ctdb/utils/pmda/pmda_ctdb.c

   1 /*
   2  * CTDB Performance Metrics Domain Agent (PMDA) for Performance Co-Pilot (PCP)
   3  *
   4  * Copyright (c) 1995,2004 Silicon Graphics, Inc.  All Rights Reserved.
   5  * Copyright (c) 2011 David Disseldorp
   6  *
   7  * This program is free software; you can redistribute it and/or modify it
   8  * under the terms of the GNU General Public License as published by the
   9  * Free Software Foundation; either version 2 of the License, or (at your
  10  * option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  14  * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15  * for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License along
  18  * with this program; if not, write to the Free Software Foundation, Inc.,
  19  * 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  20  */
  21
  22 #include "replace.h"
  23 #include "system/network.h"
  24
  25 #include <talloc.h>
  26 #include <tevent.h>
  27 #include <tdb.h>
  28
  29 #include "lib/util/time.h"
  30 #include "lib/util/blocking.h"
  31
  32 #include "client/client.h"
  33 #include "client/client_sync.h"
  34
  35 #include <pcp/pmapi.h>
  36 #include <pcp/pmda.h>
  37
  38 #ifdef HAVE___PMID_INT
  39 #include <pcp/impl.h>
  40
  41 #define pmID_cluster(id)        id->cluster
  42 #define pmID_item(id)           id->item
  43 #define pmGetProgname()         pmProgname
  44 #define pmSetProgname(a)        __pmSetProgname(a)
  45 #endif
  46
  47 #include "domain.h"
  48
  49 /*
  50  * CTDB PMDA
  51  *
  52  * This PMDA connects to the locally running ctdbd daemon and pulls
  53  * statistics for export via PCP. The ctdbd Unix domain socket path can be
  54  * specified with the CTDB_SOCKET environment variable, otherwise the default
  55  * path is used.
  56  */
  57
  58 /*
  59  * All metrics supported in this PMDA - one table entry for each.
  60  * The 4th field specifies the serial number of the instance domain
  61  * for the metric, and must be either PM_INDOM_NULL (denoting a
  62  * metric that only ever has a single value), or the serial number
  63  * of one of the instance domains declared in the instance domain table
  64  * (i.e. in indomtab, above).
  65  */
  66 static pmdaMetric metrictab[] = {
  67         /* num_clients */
  68         { NULL, { PMDA_PMID(0,0), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT,
  69                 PMDA_PMUNITS(0,0,0,0,0,0) }, },
  70         /* frozen */
  71         { NULL, { PMDA_PMID(0,1), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT,
  72                 PMDA_PMUNITS(0,0,0,0,0,0) }, },
  73         /* recovering */
  74         { NULL, { PMDA_PMID(0,2), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT,
  75                 PMDA_PMUNITS(0,0,0,0,0,0) }, },
  76         /* client_packets_sent */
  77         { NULL, { PMDA_PMID(0,3), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
  78                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
  79         /* client_packets_recv */
  80         { NULL, { PMDA_PMID(0,4), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
  81                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
  82         /* node_packets_sent */
  83         { NULL, { PMDA_PMID(0,5), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
  84                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
  85         /* node_packets_recv */
  86         { NULL, { PMDA_PMID(0,6), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
  87                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
  88         /* keepalive_packets_sent */
  89         { NULL, { PMDA_PMID(0,7), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
  90                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
  91         /* keepalive_packets_recv */
  92         { NULL, { PMDA_PMID(0,8), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
  93                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
  94         /* req_call */
  95         { NULL, { PMDA_PMID(1,0), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
  96                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
  97         /* reply_call */
  98         { NULL, { PMDA_PMID(1,1), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
