| 1 | /* |
| 2 | * This Source Code Form is subject to the terms of the Mozilla Public |
| 3 | * License, v. 2.0. If a copy of the MPL was not distributed with this |
| 4 | * file, You can obtain one at http://mozilla.org/MPL/2.0/. |
| 5 | * |
| 6 | * Copyright 1997 - July 2008 CWI, August 2008 - 2019 MonetDB B.V. |
| 7 | */ |
| 8 | |
| 9 | #include "monetdb_config.h" |
| 10 | #include <sys/types.h> |
| 11 | #include <sys/un.h> |
| 12 | #include <sys/stat.h> |
| 13 | #include <signal.h> |
| 14 | #include <unistd.h> |
| 15 | #include <string.h> /* char ** */ |
| 16 | #include <time.h> /* localtime */ |
| 17 | |
| 18 | #include "msabaoth.h" |
| 19 | #include "utils/utils.h" |
| 20 | #include "utils/glob.h" |
| 21 | #include "utils/properties.h" |
| 22 | #include "mutils.h" |
| 23 | |
| 24 | #include "merovingian.h" |
| 25 | #include "discoveryrunner.h" /* remotedb */ |
| 26 | #include "multiplex-funnel.h" /* multiplexInit */ |
| 27 | #include "forkmserver.h" |
| 28 | |
| 29 | static pthread_mutex_t fork_lock = PTHREAD_MUTEX_INITIALIZER; |
| 30 | |
| 31 | /** |
| 32 | * The terminateProcess function tries to let the given mserver process |
| 33 | * shut down gracefully within a given time-out. If that fails, it |
| 34 | * sends the deadly SIGKILL signal to the mserver process and returns. |
| 35 | */ |
| 36 | void |
| 37 | terminateProcess(pid_t pid, char *dbname, mtype type, int lock) |
| 38 | { |
| 39 | sabdb *stats; |
| 40 | char *er; |
| 41 | int i; |
| 42 | confkeyval *kv; |
| 43 | |
| 44 | if (lock) |
| 45 | pthread_mutex_lock(&fork_lock); |
| 46 | |
| 47 | er = msab_getStatus(&stats, dbname); |
| 48 | if (er != NULL) { |
| 49 | if (lock) |
| 50 | pthread_mutex_unlock(&fork_lock); |
| 51 | Mfprintf(stderr, "cannot terminate process %lld: %s\n" , |
| 52 | (long long int)pid, er); |
| 53 | free(er); |
| 54 | free(dbname); |
| 55 | return; |
| 56 | } |
| 57 | |
| 58 | if (stats == NULL) { |
| 59 | if (lock) |
| 60 | pthread_mutex_unlock(&fork_lock); |
| 61 | Mfprintf(stderr, "strange, process %lld serves database '%s' " |
| 62 | "which does not exist\n" , (long long int)pid, dbname); |
| 63 | free(dbname); |
| 64 | return; |
| 65 | } |
| 66 | |
| 67 | switch (stats->state) { |
| 68 | case SABdbRunning: |
| 69 | /* ok, what we expect */ |
| 70 | break; |
| 71 | case SABdbCrashed: |
| 72 | if (lock) |
| 73 | pthread_mutex_unlock(&fork_lock); |
| 74 | Mfprintf(stderr, "cannot shut down database '%s', mserver " |
| 75 | "(pid %lld) has crashed\n" , |
| 76 | dbname, (long long int)pid); |
| 77 | msab_freeStatus(&stats); |
| 78 | free(dbname); |
| 79 | return; |
| 80 | case SABdbInactive: |
| 81 | if (lock) |
| 82 | pthread_mutex_unlock(&fork_lock); |
| 83 | Mfprintf(stdout, "database '%s' appears to have shut down already\n" , |
| 84 | dbname); |
| 85 | fflush(stdout); |
| 86 | msab_freeStatus(&stats); |
| 87 | free(dbname); |
| 88 | return; |
| 89 | case SABdbStarting: |
| 90 | Mfprintf(stderr, "database '%s' appears to be starting up\n" , |
| 91 | dbname); |
| 92 | /* starting up, so we'll go to the shut down phase */ |
| 93 | break; |
| 94 | default: |
| 95 | if (lock) |
| 96 | pthread_mutex_unlock(&fork_lock); |
| 97 | Mfprintf(stderr, "unknown state: %d\n" , (int)stats->state); |
| 98 | msab_freeStatus(&stats); |
| 99 | free(dbname); |
| 100 | return; |
| 101 | } |
| 102 | |
| 103 | if (type == MEROFUN) { |
| 104 | if (lock) |
| 105 | pthread_mutex_unlock(&fork_lock); |
| 106 | multiplexDestroy(dbname); |
| 107 | msab_freeStatus(&stats); |
| 108 | free(dbname); |
| 109 | return; |
| 110 | } else if (type != MERODB) { |
| 111 | /* barf */ |
| 112 | if (lock) |
| 113 | pthread_mutex_unlock(&fork_lock); |
| 114 | Mfprintf(stderr, "cannot stop merovingian process role: %s\n" , dbname); |
| 115 | msab_freeStatus(&stats); |
| 116 | free(dbname); |
| 117 | return; |
| 118 | } |
| 119 | |
| 120 | /* ok, once we get here, we'll be shutting down the server */ |
| 121 | Mfprintf(stdout, "sending process %lld (database '%s') the " |
| 122 | "TERM signal\n" , (long long int)pid, dbname); |
| 123 | kill(pid, SIGTERM); |
| 124 | kv = findConfKey(_mero_props, "exittimeout" ); |
| 125 | for (i = 0; i < atoi(kv->val) * 2; i++) { |
| 126 | if (stats != NULL) |
| 127 | msab_freeStatus(&stats); |
| 128 | sleep_ms(500); |
| 129 | er = msab_getStatus(&stats, dbname); |
| 130 | if (er != NULL) { |
| 131 | Mfprintf(stderr, "unexpected problem: %s\n" , er); |
| 132 | free(er); |
| 133 | /* don't die, just continue, so we KILL in the end */ |
| 134 | } else if (stats == NULL) { |
| 135 | Mfprintf(stderr, "hmmmm, database '%s' suddenly doesn't exist " |
| 136 | "any more\n" , dbname); |
| 137 | } else { |
| 138 | switch (stats->state) { |
| 139 | case SABdbRunning: |
| 140 | case SABdbStarting: |
| 141 | /* ok, try again */ |
| 142 | break; |
| 143 | case SABdbCrashed: |
| 144 | if (lock) |
| 145 | pthread_mutex_unlock(&fork_lock); |
| 146 | Mfprintf (stderr, "database '%s' crashed after SIGTERM\n" , |
| 147 | dbname); |
| 148 | msab_freeStatus(&stats); |
| 149 | free(dbname); |
| 150 | return; |
| 151 | case SABdbInactive: |
| 152 | if (lock) |
| 153 | pthread_mutex_unlock(&fork_lock); |
| 154 | Mfprintf(stdout, "database '%s' has shut down\n" , dbname); |
| 155 | fflush(stdout); |
| 156 | msab_freeStatus(&stats); |
| 157 | free(dbname); |
| 158 | return; |
| 159 | default: |
| 160 | Mfprintf(stderr, "unknown state: %d\n" , (int)stats->state); |
| 161 | break; |
| 162 | } |
| 163 | } |
| 164 | } |
