@cryptotaxi247 / netdata-1 / commits / da4641b30

health now properly tracks alarm transitions; code cleanup;

Costa Tsaousis committed Aug 15, 2016 at 16:55 UTC da4641b30cf5670592a9f17790b0369d3ac57ccb

4 files changed +356 -418

src/eval.c

@@ -845,6 +845,8 @@ int expression_evaluate(EVAL_EXPRESSION *exp) {

845 exp->result = eval_node(exp, (EVAL_NODE *)exp->nodes, &exp->error);

846

847 if(exp->error != EVAL_ERROR_OK) {

848 + exp->result = NAN;

849 +

850 if(buffer_strlen(exp->error_msg))

851 buffer_strcat(exp->error_msg, "; ");

852

src/health.c

+332 -390

@@ -1,6 +1,6 @@

1 #include "common.h"

3 -#define RRDSETVAR_ID_MAX 1024

3 +#define RRDVAR_MAX_LENGTH 1024

5 static const char *health_default_exec = PLUGINS_DIR "/alarm.sh";

6 int health_enabled = 1;

@@ -8,6 +8,20 @@ int health_enabled = 1;

8 // ----------------------------------------------------------------------------

9 // RRDVAR management

11 +static inline int rrdvar_fix_name(char *variable) {

12 + int fixed = 0;

13 + while(*variable) {

14 + if (!isalnum(*variable) && *variable != '.' && *variable != '_') {

15 + *variable++ = '_';

16 + fixed++;

17 + }

18 + else

19 + variable++;

20 + }

21 +

22 + return fixed;

23 +}

24 +

25 int rrdvar_compare(void* a, void* b) {

26 if(((RRDVAR *)a)->hash < ((RRDVAR *)b)->hash) return -1;

27 else if(((RRDVAR *)a)->hash > ((RRDVAR *)b)->hash) return 1;

@@ -25,7 +39,7 @@ static inline RRDVAR *rrdvar_index_add(avl_tree_lock *tree, RRDVAR *rv) {

39 static inline RRDVAR *rrdvar_index_del(avl_tree_lock *tree, RRDVAR *rv) {

40 RRDVAR *ret = (RRDVAR *)avl_remove_lock(tree, (avl *)(rv));

41 if(!ret)

28 - fatal("Request to remove RRDVAR '%s' from index failed. Not Found.", rv->name);

42 + error("Request to remove RRDVAR '%s' from index failed. Not Found.", rv->name);

44 return ret;

45 }

@@ -38,45 +52,45 @@ static inline RRDVAR *rrdvar_index_find(avl_tree_lock *tree, const char *name, u

52 return (RRDVAR *)avl_search_lock(tree, (avl *)&tmp);

53 }

41 -static inline RRDVAR *rrdvar_create(const char *name, uint32_t hash, int type, calculated_number *value) {

42 - RRDVAR *rv = callocz(1, sizeof(RRDVAR));

55 +static inline void rrdvar_free(RRDHOST *host, avl_tree_lock *tree, RRDVAR *rv) {

56 + (void)host;

44 - rv->name = (char *)name;

45 - rv->hash = (hash)?hash:simple_hash((rv->name));

46 - rv->type = type;

47 - rv->value = value;

48 -

49 - return rv;

50 -}

58 + if(!rv) return;

52 -static inline void rrdvar_free(RRDHOST *host, RRDVAR *rv) {

53 - if(host) {

54 - // FIXME: we may need some kind of locking here

55 - EVAL_VARIABLE *rf;

56 - for (rf = host->references; rf; rf = rf->next)

57 - if (rf->rrdvar == rv) rf->rrdvar = NULL;

58 - }

60 + if(tree)

61 + rrdvar_index_del(tree, rv);

63 + freez(rv->name);

64 freez(rv);

65 }

63 -static inline RRDVAR *rrdvar_create_and_index(const char *scope, avl_tree_lock *tree, const char *name, uint32_t hash, int type, calculated_number *value) {

64 - RRDVAR *rv = rrdvar_index_find(tree, name, hash);

67 +static inline RRDVAR *rrdvar_create_and_index(const char *scope, avl_tree_lock *tree, const char *name, int type, calculated_number *value) {

68 + char *variable = strdupz(name);

69 + rrdvar_fix_name(variable);

70 + uint32_t hash = simple_hash(variable);

71 +

72 + RRDVAR *rv = rrdvar_index_find(tree, variable, hash);

73 if(unlikely(!rv)) {

66 - debug(D_VARIABLES, "Variable '%s' not found in scope '%s'. Creating a new one.", name, scope);

74 + debug(D_VARIABLES, "Variable '%s' not found in scope '%s'. Creating a new one.", variable, scope);