  99                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 100         /* req_dmaster */
 101         { NULL, { PMDA_PMID(1,2), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 102                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 103         /* reply_dmaster */
 104         { NULL, { PMDA_PMID(1,3), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 105                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 106         /* reply_error */
 107         { NULL, { PMDA_PMID(1,4), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 108                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 109         /* req_message */
 110         { NULL, { PMDA_PMID(1,5), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 111                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 112         /* req_control */
 113         { NULL, { PMDA_PMID(1,6), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 114                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 115         /* reply_control */
 116         { NULL, { PMDA_PMID(1,7), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 117                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 118         /* req_call */
 119         { NULL, { PMDA_PMID(2,0), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 120                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 121         /* req_message */
 122         { NULL, { PMDA_PMID(2,1), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 123                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 124         /* req_control */
 125         { NULL, { PMDA_PMID(2,2), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 126                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 127         /* call */
 128         { NULL, { PMDA_PMID(3,0), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 129                 PMDA_PMUNITS(0,0,1,0,0,0) }, },
 130         /* control */
 131         { NULL, { PMDA_PMID(3,1), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 132                 PMDA_PMUNITS(0,0,1,0,0,0) }, },
 133         /* traverse */
 134         { NULL, { PMDA_PMID(3,2), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 135                 PMDA_PMUNITS(0,0,1,0,0,0) }, },
 136         /* total_calls */
 137         { NULL, { PMDA_PMID(0,9), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 138                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 139         /* pending_calls */
 140         { NULL, { PMDA_PMID(0,10), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT,
 141                 PMDA_PMUNITS(0,0,0,0,0,0) }, },
 142         /* locks.num_calls */
 143         { NULL, { PMDA_PMID(0,11), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 144                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 145         /* locks.num_pending */
 146         { NULL, { PMDA_PMID(0,12), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT,
 147                 PMDA_PMUNITS(0,0,0,0,0,0) }, },
 148         /* childwrite_calls */
 149         { NULL, { PMDA_PMID(0,13), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_COUNTER,
 150                 PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 151         /* pending_childwrite_calls */
 152         { NULL, { PMDA_PMID(0,14), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT,
 153                 PMDA_PMUNITS(0,0,0,0,0,0) }, },
 154         /* memory_used */
 155         { NULL, { PMDA_PMID(0,15), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT,
 156                 PMDA_PMUNITS(1,0,0,PM_SPACE_BYTE,0,0) }, },
 157         /* max_hop_count */
 158         { NULL, { PMDA_PMID(0,16), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT,
 159                 PMDA_PMUNITS(0,0,0,0,0,0) }, },
 160         /* reclock.ctdbd.max */
 161         { NULL, { PMDA_PMID(0,17), PM_TYPE_DOUBLE, PM_INDOM_NULL, PM_SEM_INSTANT,
 162                 PMDA_PMUNITS(0,1,0,0,PM_TIME_SEC,0) }, },
 163         /* reclock.recd.max */
 164         { NULL, { PMDA_PMID(0,18), PM_TYPE_DOUBLE, PM_INDOM_NULL, PM_SEM_INSTANT,
 165                 PMDA_PMUNITS(0,1,0,0,PM_TIME_SEC,0) }, },
 166         /* call_latency.max */
 167         { NULL, { PMDA_PMID(0,19), PM_TYPE_DOUBLE, PM_INDOM_NULL, PM_SEM_INSTANT,
 168                 PMDA_PMUNITS(0,1,0,0,PM_TIME_SEC,0) }, },
