 433         for (;;) {
 434                 if (resend) {
 435                         /* Resend complete or partial history */
 436                         to_send = next_command(last_reply_id);
 437                 } else {
 438                         /* Wait for a new command. */
 439                         while (last_cmd_count == cmd_count)
 440                                 pthread_cond_wait(&cmd_cond, &slave_lock);
 441                         to_send = gtp_cmd;
 442                 }
 443
 444                 /* Command available, send it to slave machine.
 445                  * If slave was out of sync, send the history.
 446                  * But first get binary arguments if necessary. */
 447                 int bin_size = 0;
 448                 void *bin_buf = get_binary_arg(sstate, gtp_cmd,
 449                                                gtp_cmds + CMDS_SIZE - gtp_cmd,
 450                                                &bin_size);
 451                 /* Check that the command is still valid. */
 452                 resend = true;
 453                 if (!bin_buf) continue;
 454
 455                 /* Send the command and get the reply, which always ends with \n\n
 456                  * The slave machine sends "=id reply" or "?id reply"
 457                  * with id == cmd_id if it is in sync. */
 458                 last_cmd_count = cmd_count;
 459                 char buf[CMDS_SIZE];
 460                 int reply_id = send_command(to_send, bin_buf, &bin_size, f,
 461                                             sstate, buf);
 462                 if (reply_id == -1) return;
 463
 464                 resend = process_reply(reply_id, buf, reply_buf, bin_buf, bin_size,
 465                                        &last_reply_id, &reply_slot, sstate);
 466         }
 467 }
 468
 469 /* Thread sending gtp commands to one slave machine, and
 470  * reading replies. If a slave machine dies, this thread waits
 471  * for a connection from another slave.
 472  * The large buffers are allocated only once we get a first
 473  * connection, to avoid wasting memory if max_slaves is too large.
 474  * We do not invalidate the received buffers if a slave disconnects;
 475  * they are still useful for other slaves. */
 476 static void *
 477 slave_thread(void *arg)
 478 {
 479         struct slave_state sstate = default_sstate;
 480         sstate.thread_id = (long)arg;
 481
 482         assert(sstate.slave_sock >= 0);
 483         char reply_buf[CMDS_SIZE];
 484         bool resend = false;
 485
 486         for (;;) {
 487                 /* Wait for a connection from any slave. */
 488                 struct in_addr client;
 489                 int conn = open_server_connection(sstate.slave_sock, &client);
 490
 491                 FILE *f = fdopen(conn, "r+");
 492                 if (DEBUGL(2)) {
 493                         snprintf(reply_buf, sizeof(reply_buf),
 494                                  "new slave, id %d\n", sstate.thread_id);
 495                         logline(&client, "= ", reply_buf);
 496                 }
 497
 498                 /* Minimal check of the slave identity. */
 499                 fputs("name\n", f);
 500                 if (!fgets(reply_buf, sizeof(reply_buf), f)
 501                     || strncasecmp(reply_buf, "= Pachi", 7)
 502                     || !fgets(reply_buf, sizeof(reply_buf), f)
 503                     || strcmp(reply_buf, "\n")) {
 504                         logline(&client, "? ", "bad slave\n");
 505                         fclose(f);
 506                         continue;
 507                 }
 508
 509                 if (!resend) slave_state_alloc(&sstate);
 510                 sstate.client = client;
 511
 512                 pthread_mutex_lock(&slave_lock);
 513                 active_slaves++;
 514                 slave_loop(f, reply_buf, &sstate, resend);
 515
 516                 assert(active_slaves > 0);
 517                 active_slaves--;
 518                 // Unblock main thread if it was waiting for this slave.
 519                 pthread_cond_signal(&reply_cond);
 520                 pthread_mutex_unlock(&slave_lock);
 521
 522                 resend = true;
 523                 if (DEBUGL(2))
 524                         logline(&client, "= ", "lost slave\n");
 525                 fclose(f);
 526         }
 527 }
 528
 529 /* Create a new gtp command for all slaves. The slave lock is held
 530  * upon entry and upon return, so the command will actually be
 531  * sent when the lock is released. The last command is overwritten
 532  * if gtp_cmd points to a non-empty string. cmd is a single word;
 533  * args has all arguments and is empty or has a trailing \n */
 534 void
 535 update_cmd(struct board *b, char *cmd, char *args, bool new_id)
 536 {
 537         assert(gtp_cmd);
 538         /* To make sure the slaves are in sync, we ignore the original id
 539          * and use the board number plus some random bits as gtp id. */
 540         static int gtp_id = -1;
 541         int moves = is_reset(cmd) ? 0 : b->moves;
 542         if (new_id) {
 543                 int prev_id = gtp_id;
 544                 do {
 545                         /* fast_random() is 16-bit only so the multiplication can't overflow. */
 546                         gtp_id = force_reply(moves + fast_random(65535) * DIST_GAMELEN);
 547                 } while (gtp_id == prev_id);
 548                 reply_count = 0;
 549         }
 550         snprintf(gtp_cmd, gtp_cmds + CMDS_SIZE - gtp_cmd, "%d %s %s",
 551                  gtp_id, cmd, *args ? args : "\n");
 552         cmd_count++;
 553
 554         /* Remember history for out-of-sync slaves. */
 555         static int slot = 0;
 556         static struct cmd_history *last = NULL;
 557         if (new_id) {
 558                 if (last) last->next_cmd = gtp_cmd;
 559                 slot = (slot + 1) % MAX_CMDS_PER_MOVE;
 560                 last = &history[moves][slot];
 561                 last->gtp_id = gtp_id;
 562                 last->next_cmd = NULL;
 563         }
 564         // Notify the slave threads about the new command.