75 +

76 + rv = callocz(1, sizeof(RRDVAR));

77 + rv->name = variable;

78 + rv->hash = hash;

79 + rv->type = type;

80 + rv->value = value;

68 - rv = rrdvar_create(name, hash, type, value);

82 RRDVAR *ret = rrdvar_index_add(tree, rv);

83 if(unlikely(ret != rv)) {

71 - debug(D_VARIABLES, "Variable '%s' in scope '%s' already exists", name, scope);

72 - rrdvar_free(NULL, rv);

84 + debug(D_VARIABLES, "Variable '%s' in scope '%s' already exists", variable, scope);

85 + rrdvar_free(NULL, NULL, rv);

86 rv = NULL;

87 }

88 else

76 - debug(D_VARIABLES, "Variable '%s' created in scope '%s'", name, scope);

89 + debug(D_VARIABLES, "Variable '%s' created in scope '%s'", variable, scope);

90 }

91 else {

92 // already exists

93 + freez(variable);

94 rv = NULL;

95 }

@@ -120,7 +134,7 @@ calculated_number rrdvar2number(RRDVAR *rv) {

134

135 void dump_variable(void *data) {

136 RRDVAR *rv = (RRDVAR *)data;

123 - debug(D_HEALTH, "%30s : " CALCULATED_NUMBER_FORMAT, rv->name, rrdvar2number(rv));

137 + debug(D_HEALTH, "%50s : %20.5Lf", rv->name, rrdvar2number(rv));

138 }

139

140 int health_variable_lookup(const char *variable, uint32_t hash, RRDCALC *rc, calculated_number *result) {

@@ -166,28 +180,25 @@ RRDSETVAR *rrdsetvar_create(RRDSET *st, const char *variable, int type, void *va

180 debug(D_VARIABLES, "RRDVARSET create for chart id '%s' name '%s' with variable name '%s'", st->id, st->name, variable);

181 RRDSETVAR *rs = (RRDSETVAR *)callocz(1, sizeof(RRDSETVAR));

182

169 - char buffer[RRDSETVAR_ID_MAX + 1];

170 - snprintfz(buffer, RRDSETVAR_ID_MAX, "%s.%s", st->id, variable);

183 + char buffer[RRDVAR_MAX_LENGTH + 1];

184 + snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->id, variable);

185 rs->fullid = strdupz(buffer);

172 - rs->hash_fullid = simple_hash(rs->fullid);

186

174 - snprintfz(buffer, RRDSETVAR_ID_MAX, "%s.%s", st->name, variable);

187 + snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, variable);

188 rs->fullname = strdupz(buffer);

176 - rs->hash_fullname = simple_hash(rs->fullname);

189

190 rs->variable = strdupz(variable);

179 - rs->hash_variable = simple_hash(rs->variable);

191

192 rs->type = type;

193 rs->value = value;

194 rs->options = options;

195 rs->rrdset = st;

196

186 - rs->local = rrdvar_create_and_index("local", &st->variables_root_index, rs->variable, rs->hash_variable, rs->type, rs->value);

187 - rs->context = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->fullid, rs->hash_fullid, rs->type, rs->value);

188 - rs->host = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullid, rs->hash_fullid, rs->type, rs->value);

189 - rs->context_name = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->fullname, rs->hash_fullname, rs->type, rs->value);

190 - rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->hash_fullname, rs->type, rs->value);

197 + rs->local = rrdvar_create_and_index("local", &st->variables_root_index, rs->variable, rs->type, rs->value);

198 + rs->context = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->fullid, rs->type, rs->value);

199 + rs->host = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullid, rs->type, rs->value);

200 + rs->context_name = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->fullname, rs->type, rs->value);

201 + rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);

202

203 rs->next = st->variables;

204 st->variables = rs;

@@ -202,30 +213,22 @@ void rrdsetvar_rename_all(RRDSET *st) {

213 // rs->context_name

214 // rs->host_name

215

205 - char buffer[RRDSETVAR_ID_MAX + 1];

216 + char buffer[RRDVAR_MAX_LENGTH + 1];

217 RRDSETVAR *rs, *next = st->variables;

218 while((rs = next)) {

219 next = rs->next;

220

210 - snprintfz(buffer, RRDSETVAR_ID_MAX, "%s.%s", st->name, rs->variable);

221 + snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rs->variable);

222

223 if (strcmp(buffer, rs->fullname)) {

224 // name changed

214 - if (rs->context_name) {

215 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context_name);

216 - rrdvar_free(st->rrdhost, rs->context_name);