 169         /* locks.latency.max */
 170         { NULL, { PMDA_PMID(0,20), PM_TYPE_DOUBLE, PM_INDOM_NULL, PM_SEM_INSTANT,
 171                 PMDA_PMUNITS(0,1,0,0,PM_TIME_SEC,0) }, },
 172         /* childwrite_latency.max */
 173         { NULL, { PMDA_PMID(0,21), PM_TYPE_DOUBLE, PM_INDOM_NULL, PM_SEM_INSTANT,
 174                 PMDA_PMUNITS(0,1,0,0,PM_TIME_SEC,0) }, },
 175         /* num_recoveries */
 176         { NULL, { PMDA_PMID(0,22), PM_TYPE_U32, PM_INDOM_NULL, PM_SEM_INSTANT,
 177                 PMDA_PMUNITS(0,0,0,0,0,0) }, },
 178 };
 179
 180 static struct tevent_context *ev;
 181 static struct ctdb_client_context *client;
 182 static struct ctdb_statistics *stats;
 183
 184 static void
 185 pmda_ctdb_disconnected(void *args)
 186 {
 187         fprintf(stderr, "ctdbd unreachable\n");
 188         TALLOC_FREE(client);
 189 }
 190
 191
 192 static int
 193 pmda_ctdb_daemon_connect(void)
 194 {
 195         const char *socket_name;
 196         int ret;
 197
 198         ev = tevent_context_init(NULL);
 199         if (ev == NULL) {
 200                 fprintf(stderr, "Failed to init event ctx\n");
 201                 return -1;
 202         }
 203
 204         socket_name = getenv("CTDB_SOCKET");
 205         if (socket_name == NULL) {
 206                 socket_name = CTDB_SOCKET;
 207         }
 208
 209         ret = ctdb_client_init(ev, ev, socket_name, &client);
 210         if (ret != 0) {
 211                 fprintf(stderr, "Failed to connect to ctdb daemon via %s\n",
 212                         socket_name);
 213                 goto err_ev;
 214         }
 215
 216         ctdb_client_set_disconnect_callback(client, pmda_ctdb_disconnected,
 217                                             NULL);
 218
 219         return 0;
 220
 221 err_ev:
 222         talloc_free(ev);
 223         client = NULL;
 224         return -1;
 225 }
 226
 227 static void
 228 pmda_ctdb_daemon_disconnect(void)
 229 {
 230         TALLOC_FREE(client);
 231         talloc_free(ev);
 232 }
 233
 234 static int
 235 fill_base(unsigned int item, pmAtomValue *atom)
 236 {
 237         switch (item) {
 238         case 0:
 239                 atom->ul = stats->num_clients;
 240                 break;
 241         case 1:
 242                 atom->ul = stats->frozen;
 243                 break;
 244         case 2:
 245                 atom->ul = stats->recovering;
 246                 break;
 247         case 3:
 248                 atom->ul = stats->client_packets_sent;
 249                 break;
 250         case 4:
 251                 atom->ul = stats->client_packets_recv;
 252                 break;
 253         case 5:
 254                 atom->ul = stats->node_packets_sent;
 255                 break;
 256         case 6:
 257                 atom->ul = stats->node_packets_recv;
 258                 break;
 259         case 7:
 260                 atom->ul = stats->keepalive_packets_sent;
 261                 break;
 262         case 8:
 263                 atom->ul = stats->keepalive_packets_recv;
 264                 break;
 265         case 9:
 266                 atom->ul = stats->total_calls;
 267                 break;
 268         case 10:
 269                 atom->ul = stats->pending_calls;
 270                 break;
 271         case 11:
 272                 atom->ul = stats->locks.num_calls;
 273                 break;
 274         case 12:
 275                 atom->ul = stats->locks.num_pending;
 276                 break;
 277         case 13:
 278                 atom->ul = stats->childwrite_calls;
 279                 break;
 280         case 14:
 281                 atom->ul = stats->pending_childwrite_calls;
 282                 break;
 283         case 15:
 284                 atom->ul = stats->memory_used;
 285                 break;
 286         case 16:
 287                 atom->ul = stats->max_hop_count;
 288                 break;
 289         case 17:
 290                 atom->d = stats->reclock.ctdbd.max;
 291                 break;
 292         case 18:
 293                 atom->d = stats->reclock.recd.max;
 294                 break;
 295         case 19:
 296                 atom->d = stats->call_latency.max;
 297                 break;
 298         case 20:
 299                 atom->d = stats->locks.latency.max;
 300                 break;