 565         pthread_cond_broadcast(&cmd_cond);
 566 }
 567
 568 /* Update the command history, then create a new gtp command
 569  * for all slaves. The slave lock is held upon entry and
 570  * upon return, so the command will actually be sent when the
 571  * lock is released. cmd is a single word; args has all
 572  * arguments and is empty or has a trailing \n */
 573 void
 574 new_cmd(struct board *b, char *cmd, char *args)
 575 {
 576         // Clear the history when a new game starts:
 577         if (!gtp_cmd || is_gamestart(cmd)) {
 578                 gtp_cmd = gtp_cmds;
 579                 memset(history, 0, sizeof(history));
 580         } else {
 581                 /* Preserve command history for new slaves.
 582                  * To indicate that the slave should only reply to
 583                  * the last command we force the id of previous
 584                  * commands to be just the move number. */
 585                 int id = prevent_reply(atoi(gtp_cmd));
 586                 int len = strspn(gtp_cmd, "0123456789");
 587                 char buf[32];
 588                 snprintf(buf, sizeof(buf), "%0*d", len, id);
 589                 memcpy(gtp_cmd, buf, len);
 590
 591                 gtp_cmd += strlen(gtp_cmd);
 592         }
 593
 594         // Let the slave threads send the new gtp command:
 595         update_cmd(b, cmd, args, true);
 596 }
 597
 598 /* Wait for at least one new reply. Return when at least
 599  * min_replies slaves have already replied, or when the
 600  * given absolute time is passed.
 601  * The replies are returned in gtp_replies[0..reply_count-1]
 602  * slave_lock is held on entry and on return. */
 603 void
 604 get_replies(double time_limit, int min_replies)
 605 {
 606         for (;;) {
 607                 if (reply_count > 0) {
 608                         struct timespec ts;
 609                         double sec;
 610                         ts.tv_nsec = (int)(modf(time_limit, &sec)*1000000000.0);
 611                         ts.tv_sec = (int)sec;
 612                         pthread_cond_timedwait(&reply_cond, &slave_lock, &ts);
 613                 } else {
 614                         pthread_cond_wait(&reply_cond, &slave_lock);
 615                 }
 616                 if (reply_count == 0) continue;
 617                 if (reply_count >= min_replies || reply_count >= active_slaves) return;
 618                 if (time_now() >= time_limit) break;
 619         }
 620         if (DEBUGL(1)) {
 621                 char buf[1024];
 622                 snprintf(buf, sizeof(buf),
 623                          "get_replies timeout %.3f >= %.3f, replies %d < min %d, active %d\n",
 624                          time_now() - start_time, time_limit - start_time,
 625                          reply_count, min_replies, active_slaves);
 626                 logline(NULL, "? ", buf);
 627         }
 628         assert(reply_count > 0);
 629 }
 630
 631 /* In a 30s move with at least 5ms per genmoves we get at most
 632  * 6000 genmoves per slave. */
 633 #define MAX_GENMOVES_PER_SLAVE 6000
 634
 635 /* Allocate the receive queue, and create the slave and proxy threads.
 636  * max_buf_size and the merge-related fields of default_sstate must
 637  * already be initialized. */
 638 void
 639 protocol_init(char *slave_port, char *proxy_port, int max_slaves)
 640 {
 641         start_time = time_now();
 642
 643         queue_max_length = max_slaves * MAX_GENMOVES_PER_SLAVE;
 644         receive_queue = calloc2(queue_max_length, sizeof(*receive_queue));
 645
 646         default_sstate.slave_sock = port_listen(slave_port, max_slaves);
 647         default_sstate.last_processed = -1;
 648
 649         for (int n = 0; n < BUFFERS_PER_SLAVE; n++) {
 650                 default_sstate.b[n].queue_index = -1;
 651         }
 652
 653         pthread_t thread;
 654         for (int id = 0; id < max_slaves; id++) {
 655                 pthread_create(&thread, NULL, slave_thread, (void *)(long)id);
 656         }
 657
 658         if (proxy_port) {
 659                 int proxy_sock = port_listen(proxy_port, max_slaves);
 660                 for (int id = 0; id < max_slaves; id++) {
 661                         pthread_create(&thread, NULL, proxy_thread, (void *)(long)proxy_sock);
 662                 }
 663         }
 664 }