| 165 | Mfprintf(stderr, "timeout of %s seconds expired, sending process %lld" |
| 166 | " (database '%s') the KILL signal\n" , |
| 167 | kv->val, (long long int)pid, dbname); |
| 168 | kill(pid, SIGKILL); |
| 169 | msab_freeStatus(&stats); |
| 170 | free(dbname); |
| 171 | if (lock) |
| 172 | pthread_mutex_unlock(&fork_lock); |
| 173 | } |
| 174 | |
| 175 | /** |
| 176 | * Fork an mserver and detach. Before forking off, Sabaoth is consulted |
| 177 | * to see if forking makes sense, or whether it is necessary at all, or |
| 178 | * forbidden by restart policy, e.g. when in maintenance. |
| 179 | */ |
| 180 | |
| 181 | #define MAX_NR_ARGS 511 |
| 182 | |
| 183 | err |
| 184 | forkMserver(char *database, sabdb** stats, bool force) |
| 185 | { |
| 186 | pid_t pid; |
| 187 | char *er; |
| 188 | sabuplog info; |
| 189 | struct tm *t; |
| 190 | char tstr[20]; |
| 191 | int pfdo[2]; |
| 192 | int pfde[2]; |
| 193 | dpair dp; |
| 194 | char vaultkey[512]; |
| 195 | struct stat statbuf; |
| 196 | char upmin[8]; |
| 197 | char upavg[8]; |
| 198 | char upmax[8]; |
| 199 | confkeyval *ckv, *kv, *list; |
| 200 | SABdbState state; |
| 201 | char *sabdbfarm; |
| 202 | char dbpath[1024]; |
| 203 | char [1024]; |
| 204 | char port[32]; |
| 205 | char listenaddr[512]; |
| 206 | char muri[512]; /* possibly undersized */ |
| 207 | char usock[512]; |
| 208 | bool mydoproxy; |
| 209 | char nthreads[32]; |
| 210 | char nclients[32]; |
| 211 | char pipeline[512]; |
| 212 | char memmaxsize[64]; |
| 213 | char vmmaxsize[64]; |
| 214 | char *readonly = NULL; |
| 215 | char *embeddedr = NULL; |
| 216 | char *embeddedpy = NULL; |
| 217 | char *embeddedc = NULL; |
| 218 | char *ipv6 = NULL; |
| 219 | char * = NULL; |
| 220 | char * = NULL; |
| 221 | char * = NULL; |
| 222 | char *argv[MAX_NR_ARGS+1]; /* for the exec arguments */ |
| 223 | char property_other[1024]; |
| 224 | int c = 0; |
| 225 | unsigned int mport; |
| 226 | |
| 227 | er = msab_getStatus(stats, database); |
| 228 | if (er != NULL) { |
| 229 | err e = newErr("%s" , er); |
| 230 | free(er); |
| 231 | return(e); |
| 232 | } |
| 233 | |
| 234 | /* NOTE: remotes also include locals through self announcement */ |
| 235 | if (*stats == NULL) { |
| 236 | *stats = getRemoteDB(database); |
| 237 | |
| 238 | if (*stats != NULL) |
| 239 | return(NO_ERR); |
| 240 | |
| 241 | return(newErr("no such database: %s" , database)); |
| 242 | } |
| 243 | |
| 244 | /* Since we ask for a specific database, it should be either there |
| 245 | * or not there. Since we checked the latter case above, it should |
| 246 | * just be there, and be the right one. There also shouldn't be |
| 247 | * more than one entry in the list, so we assume we have the right |
| 248 | * one here. */ |
| 249 | |
| 250 | if ((*stats)->state == SABdbRunning) |
| 251 | /* return before doing expensive stuff, when this db just seems |
| 252 | * to be running */ |
| 253 | return(NO_ERR); |
| 254 | |
| 255 | /* Make sure we only start one mserver5 at the same time, this is a |
| 256 | * horsedrug for preventing race-conditions where two or more |
| 257 | * clients start the same database at the same time, because they |
| 258 | * were all identified as being SABdbInactive. If this "global" |
| 259 | * lock ever becomes a problem, we can reduce it to a per-database |
| 260 | * lock instead. */ |
| 261 | pthread_mutex_lock(&fork_lock); |
| 262 | |
| 263 | /* refetch the status, as it may have changed */ |
| 264 | msab_freeStatus(stats); |
| 265 | er = msab_getStatus(stats, database); |
| 266 | if (er != NULL) { |
| 267 | err e = newErr("%s" , er); |
| 268 | free(er); |
| 269 | pthread_mutex_unlock(&fork_lock); |
| 270 | return(e); |
| 271 | } |
| 272 | |
| 273 | ckv = getDefaultProps(); |
| 274 | readAllProps(ckv, (*stats)->path); |
| 275 | kv = findConfKey(ckv, "type" ); |
| 276 | if (kv->val == NULL) |
| 277 | kv = findConfKey(_mero_db_props, "type" ); |
| 278 | |
| 279 | if ((*stats)->locked) { |
| 280 | if (!force) { |
| 281 | Mfprintf(stdout, "%s '%s' is under maintenance\n" , |
| 282 | kv->val, database); |
| 283 | freeConfFile(ckv); |
| 284 | free(ckv); |
| 285 | pthread_mutex_unlock(&fork_lock); |
| 286 | return(NO_ERR); |
| 287 | } else { |
| 288 | Mfprintf(stdout, "startup of %s under maintenance " |
| 289 | "'%s' forced\n" , kv->val, database); |
| 290 | } |
| 291 | } |
| 292 | |
| 293 | /* retrieve uplog information to print a short conclusion */ |
| 294 | er = msab_getUplogInfo(&info, *stats); |
| 295 | if (er != NULL) { |
| 296 | err e = newErr("could not retrieve uplog information: %s" , er); |
| 297 | free(er); |
| 298 | msab_freeStatus(stats); |
| 299 | freeConfFile(ckv); |
| 300 | free(ckv); |
| 301 | pthread_mutex_unlock(&fork_lock); |
| 302 | return(e); |
| 303 | } |
| 304 | |
| 305 | switch ((*stats)->state) { |
| 306 | case SABdbRunning: |
| 307 | freeConfFile(ckv); |
| 308 | free(ckv); |
| 309 | pthread_mutex_unlock(&fork_lock); |
| 310 | return(NO_ERR); |
| 311 | case SABdbCrashed: |
| 312 | t = localtime(&info.lastcrash); |
| 313 | strftime(tstr, sizeof(tstr), "%Y-%m-%d %H:%M:%S" , t); |
| 314 | secondsToString(upmin, info.minuptime, 1); |
| 315 | secondsToString(upavg, info.avguptime, 1); |
| 316 | secondsToString(upmax, info.maxuptime, 1); |
| 317 | Mfprintf(stdout, "%s '%s' has crashed after start on %s, " |
| 318 | "attempting restart, " |
| 319 | "up min/avg/max: %s/%s/%s, " |
| 320 | "crash average: %d.00 %.2f %.2f (%d-%d=%d)\n" , |
| 321 | kv->val, database, tstr, |