 301         case 21:
 302                 atom->d = stats->childwrite_latency.max;
 303                 break;
 304         case 22:
 305                 atom->ul = stats->num_recoveries;
 306                 break;
 307         default:
 308                 return PM_ERR_PMID;
 309         }
 310
 311         return 0;
 312 }
 313
 314 static int
 315 fill_node(unsigned int item, pmAtomValue *atom)
 316 {
 317         switch (item) {
 318         case 0:
 319                atom->ul = stats->node.req_call;
 320                break;
 321         case 1:
 322                atom->ul = stats->node.reply_call;
 323                break;
 324         case 2:
 325                atom->ul = stats->node.req_dmaster;
 326                break;
 327         case 3:
 328                atom->ul = stats->node.reply_dmaster;
 329                break;
 330         case 4:
 331                atom->ul = stats->node.reply_error;
 332                break;
 333         case 5:
 334                atom->ul = stats->node.req_message;
 335                break;
 336         case 6:
 337                atom->ul = stats->node.req_control;
 338                break;
 339         case 7:
 340                 atom->ul = stats->node.reply_control;
 341                 break;
 342         default:
 343                 return PM_ERR_PMID;
 344         }
 345
 346         return 0;
 347 }
 348
 349
 350 static int
 351 fill_client(unsigned int item, pmAtomValue *atom)
 352 {
 353         switch (item) {
 354         case 0:
 355                 atom->ul = stats->client.req_call;
 356                 break;
 357         case 1:
 358                 atom->ul = stats->client.req_message;
 359                 break;
 360         case 2:
 361                 atom->ul = stats->client.req_control;
 362                 break;
 363         default:
 364                 return PM_ERR_PMID;
 365         }
 366
 367         return 0;
 368 }
 369
 370 static int
 371 fill_timeout(unsigned int item, pmAtomValue *atom)
 372 {
 373         switch (item) {
 374         case 0:
 375                 atom->ul = stats->timeouts.call;
 376                 break;
 377         case 1:
 378                 atom->ul = stats->timeouts.control;
 379                 break;
 380         case 2:
 381                 atom->ul = stats->timeouts.traverse;
 382                 break;
 383         default:
 384                 return PM_ERR_PMID;
 385         }
 386
 387         return 0;
 388 }
 389
 390 /*
 391  * callback provided to pmdaFetch
 392  */
 393 static int
 394 pmda_ctdb_fetch_cb(pmdaMetric *mdesc, unsigned int inst, pmAtomValue *atom)
 395 {
 396         int ret;
 397 #ifdef HAVE___PMID_INT
 398         __pmID_int *id = (__pmID_int *)&(mdesc->m_desc.pmid);
 399 #else
 400         pmID id = *(pmID *)&(mdesc->m_desc.pmid);
 401 #endif
 402
 403         if (inst != PM_IN_NULL) {
 404                 return PM_ERR_INST;
 405         }
 406
 407         if (stats == NULL) {
 408                 fprintf(stderr, "stats not available\n");
 409                 ret = PM_ERR_VALUE;
 410                 goto err_out;
 411         }
 412
 413
 414         switch (pmID_cluster(id)) {
 415         case 0:
 416                 ret = fill_base(pmID_item(id), atom);
 417                 if (ret) {
 418                         goto err_out;
 419                 }
 420                 break;
 421         case 1:
 422                 ret = fill_node(pmID_item(id), atom);
 423                 if (ret) {
 424                         goto err_out;
 425                 }
 426                 break;
 427         case 2:
 428                 ret = fill_client(pmID_item(id), atom);
 429                 if (ret) {
 430                         goto err_out;
 431                 }
 432                 break;
 433         case 3:
 434                 ret = fill_timeout(pmID_item(id), atom);
 435                 if (ret) {
 436                         goto err_out;
 437                 }
 438                 break;
 439         default:
 440                 return PM_ERR_PMID;
 441         }
 442
 443         ret = 0;
 444 err_out:
 445         return ret;
 446 }
 447
 448 /*
 449  * This routine is called once for each pmFetch(3) operation, so is a
 450  * good place to do once-per-fetch functions, such as value caching or
 451  * instance domain evaluation.
 452  */
 453 static int
 454 pmda_ctdb_fetch(int numpmid, pmID pmidlist[], pmResult **resp, pmdaExt *pmda)
 455 {
 456         int ret;
 457
 458         if (client == NULL) {
 459                 fprintf(stderr, "attempting reconnect to ctdbd\n");
 460                 ret = pmda_ctdb_daemon_connect();
 461                 if (ret < 0) {
 462                         fprintf(stderr, "reconnect failed\n");
 463                         return PM_ERR_VALUE;
 464                 }
 465         }
 466
 467         ret = ctdb_ctrl_statistics(client, ev, client, CTDB_CURRENT_NODE,
 468                                    tevent_timeval_current_ofs(1,0), &stats);
 469         if (ret != 0) {
 470                 fprintf(stderr, "ctdb control for statistics failed, reconnecting\n");
 471                 pmda_ctdb_daemon_disconnect();
 472                 ret = PM_ERR_VALUE;
 473                 goto err_out;
 474         }
 475
 476         ret = pmdaFetch(numpmid, pmidlist, resp, pmda);
 477
 478         talloc_free(stats);
 479 err_out:
 480         return ret;
 481 }
 482
 483 void pmda_ctdb_init(pmdaInterface *dp);
 484
 485 /*
 486  * Initialise the agent
 487  */
 488 void
 489 pmda_ctdb_init(pmdaInterface *dp)
 490 {
 491         if (dp->status != 0) {
 492                 return;
 493         }
 494
 495         dp->version.two.fetch = pmda_ctdb_fetch;
 496         pmdaSetFetchCallBack(dp, pmda_ctdb_fetch_cb);
 497
 498         pmdaInit(dp, NULL, 0, metrictab,
 499                  (sizeof(metrictab) / sizeof(metrictab[0])));
 500 }
 501
 502 static char *
 503 helpfile(void)
 504 {
 505         static char buf[MAXPATHLEN];
 506
 507         if (!buf[0]) {
 508                 snprintf(buf, sizeof(buf), "%s/ctdb/help",
 509                          pmGetConfig("PCP_PMDAS_DIR"));
 510         }
 511         return buf;
 512 }
 513
 514 static void
 515 usage(void)
 516 {
 517         fprintf(stderr, "Usage: %s [options]\n\n", pmGetProgname());
 518         fputs("Options:\n"
 519           "  -d domain        use domain (numeric) for metrics domain of PMDA\n"
 520           "  -l logfile       write log into logfile rather than using default log name\n"
 521           "\nExactly one of the following options may appear:\n"
 522           "  -i port          expect PMCD to connect on given inet port (number or name)\n"
 523           "  -p               expect PMCD to supply stdin/stdout (pipe)\n"
 524           "  -u socket        expect PMCD to connect on given unix domain socket\n",
 525           stderr);
 526         exit(1);
 527 }
 528
 529 /*
 530  * Set up the agent if running as a daemon.
 531  */
 532 int
 533 main(int argc, char **argv)
 534 {
 535         int err = 0;
 536         char log_file[] = "pmda_ctdb.log";
 537         pmdaInterface dispatch;
 538
 539         pmSetProgname(argv[0]);
 540
 541         pmdaDaemon(&dispatch, PMDA_INTERFACE_2, argv[0], CTDB,
 542                    log_file, helpfile());
 543
 544         if (pmdaGetOpt(argc, argv, "d:i:l:pu:?", &dispatch, &err) != EOF) {
 545                 err++;
 546         }
 547
 548         if (err) {
 549                 usage();
 550         }
 551
 552         pmdaOpenLog(&dispatch);
 553         pmda_ctdb_init(&dispatch);
 554         pmdaConnect(&dispatch);
 555         pmdaMain(&dispatch);
 556
 557         exit(0);
 558 }
 559