| 322 | upmin, upavg, upmax, |
| 323 | info.crashavg1, info.crashavg10, info.crashavg30, |
| 324 | info.startcntr, info.stopcntr, info.crashcntr); |
| 325 | break; |
| 326 | case SABdbInactive: |
| 327 | secondsToString(upmin, info.minuptime, 1); |
| 328 | secondsToString(upavg, info.avguptime, 1); |
| 329 | secondsToString(upmax, info.maxuptime, 1); |
| 330 | Mfprintf(stdout, "starting %s '%s', " |
| 331 | "up min/avg/max: %s/%s/%s, " |
| 332 | "crash average: %d.00 %.2f %.2f (%d-%d=%d)\n" , |
| 333 | kv->val, database, |
| 334 | upmin, upavg, upmax, |
| 335 | info.crashavg1, info.crashavg10, info.crashavg30, |
| 336 | info.startcntr, info.stopcntr, info.crashcntr); |
| 337 | break; |
| 338 | default: |
| 339 | /* this also includes SABdbStarting, which we shouldn't ever |
| 340 | * see due to the global starting lock */ |
| 341 | state = (*stats)->state; |
| 342 | msab_freeStatus(stats); |
| 343 | freeConfFile(ckv); |
| 344 | free(ckv); |
| 345 | pthread_mutex_unlock(&fork_lock); |
| 346 | return(newErr("unknown or impossible state: %d" , |
| 347 | (int)state)); |
| 348 | } |
| 349 | |
| 350 | /* create the pipes (filedescriptors) now, such that we and the |
| 351 | * child have the same descriptor set */ |
| 352 | if (pipe(pfdo) == -1) { |
| 353 | int e = errno; |
| 354 | msab_freeStatus(stats); |
| 355 | freeConfFile(ckv); |
| 356 | free(ckv); |
| 357 | pthread_mutex_unlock(&fork_lock); |
| 358 | return(newErr("unable to create pipe: %s" , strerror(e))); |
| 359 | } |
| 360 | if (pipe(pfde) == -1) { |
| 361 | int e = errno; |
| 362 | close(pfdo[0]); |
| 363 | close(pfdo[1]); |
| 364 | msab_freeStatus(stats); |
| 365 | freeConfFile(ckv); |
| 366 | free(ckv); |
| 367 | pthread_mutex_unlock(&fork_lock); |
| 368 | return(newErr("unable to create pipe: %s" , strerror(e))); |
| 369 | } |
| 370 | |
| 371 | /* a multiplex-funnel means starting a separate thread */ |
| 372 | if (strcmp(kv->val, "mfunnel" ) == 0) { |
| 373 | FILE *f1, *f2; |
| 374 | /* create a dpair entry */ |
| 375 | pthread_mutex_lock(&_mero_topdp_lock); |
| 376 | |
| 377 | dp = _mero_topdp; |
| 378 | while (dp->next != NULL) |
| 379 | dp = dp->next; |
| 380 | dp = dp->next = malloc(sizeof(struct _dpair)); |
| 381 | dp->out = pfdo[0]; |
| 382 | dp->err = pfde[0]; |
| 383 | dp->next = NULL; |
| 384 | dp->type = MEROFUN; |
| 385 | dp->pid = getpid(); |
| 386 | dp->dbname = strdup(database); |
| 387 | dp->flag = 0; |
| 388 | |
| 389 | pthread_mutex_unlock(&_mero_topdp_lock); |
| 390 | |
| 391 | kv = findConfKey(ckv, "mfunnel" ); |
| 392 | if(!(f1 = fdopen(pfdo[1], "a" ))) { |
| 393 | freeConfFile(ckv); |
| 394 | free(ckv); |
| 395 | pthread_mutex_unlock(&fork_lock); |
| 396 | return newErr("Failed to open file descriptor\n" ); |
| 397 | } |
| 398 | if(!(f2 = fdopen(pfde[1], "a" ))) { |
| 399 | fclose(f1); |
| 400 | freeConfFile(ckv); |
| 401 | free(ckv); |
| 402 | pthread_mutex_unlock(&fork_lock); |
| 403 | return newErr("Failed to open file descriptor\n" ); |
| 404 | } |
| 405 | if ((er = multiplexInit(database, kv->val, f1, f2)) != NO_ERR) { |
| 406 | Mfprintf(stderr, "failed to create multiplex-funnel: %s\n" , |
| 407 | getErrMsg(er)); |
| 408 | freeConfFile(ckv); |
| 409 | free(ckv); |
| 410 | pthread_mutex_unlock(&fork_lock); |
| 411 | return(er); |
| 412 | } |
| 413 | freeConfFile(ckv); |
| 414 | free(ckv); |
| 415 | |
| 416 | /* refresh stats, now we will have a connection registered */ |
| 417 | msab_freeStatus(stats); |
| 418 | er = msab_getStatus(stats, database); |
| 419 | pthread_mutex_unlock(&fork_lock); |
| 420 | if (er != NULL) { |
| 421 | /* since the client mserver lives its own life anyway, |
| 422 | * it's not really a problem we exit here */ |
| 423 | err e = newErr("%s" , er); |
| 424 | free(er); |
| 425 | return(e); |
| 426 | } |
| 427 | return(NO_ERR); |
| 428 | } |
| 429 | |
| 430 | /* check if the vaultkey is there, otherwise abort early (value |
| 431 | * lateron reused when server is started) */ |
| 432 | snprintf(vaultkey, sizeof(vaultkey), "%s/.vaultkey" , (*stats)->path); |
| 433 | if (stat(vaultkey, &statbuf) == -1) { |
| 434 | msab_freeStatus(stats); |
| 435 | freeConfFile(ckv); |
| 436 | free(ckv); |
| 437 | pthread_mutex_unlock(&fork_lock); |
| 438 | return(newErr("cannot start database '%s': no .vaultkey found " |
| 439 | "(did you create the database with `monetdb create %s`?)" , |
| 440 | database, database)); |
| 441 | } |
| 442 | |
| 443 | er = msab_getDBfarm(&sabdbfarm); |
| 444 | if (er != NULL) { |
| 445 | freeConfFile(ckv); |
| 446 | free(ckv); |
| 447 | pthread_mutex_unlock(&fork_lock); |
| 448 | return(er); |
| 449 | } |
| 450 | |
| 451 | mydoproxy = strcmp(getConfVal(_mero_props, "forward" ), "proxy" ) == 0; |
| 452 | |
| 453 | kv = findConfKey(ckv, "nthreads" ); |
| 454 | if (kv->val == NULL) |
| 455 | kv = findConfKey(_mero_db_props, "nthreads" ); |
| 456 | if (kv->val != NULL) { |
| 457 | snprintf(nthreads, sizeof(nthreads), "gdk_nr_threads=%s" , kv->val); |
| 458 | } else { |
| 459 | nthreads[0] = '\0'; |
| 460 | } |
| 461 | |
| 462 | kv = findConfKey(ckv, "nclients" ); |
| 463 | if (kv->val == NULL) |
| 464 | kv = findConfKey(_mero_db_props, "nclients" ); |
| 465 | if (kv->val != NULL) { |
| 466 | snprintf(nclients, sizeof(nclients), "max_clients=%s" , kv->val); |
| 467 | } else { |
| 468 | nclients[0] = '\0'; |
| 469 | } |
| 470 | |
| 471 | kv = findConfKey(ckv, "optpipe" ); |
| 472 | if (kv->val == NULL) |
| 473 | kv = findConfKey(_mero_db_props, "optpipe" ); |
| 474 | if (kv->val != NULL) { |
| 475 | snprintf(pipeline, sizeof(pipeline), "sql_optimizer=%s" , kv->val); |
| 476 | } else { |
| 477 | pipeline[0] = '\0'; |
| 478 | } |
| 479 | |
| 480 | kv = findConfKey(ckv, "memmaxsize" ); |
| 481 | if (kv->val != NULL) { |
| 482 | snprintf(memmaxsize, sizeof(memmaxsize), "gdk_mem_maxsize=%s" , kv->val); |
| 483 | } else { |
| 484 | memmaxsize[0] = '\0'; |
| 485 | } |
| 486 | |
| 487 | kv = findConfKey(ckv, "vmmaxsize" ); |
| 488 | if (kv->val != NULL) { |
| 489 | snprintf(vmmaxsize, sizeof(vmmaxsize), "gdk_vm_maxsize=%s" , kv->val); |
| 490 | } else { |
| 491 | vmmaxsize[0] = '\0'; |
| 492 | } |
| 493 | |
| 494 | kv = findConfKey(ckv, "readonly" ); |
| 495 | if (kv->val != NULL && strcmp(kv->val, "no" ) != 0) |
| 496 | readonly = "--readonly" ; |
| 497 | |
| 498 | kv = findConfKey(ckv, "embedr" ); |
| 499 | if (kv->val != NULL && strcmp(kv->val, "no" ) != 0) |
| 500 | embeddedr = "embedded_r=true" ; |
| 501 | |
| 502 | kv = findConfKey(ckv, "embedpy" ); |
| 503 | if (kv->val != NULL && strcmp(kv->val, "no" ) != 0) |
| 504 | embeddedpy = "embedded_py=2" ; |
| 505 | |
| 506 | kv = findConfKey(ckv, "embedpy3" ); |
| 507 | if (kv->val != NULL && strcmp(kv->val, "no" ) != 0) { |
| 508 | if (embeddedpy) { |
| 509 | // only one python version can be active at a time |
| 510 | freeConfFile(ckv); |
| 511 | free(ckv); |
| 512 | pthread_mutex_unlock(&fork_lock); |
| 513 | free(sabdbfarm); |
| 514 | return newErr("attempting to start mserver with both embedded python2 and embedded python3; only one python version can be active at a time\n" ); |
| 515 | } |
| 516 | embeddedpy = "embedded_py=3" ; |
| 517 | } |
| 518 | kv = findConfKey(ckv, "embedc" ); |
| 519 | if (kv->val != NULL && strcmp(kv->val, "no" ) != 0) |
| 520 | embeddedc = "embedded_c=true" ; |
| 521 | kv = findConfKey(ckv, "dbextra" ); |
| 522 | if (kv != NULL && kv->val != NULL) { |
| 523 | dbextra = kv->val; |
| 524 | } |
| 525 | |
| 526 | kv = findConfKey(ckv, "listenaddr" ); |
| 527 | if (kv->val != NULL) { |
| 528 | if (mydoproxy) { |
| 529 | // listenaddr is only available on forwarding method |
| 530 | freeConfFile(ckv); |
| 531 | free(ckv); |
| 532 | pthread_mutex_unlock(&fork_lock); |
| 533 | free(sabdbfarm); |
| 534 | return newErr("attempting to start mserver with listening address while being proxied by monetdbd; this option is only possible on forward method\n" ); |
| 535 | } |
| 536 | snprintf(listenaddr, sizeof(listenaddr), "mapi_listenaddr=%s" , kv->val); |
| 537 | } else { |
| 538 | listenaddr[0] = '\0'; |
| 539 | } |
| 540 | mport = (unsigned int)getConfNum(_mero_props, "port" ); |
| 541 | ipv6 = getConfNum(_mero_props, "ipv6" ) == 1 ? "mapi_ipv6=true" : "mapi_ipv6=false" ; |
| 542 | |
| 543 | /* ok, now exec that mserver we want */ |
| 544 | snprintf(dbpath, sizeof(dbpath), |
| 545 | "--dbpath=%s/%s" , sabdbfarm, database); |
| 546 | free(sabdbfarm); |
| 547 | snprintf(vaultkey, sizeof(vaultkey), |
| 548 | "monet_vault_key=%s/.vaultkey" , (*stats)->path); |
| 549 | snprintf(muri, sizeof(muri), |
| 550 | "merovingian_uri=mapi:monetdb://%s:%u/%s" , |
| 551 | _mero_hostname, mport, database); |
| 552 | argv[c++] = _mero_mserver; |
| 553 | argv[c++] = dbpath; |
| 554 | argv[c++] = "--set" ; argv[c++] = muri; |
| 555 | if (dbextra != NULL) { |
| 556 | snprintf(dbextra_path, sizeof(dbextra_path), |
| 557 | "--dbextra=%s" , dbextra); |
| 558 | argv[c++] = dbextra_path; |
| 559 | } |
| 560 | if (mydoproxy) { |
| 561 | struct sockaddr_un s; /* only for sizeof(s.sun_path) :( */ |
| 562 | argv[c++] = "--set" ; argv[c++] = "mapi_open=false" ; |
| 563 | /* we "proxy", so we can just solely use UNIX domain sockets |
| 564 | * internally. Before we hit our head, check if we can |
| 565 | * actually use a UNIX socket (due to pathlength) */ |
| 566 | if (strlen((*stats)->path) + 11 < sizeof(s.sun_path)) { |
| 567 | snprintf(port, sizeof(port), "mapi_port=0" ); |
| 568 | snprintf(usock, sizeof(usock), "mapi_usock=%s/.mapi.sock" , |
| 569 | (*stats)->path); |
| 570 | } else { |
| 571 | argv[c++] = "--set" ; argv[c++] = "mapi_autosense=true" ; |
| 572 | /* for logic here, see comment below */ |
| 573 | snprintf(port, sizeof(port), "mapi_port=%u" , mport + 1); |
| 574 | snprintf(usock, sizeof(usock), "mapi_usock=" ); |
| 575 | } |
| 576 | } else { |
| 577 | if (listenaddr[0] != '\0') { |
| 578 | argv[c++] = "--set" ; argv[c++] = listenaddr; |
| 579 | } else { |
| 580 | argv[c++] = "--set" ; argv[c++] = "mapi_open=true" ; |
| 581 | } |
| 582 | argv[c++] = "--set" ; argv[c++] = "mapi_autosense=true" ; |
| 583 | /* avoid this mserver binding to the same port as merovingian |
| 584 | * but on another interface, (INADDR_ANY ... sigh) causing |
| 585 | * endless redirects since 0.0.0.0 is not a valid address to |
| 586 | * connect to, and hence the hostname is advertised instead */ |
| 587 | snprintf(port, sizeof(port), "mapi_port=%u" , mport + 1); |
| 588 | snprintf(usock, sizeof(usock), "mapi_usock=" ); |
| 589 | } |
| 590 | argv[c++] = "--set" ; argv[c++] = ipv6; |
| 591 | argv[c++] = "--set" ; argv[c++] = port; |
| 592 | argv[c++] = "--set" ; argv[c++] = usock; |
| 593 | argv[c++] = "--set" ; argv[c++] = vaultkey; |
| 594 | if (nthreads[0] != '\0') { |
| 595 | argv[c++] = "--set" ; argv[c++] = nthreads; |
| 596 | } |
| 597 | if (nclients[0] != '\0') { |
| 598 | argv[c++] = "--set" ; argv[c++] = nclients; |
| 599 | } |
| 600 | if (pipeline[0] != '\0') { |
| 601 | argv[c++] = "--set" ; argv[c++] = pipeline; |
| 602 | } |
| 603 | if (memmaxsize[0] != '\0') { |
| 604 | argv[c++] = "--set" ; argv[c++] = memmaxsize; |
| 605 | } |
| 606 | if (vmmaxsize[0] != '\0') { |
| 607 | argv[c++] = "--set" ; argv[c++] = vmmaxsize; |
| 608 | } |
| 609 | if (embeddedr != NULL) { |
| 610 | argv[c++] = "--set" ; argv[c++] = embeddedr; |
| 611 | } |
| 612 | if (embeddedpy != NULL) { |
| 613 | argv[c++] = "--set" ; argv[c++] = embeddedpy; |
| 614 | } |
| 615 | if (embeddedc != NULL) { |
| 616 | argv[c++] = "--set" ; argv[c++] = embeddedc; |
| 617 | } |
| 618 | if (readonly != NULL) { |
| 619 | argv[c++] = readonly; |
| 620 | } |
| 621 | /* get the rest (non-default) mserver props set in the conf file */ |
| 622 | list = ckv; |
| 623 | while (list->key != NULL) { |
| 624 | if (list->val != NULL && !defaultProperty(list->key)) { |
| 625 | argv[c++] = "--set" ; |
| 626 | snprintf(property_other, sizeof(property_other), "%s=%s" , list->key, list->val); |
| 627 | argv[c++] = strdup(property_other); |
| 628 | } |
| 629 | list++; |
| 630 | } |
| 631 | |
| 632 | /* Let's get extra mserver5 args from the environment */ |
| 633 | mserver5_extra = getenv("MSERVER5_EXTRA_ARGS" ); |
| 634 | while (c < MAX_NR_ARGS && (mserver5_extra_token = strsep(&mserver5_extra, " " ))) |
| 635 | argv[c++] = mserver5_extra_token; |
| 636 | |
| 637 | argv[c++] = NULL; |
| 638 | |
| 639 | freeConfFile(ckv); |
| 640 | free(ckv); /* can make ckv static and reuse it all the time */ |
| 641 | |
| 642 | /* make sure no entries are shot while adding and that we |
| 643 | * deliver a consistent state */ |
| 644 | pthread_mutex_lock(&_mero_topdp_lock); |
| 645 | |
| 646 | pid = fork(); |
| 647 | if (pid == 0) { |
| 648 | /* redirect stdout and stderr to a new pair of fds for |
| 649 | * logging help */ |
| 650 | ssize_t write_error; /* to avoid compiler warning */ |
| 651 | int dup_err; |
| 652 | close(pfdo[0]); |
| 653 | dup_err = dup2(pfdo[1], 1); |
| 654 | close(pfdo[1]); |
| 655 | |
| 656 | close(pfde[0]); |
| 657 | if(dup_err == -1) |
| 658 | perror("dup2" ); |
| 659 | dup_err = dup2(pfde[1], 2); |
| 660 | close(pfde[1]); |
| 661 | if(dup_err == -1) |
| 662 | perror("dup2" ); |
| 663 | |
| 664 | write_error = write(1, "arguments:" , 10); |
| 665 | for (c = 0; argv[c] != NULL; c++) { |
| 666 | /* very stupid heuristic to make copy/paste easier from |
| 667 | * merovingian's log */ |
| 668 | int q = strchr(argv[c], ' ') != NULL; |
| 669 | write_error |= write(1, " \"" , 1 + q); |
| 670 | write_error |= write(1, argv[c], strlen(argv[c])); |
| 671 | if (q) |
| 672 | write_error |= write(1, "\"" , 1); |
| 673 | } |
| 674 | write_error |= write(1, "\n" , 1); |
| 675 | if (write_error < 0) |
| 676 | perror("write" ); |
| 677 | |
| 678 | execv(_mero_mserver, argv); |
| 679 | /* if the exec returns, it is because of a failure */ |
| 680 | perror("executing failed" ); |
| 681 | exit(1); |
| 682 | } else if (pid > 0) { |
| 683 | /* parent: fine, let's add the pipes for this child */ |
| 684 | dp = _mero_topdp; |
| 685 | while (dp->next != NULL) |
| 686 | dp = dp->next; |
| 687 | dp = dp->next = malloc(sizeof(struct _dpair)); |
| 688 | dp->out = pfdo[0]; |
| 689 | close(pfdo[1]); |
| 690 | dp->err = pfde[0]; |
| 691 | close(pfde[1]); |
| 692 | dp->next = NULL; |
| 693 | dp->type = MERODB; |
| 694 | dp->pid = pid; |
| 695 | dp->dbname = strdup(database); |
| 696 | dp->flag = 0; |
| 697 | |
| 698 | pthread_mutex_unlock(&_mero_topdp_lock); |
| 699 | |
| 700 | /* wait for the child to finish starting, at some point we |
| 701 | * decided that we should wait indefinitely here because if the |
| 702 | * mserver needs time to start up, we shouldn't interrupt it, |
| 703 | * and if it hangs, we're just doomed, with the drawback that we |
| 704 | * completely kill the functionality of monetdbd too */ |
| 705 | do { |
| 706 | /* give the database a break */ |
| 707 | sleep_ms(500); |
| 708 | |
| 709 | /* in the meanwhile, if the server has stopped, it will |
| 710 | * have been removed from the dpair list, so check if |
| 711 | * it's still there. */ |
| 712 | pthread_mutex_lock(&_mero_topdp_lock); |
| 713 | dp = _mero_topdp; |
| 714 | while (dp != NULL && dp->pid != pid) |
| 715 | dp = dp->next; |
| 716 | pthread_mutex_unlock(&_mero_topdp_lock); |
| 717 | |
| 718 | /* stats cannot be NULL, as we don't allow starting non |
| 719 | * existing databases, note that we need to run this loop at |
| 720 | * least once not to leak */ |
| 721 | msab_freeStatus(stats); |
| 722 | er = msab_getStatus(stats, database); |
| 723 | if (er != NULL) { |
| 724 | /* since the client mserver lives its own life anyway, |
| 725 | * it's not really a problem we exit here */ |
| 726 | err e = newErr("%s" , er); |
| 727 | free(er); |
| 728 | pthread_mutex_unlock(&fork_lock); |
| 729 | return(e); |
| 730 | } |
| 731 | |
| 732 | /* server doesn't run, no need to wait any longer */ |
| 733 | if (dp == NULL) |
| 734 | break; |
| 735 | } while ((*stats)->state != SABdbRunning); |
| 736 | |
| 737 | /* check if the SQL scenario was loaded */ |
| 738 | if (dp != NULL && (*stats)->state == SABdbRunning && |
| 739 | (*stats)->conns != NULL && |
| 740 | (*stats)->conns->val != NULL && |
| 741 | (*stats)->scens != NULL && |
| 742 | (*stats)->scens->val != NULL) { |
| 743 | sablist *scen = (*stats)->scens; |
| 744 | do { |
| 745 | if (scen->val != NULL && strcmp(scen->val, "sql" ) == 0) |
| 746 | break; |
| 747 | } while ((scen = scen->next) != NULL); |
| 748 | if (scen == NULL) { |
| 749 | /* we don't know what it's doing, but we don't like it |
| 750 | * any case, so kill it */ |
| 751 | terminateProcess(pid, strdup(database), MERODB, 0); |
| 752 | msab_freeStatus(stats); |
| 753 | pthread_mutex_unlock(&fork_lock); |
| 754 | return(newErr("database '%s' did not initialise the sql " |
| 755 | "scenario" , database)); |
| 756 | } |
| 757 | } else if (dp != NULL) { |
| 758 | terminateProcess(pid, strdup(database), MERODB, 0); |
| 759 | msab_freeStatus(stats); |
| 760 | pthread_mutex_unlock(&fork_lock); |
| 761 | return(newErr( |
| 762 | "database '%s' started up, but failed to open up " |
| 763 | "a communication channel" , database)); |
| 764 | } |
| 765 | |
| 766 | pthread_mutex_unlock(&fork_lock); |
| 767 | |
| 768 | /* try to be clear on why starting the database failed */ |
| 769 | if (dp == NULL) { |
| 770 | state = (*stats)->state; |
| 771 | |
| 772 | /* starting failed */ |
| 773 | msab_freeStatus(stats); |
| 774 | |
| 775 | switch ((int)state) { |
| 776 | case SABdbRunning: |
| 777 | /* right, it's not there, but it's running */ |
| 778 | return(newErr( |
| 779 | "database '%s' has inconsistent state " |
| 780 | "(sabaoth administration reports running, " |
| 781 | "but process seems gone), " |
| 782 | "review monetdbd's " |
| 783 | "logfile (%s) for any peculiarities" , database, |
| 784 | getConfVal(_mero_props, "logfile" ))); |
| 785 | case SABdbCrashed: |
| 786 | return(newErr( |
| 787 | "database '%s' has crashed after starting, " |
| 788 | "manual intervention needed, " |
| 789 | "check monetdbd's logfile (%s) for details" , |
| 790 | database, getConfVal(_mero_props, "logfile" ))); |
| 791 | case SABdbInactive: |
| 792 | return(newErr( |
| 793 | "database '%s' appears to shut " |
| 794 | "itself down after starting, " |
| 795 | "check monetdbd's logfile (%s) for possible " |
| 796 | "hints" , database, |
| 797 | getConfVal(_mero_props, "logfile" ))); |
| 798 | case SABdbStarting: |
| 799 | return(newErr( |
| 800 | "database '%s' has inconsistent state " |
| 801 | "(sabaoth administration reports starting up, " |
| 802 | "but process seems gone), " |
| 803 | "review monetdbd's " |
| 804 | "logfile (%s) for any peculiarities" , database, |
| 805 | getConfVal(_mero_props, "logfile" ))); |
| 806 | default: |
| 807 | return(newErr("unknown state: %d" , (int)state)); |
| 808 | } |
| 809 | } |
| 810 | |
| 811 | if ((*stats)->locked) { |
| 812 | Mfprintf(stdout, "database '%s' has been put into maintenance " |
| 813 | "mode during startup\n" , database); |
| 814 | } |
| 815 | |
| 816 | return(NO_ERR); |
| 817 | } |
| 818 | int e = errno; |
| 819 | pthread_mutex_unlock(&_mero_topdp_lock); |
| 820 | |
| 821 | /* forking failed somehow, cleanup the pipes */ |
| 822 | close(pfdo[0]); |
| 823 | close(pfdo[1]); |
| 824 | close(pfde[0]); |
| 825 | close(pfde[1]); |
| 826 | pthread_mutex_unlock(&fork_lock); |
| 827 | return(newErr("%s" , strerror(e))); |
| 828 | } |
| 829 | |
| 830 | #define BUFLEN 1024 |
| 831 | |
| 832 | /** |
| 833 | * Fork stethoscope and detatch, after performing sanity checks. The assumption |
| 834 | * is that each mserver5 process can have at most one stethoscope process |
| 835 | * attached to it. |
| 836 | */ |
| 837 | err |
| 838 | fork_profiler(char *dbname, sabdb **stats, char **log_path) |
| 839 | { |
| 840 | pid_t pid; |
| 841 | char *error = NO_ERR; |
| 842 | char *pidfilename = NULL; |
| 843 | confkeyval *ckv = NULL, *kv; |
| 844 | size_t pidfnlen; |
| 845 | FILE *pidfile; |
| 846 | char *profiler_executable; |
| 847 | char *beat_frequency = NULL; |
| 848 | char *tmp_exe; |
| 849 | struct stat path_info; |
| 850 | int error_code; |
| 851 | |
| 852 | error = msab_getStatus(stats, dbname); |
| 853 | if (error != NULL) { |
| 854 | return error; |
| 855 | } |
| 856 | |
| 857 | if (*stats == NULL) { |
| 858 | /* TODO: What now? */ |
| 859 | error = newErr("Null stats for db %s" , dbname); |
| 860 | return error; |
| 861 | } |
| 862 | |
| 863 | /* Find the profiler executable. The mserver is running as |
| 864 | * /path/to/installation/mserver5 |
| 865 | * and the profiler executable should be: |
| 866 | * /path/to/installation/stethoscope |
| 867 | */ |
| 868 | tmp_exe = strdup(_mero_mserver); |
| 869 | if (tmp_exe == NULL) { |
| 870 | error = newErr("Cannot find the profiler executable" ); |
| 871 | return error; |
| 872 | } else { |
| 873 | char *server_filename = "mserver5" ; |
| 874 | char *profiler_filename = "stethoscope" ; |
| 875 | char *s = strstr(tmp_exe, server_filename); |
| 876 | size_t executable_len = 0; |
| 877 | |
| 878 | if (s == NULL || strncmp(s, server_filename, strlen(server_filename)) != 0) { |
| 879 | error = newErr("Unexpected executable (missing the string \"%s\")" , server_filename); |
| 880 | free(tmp_exe); |
| 881 | return error; |
| 882 | } |
| 883 | |
| 884 | executable_len = strlen(tmp_exe) + strlen(profiler_filename) - strlen(server_filename) + 1; |
| 885 | *s = '\0'; |
| 886 | profiler_executable = malloc(executable_len); |
| 887 | snprintf(profiler_executable, executable_len, "%s%s%s" , |
| 888 | tmp_exe, profiler_filename, s + 8); |
| 889 | free(tmp_exe); |
| 890 | if (stat(profiler_executable, &path_info) == -1) { |
| 891 | error = newErr("Cannot find profiler executable" ); |
| 892 | goto cleanup; |
| 893 | } |
| 894 | /* free(tmp_exe); */ |
| 895 | } |
| 896 | |
| 897 | pthread_mutex_lock(&fork_lock); |
| 898 | |
| 899 | /* Verify that the requested db is running */ |
| 900 | if ((*stats)->state != SABdbRunning) { |
| 901 | /* server is not running, shoo */ |
| 902 | error = newErr("Database %s is not running." , dbname); |
| 903 | goto cleanup; |
| 904 | } |
| 905 | |
| 906 | /* find the path that the profiler will be storing files */ |
| 907 | ckv = getDefaultProps(); |
| 908 | readAllProps(ckv, (*stats)->path); |
| 909 | kv = findConfKey(ckv, PROFILERBEATFREQ); |
| 910 | if (kv) { |
| 911 | beat_frequency = kv->val; |
| 912 | } |
| 913 | kv = findConfKey(ckv, PROFILERLOGPROPERTY); |
| 914 | |
| 915 | if (kv == NULL) { |
| 916 | error = newErr("Property '" PROFILERLOGPROPERTY"' not set for db %s\n" , |
| 917 | dbname); |
| 918 | goto cleanup; |
| 919 | } |
| 920 | |
| 921 | *log_path = strdup(kv->val); |
| 922 | |
| 923 | /* Check that the log_path exists and create it if it does not */ |
| 924 | error_code = stat(*log_path, &path_info); |
| 925 | if (error_code == -1) { /* stat failed */ |
| 926 | if (errno == ENOENT) { /* dir does not exist, create it */ |
| 927 | mode_t mode = 0755; |
| 928 | if (mkdir(*log_path, mode) == -1) { /* mkdir failed, bail out */ |
| 929 | char error_message[BUFSIZ]; |
| 930 | if (strerror_r(errno, error_message, BUFSIZ) != 0) |
| 931 | strcpy(error_message, "unknown error" ); |
| 932 | error = newErr("%s" , error_message); |
| 933 | free(*log_path); |
| 934 | *log_path = NULL; |
| 935 | goto cleanup; |
| 936 | } |
| 937 | } else { /* Something else went wrong, can't handle the heat */ |
| 938 | char error_message[BUFSIZ]; |
| 939 | if (strerror_r(errno, error_message, BUFSIZ) != 0) |
| 940 | strcpy(error_message, "unknown error" ); |
| 941 | error = newErr("%s" , error_message); |
| 942 | free(*log_path); |
| 943 | *log_path = NULL; |
| 944 | goto cleanup; |
| 945 | } |
| 946 | } else { /* stat succeeded */ |
| 947 | if(!S_ISDIR(path_info.st_mode)) { /* file exists but is not a directory, bail out */ |
| 948 | error = newErr("File %s exists but is not a directory." , *log_path); |
| 949 | free(*log_path); |
| 950 | *log_path = NULL; |
| 951 | goto cleanup; |
| 952 | } |
| 953 | } |
| 954 | |
| 955 | /* construct the filename of the pid file */ |
| 956 | pidfnlen = strlen(*log_path) + strlen("/profiler.pid" ) + 1; |
| 957 | pidfilename = malloc(pidfnlen); |
| 958 | if (pidfilename == NULL) { |
| 959 | error = newErr("Cannot allocate buffer while starting profiler" ); |
| 960 | goto cleanup; |
| 961 | } |
| 962 | snprintf(pidfilename, pidfnlen, "%s/profiler.pid" , *log_path); |
| 963 | |
| 964 | /* Make sure another instance of stethoscope is not running. */ |
| 965 | error_code = stat(pidfilename, &path_info); |
| 966 | if (error_code != -1) { |
| 967 | char buf[8]; |
| 968 | long pid; |
| 969 | /* The pid file exists. See if a process with this pid exists, |
| 970 | * and if yes, if it is stethoscope. |
| 971 | */ |
| 972 | |
| 973 | /* We cannot open the pidfile, bail out */ |
| 974 | if ((pidfile = fopen(pidfilename, "r" )) == NULL) { |
| 975 | error = newErr("pid file %s already exists, but is not accessible. Is the profiler already running?" , |
| 976 | pidfilename); |
| 977 | free(*log_path); |
| 978 | *log_path = NULL; |
| 979 | goto cleanup; |
| 980 | } |
| 981 | |
| 982 | if (fgets(buf, sizeof(buf), pidfile) == NULL) { |
| 983 | error = newErr("cannot read from pid file %s: %s\n" , |
| 984 | pidfilename, strerror(errno)); |
| 985 | fclose(pidfile); |
| 986 | free(*log_path); |
| 987 | *log_path = NULL; |
| 988 | goto cleanup; |
| 989 | } |
| 990 | fclose(pidfile); |
| 991 | |
| 992 | /* Verify that what we read is actually a number */ |
| 993 | errno = 0; |
| 994 | pid = strtol(buf, NULL, 10); |
| 995 | if (errno != 0) { |
| 996 | error = newErr("contents of the pid file are not correct: %s\n" , |
| 997 | strerror(errno)); |
| 998 | free(*log_path); |
| 999 | *log_path = NULL; |
| 1000 | goto cleanup; |
| 1001 | } |
| 1002 | |
| 1003 | // Open /proc/<pid>/comm and compare the contents to "stethoscope" |
| 1004 | // This of course is specific to Linux |
| 1005 | size_t fn_size = strlen("/proc/comm" ) + 9; |
| 1006 | char *filename = malloc(fn_size); |
| 1007 | if (filename == NULL) { |
| 1008 | error = newErr("cannot allocate %zu bytes: %s\n" , |
| 1009 | fn_size, strerror(errno)); |
| 1010 | free(*log_path); |
| 1011 | *log_path = NULL; |
| 1012 | goto cleanup; |
| 1013 | } |
| 1014 | snprintf(filename, fn_size, "/proc/%ld/comm" , pid); |
| 1015 | |
| 1016 | FILE *comm = fopen(filename, "r" ); |
| 1017 | if (comm == NULL) { |
| 1018 | /* We cannot open the file for the process with the specified pid, |
| 1019 | * so the process is not running. |
| 1020 | */ |
| 1021 | free(filename); |
| 1022 | goto startup; |
| 1023 | } |
| 1024 | char buf2[BUFLEN]; |
| 1025 | size_t len = fread(buf2, 1, BUFLEN, comm); |
| 1026 | |
| 1027 | if(ferror(comm)) { |
| 1028 | error = newErr("cannot read from file %s\n" , filename); |
| 1029 | free(filename); |
| 1030 | free(*log_path); |
| 1031 | fclose(comm); |
| 1032 | *log_path = NULL; |
| 1033 | goto cleanup; |
| 1034 | } |
| 1035 | if (len == BUFLEN) |
| 1036 | len--; |
| 1037 | buf2[len] = 0; |
| 1038 | |
| 1039 | char expected_command[] = "stethoscope" ; |
| 1040 | size_t command_len = strlen(expected_command); |
| 1041 | if (strncmp(buf2, expected_command, command_len) == 0) { |
| 1042 | error = newErr("profiler already running for %s\n" , dbname); |
| 1043 | free(filename); |
| 1044 | free(*log_path); |
| 1045 | fclose(comm); |
| 1046 | *log_path = NULL; |
| 1047 | goto cleanup; |
| 1048 | } |
| 1049 | |
| 1050 | fclose(comm); |
| 1051 | free(filename); |
| 1052 | } |
| 1053 | |
| 1054 | startup: |
| 1055 | /* Open the pid file */ |
| 1056 | if ((pidfile = fopen(pidfilename, "w" )) == NULL) { |
| 1057 | error = newErr("unable to open %s for writing" , pidfilename); |
| 1058 | free(*log_path); |
| 1059 | *log_path = NULL; |
| 1060 | goto cleanup; |
| 1061 | } |
| 1062 | |
| 1063 | pid = fork(); |
| 1064 | if (pid == 0) { |
| 1065 | char timestamp[20]; |
| 1066 | char *argv[512]; |
| 1067 | int arg_idx = 0; |
| 1068 | size_t log_filename_len; |
| 1069 | char *log_filename; |
| 1070 | time_t current_time; |
| 1071 | struct tm *tm_ctime; |
| 1072 | |
| 1073 | fclose(pidfile); |
| 1074 | /* construct the log output file */ |
| 1075 | log_filename_len = strlen(*log_path) + strlen("/proflog_" ) + strlen(dbname) + 26; |
| 1076 | log_filename = malloc(log_filename_len); |
| 1077 | if (log_filename == NULL) { |
| 1078 | /* TODO What now? */ |
| 1079 | Mfprintf(stderr, "failed to allocate buffer\n" ); |
| 1080 | exit(1); |
| 1081 | } |
| 1082 | current_time = time(NULL); |
| 1083 | tm_ctime = localtime(¤t_time); |
| 1084 | strftime(timestamp, sizeof(timestamp), "%Y-%m-%d_%H:%M:%S" , tm_ctime); |
| 1085 | snprintf(log_filename, log_filename_len, "%s/proflog_%s_%s.json" , |
| 1086 | *log_path, dbname, timestamp); |
| 1087 | |
| 1088 | /* build the arguments */ |
| 1089 | argv[arg_idx++] = profiler_executable; |
| 1090 | argv[arg_idx++] = "-j" ; /* JSON output */ |
| 1091 | argv[arg_idx++] = "-d" ; |
| 1092 | argv[arg_idx++] = dbname; |
| 1093 | if (beat_frequency) { |
| 1094 | argv[arg_idx++] = "-b" ; |
| 1095 | argv[arg_idx++] = beat_frequency; |
| 1096 | } |
| 1097 | argv[arg_idx++] = "-o" ; |
| 1098 | argv[arg_idx++] = log_filename; |
| 1099 | /* execute */ |
| 1100 | execv(profiler_executable, argv); |
| 1101 | exit(1); |
| 1102 | } else { |
| 1103 | /* write pid of stethoscope */ |
| 1104 | Mfprintf(pidfile, "%d" , (int)pid); |
| 1105 | fclose(pidfile); |
| 1106 | } |
| 1107 | |
| 1108 | cleanup: |
| 1109 | freeConfFile(ckv); |
| 1110 | free(ckv); |
| 1111 | free(profiler_executable); |
| 1112 | free(pidfilename); |
| 1113 | pthread_mutex_unlock(&fork_lock); |
| 1114 | return error; |
| 1115 | } |
| 1116 | |
| 1117 | err |
| 1118 | shutdown_profiler(char *dbname, sabdb **stats) |
| 1119 | { |
| 1120 | err error=NO_ERR; |
| 1121 | confkeyval *ckv = NULL, *kv; |
| 1122 | size_t pidfnlen = 0; |
| 1123 | char *pidfilename = NULL; |
| 1124 | FILE *pidfile; |
| 1125 | char buf[BUFSIZ]; |
| 1126 | size_t nbytes; |
| 1127 | pid_t pid; |
| 1128 | |
| 1129 | error = msab_getStatus(stats, dbname); |
| 1130 | if (error != NULL) { |
| 1131 | return error; |
| 1132 | } |
| 1133 | |
| 1134 | if (*stats == NULL) { |
| 1135 | /* TODO: What now? */ |
| 1136 | error = newErr("Null stats for db %s" , dbname); |
| 1137 | return error; |
| 1138 | } |
| 1139 | |
| 1140 | /* Verify that the requested db is running */ |
| 1141 | if ((*stats)->state != SABdbRunning) { |
| 1142 | /* server is not running, shoo */ |
| 1143 | error = newErr("Database %s is not running." , dbname); |
| 1144 | goto cleanup; |
| 1145 | } |
| 1146 | |
| 1147 | /* Find the pid file and make sure the profiler is running */ |
| 1148 | ckv = getDefaultProps(); |
| 1149 | readAllProps(ckv, (*stats)->path); |
| 1150 | kv = findConfKey(ckv, PROFILERLOGPROPERTY); |
| 1151 | |
| 1152 | if (kv == NULL) { |
| 1153 | error = newErr("Property '" PROFILERLOGPROPERTY"' not set for db %s\n" , |
| 1154 | dbname); |
| 1155 | goto cleanup; |
| 1156 | } |
| 1157 | |
| 1158 | /* construct the filename of the pid file */ |
| 1159 | pidfnlen = strlen(kv->val) + strlen("/profiler.pid" ) + 1; |
| 1160 | pidfilename = malloc(pidfnlen); |
| 1161 | if (pidfilename == NULL) { |
| 1162 | error = newErr("Cannot allocate buffer while shutting down of profiler" ); |
| 1163 | goto cleanup; |
| 1164 | } |
| 1165 | snprintf(pidfilename, pidfnlen, "%s/profiler.pid" , kv->val); |
| 1166 | |
| 1167 | if ((pidfile = fopen(pidfilename, "r" )) == NULL) { |
| 1168 | error = newErr("Unable to open %s for reading" , pidfilename); |
| 1169 | goto cleanup; |
| 1170 | } |
| 1171 | |
| 1172 | clearerr(pidfile); |
| 1173 | nbytes = fread(buf, 1, BUFSIZ, pidfile); |
| 1174 | |
| 1175 | if (ferror(pidfile)) { |
| 1176 | error = newErr("Cannot read pid (%s) from file %s" , buf, pidfilename); |
| 1177 | fclose(pidfile); |
| 1178 | goto cleanup; |
| 1179 | } |
| 1180 | fclose(pidfile); |
| 1181 | |
| 1182 | if (buf[nbytes - 1] == '\n') { |
| 1183 | buf[nbytes - 1] = '\0'; |
| 1184 | } |
| 1185 | |
| 1186 | pid = (pid_t)strtol(buf, NULL, 10); |
| 1187 | if (pid == 0 && errno == EINVAL) { |
| 1188 | error = newErr("File contents %s not a valid pid" , buf); |
| 1189 | goto cleanup; |
| 1190 | } |
| 1191 | |
| 1192 | if (kill(pid, SIGTERM) == -1) { |
| 1193 | char error_message[BUFSIZ]; |
| 1194 | if (strerror_r(errno, error_message, BUFSIZ) != 0) |
| 1195 | strcpy(error_message, "unknown error" ); |
| 1196 | error = newErr("%s" , error_message); |
| 1197 | goto cleanup; |
| 1198 | } |
| 1199 | |
| 1200 | /* All went well. Remove the pidfile */ |
| 1201 | if (remove(pidfilename) != 0) { |
| 1202 | error = newErr("Profiler seems to have stopped, but cannot remove pid file." ); |
| 1203 | } |
| 1204 | |
| 1205 | cleanup: |
| 1206 | freeConfFile(ckv); |
| 1207 | free(pidfilename); |
| 1208 | return error; |
| 1209 | } |
| 1210 | |
| 1211 | /* vim:set ts=4 sw=4 noexpandtab: */ |
| 1212 | |