217 - }

218 -

219 - if (rs->host_name) {

220 - rrdvar_index_del(&st->rrdhost->variables_root_index, rs->host_name);

221 - rrdvar_free(st->rrdhost, rs->host_name);

222 - }

225 + rrdvar_free(st->rrdhost, &st->rrdcontext->variables_root_index, rs->context_name);

226 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);

227

228 freez(rs->fullname);

229 rs->fullname = strdupz(st->name);

226 - rs->hash_fullname = simple_hash(rs->fullname);

227 - rs->context_name = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->fullname, rs->hash_fullname, rs->type, rs->value);

228 - rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->hash_fullname, rs->type, rs->value);

230 + rs->context_name = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->fullname, rs->type, rs->value);

231 + rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);

232 }

233 }

234

@@ -236,30 +239,11 @@ void rrdsetvar_free(RRDSETVAR *rs) {

239 RRDSET *st = rs->rrdset;

240 debug(D_VARIABLES, "RRDSETVAR free for chart id '%s' name '%s', variable '%s'", st->id, st->name, rs->variable);

241

239 - if(rs->local) {

240 - rrdvar_index_del(&st->variables_root_index, rs->local);

241 - rrdvar_free(st->rrdhost, rs->local);

242 - }

243 -

244 - if(rs->context) {

245 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context);

246 - rrdvar_free(st->rrdhost, rs->context);

247 - }

248 -

249 - if(rs->host) {

250 - rrdvar_index_del(&st->rrdhost->variables_root_index, rs->host);

251 - rrdvar_free(st->rrdhost, rs->host);

252 - }

253 -

254 - if(rs->context_name) {

255 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context_name);

256 - rrdvar_free(st->rrdhost, rs->context_name);

257 - }

258 -

259 - if(rs->host_name) {

260 - rrdvar_index_del(&st->rrdhost->variables_root_index, rs->host_name);

261 - rrdvar_free(st->rrdhost, rs->host_name);

262 - }

242 + rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local);

243 + rrdvar_free(st->rrdhost, &st->rrdcontext->variables_root_index, rs->context);

244 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host);

245 + rrdvar_free(st->rrdhost, &st->rrdcontext->variables_root_index, rs->context_name);

246 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);

247

248 if(st->variables == rs) {

249 st->variables = rs->next;

@@ -298,48 +282,37 @@ RRDDIMVAR *rrddimvar_create(RRDDIM *rd, int type, const char *prefix, const char

282

283 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->id, rs->suffix);

284 rs->id = strdupz(buffer);

301 - rs->hash = simple_hash(rs->id);

285

286 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);

287 rs->name = strdupz(buffer);

305 - rs->hash_name = simple_hash(rs->name);

288

289 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->id);

290 rs->fullidid = strdupz(buffer);

309 - rs->hash_fullidid = simple_hash(rs->fullidid);

291

292 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->name);

293 rs->fullidname = strdupz(buffer);

313 - rs->hash_fullidname = simple_hash(rs->fullidname);

294

295 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->id);

296 rs->fullnameid = strdupz(buffer);

317 - rs->hash_fullnameid = simple_hash(rs->fullnameid);

297

298 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->name);

299 rs->fullnamename = strdupz(buffer);

321 - rs->hash_fullnamename = simple_hash(rs->fullnamename);

300

301 rs->type = type;

302 rs->value = value;

303 rs->options = options;

304 rs->rrddim = rd;

305

328 - rs->local_id = rrdvar_create_and_index("local", &st->variables_root_index, rs->id, rs->hash, rs->type, rs->value);

329 - rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->hash_name, rs->type, rs->value);

330 -

331 - rs->context_id = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->id, rs->hash, rs->type, rs->value);

332 - rs->context_name = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->name, rs->hash_name, rs->type, rs->value);

306 + rs->local_id = rrdvar_create_and_index("local", &st->variables_root_index, rs->id, rs->type, rs->value);

307 + rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);

308

334 - rs->context_fullidid = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->fullidid, rs->hash_fullidid, rs->type, rs->value);

335 - rs->context_fullidname = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->fullidname, rs->hash_fullidname, rs->type, rs->value);

336 - rs->context_fullnameid = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->fullnameid, rs->hash_fullnameid, rs->type, rs->value);

337 - rs->context_fullnamename = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->fullnamename, rs->hash_fullnamename, rs->type, rs->value);

309 + rs->context_id = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->id, rs->type, rs->value);

310 + rs->context_name = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rs->name, rs->type, rs->value);

311

339 - rs->host_fullidid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidid, rs->hash_fullidid, rs->type, rs->value);

340 - rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidname, rs->hash_fullidname, rs->type, rs->value);

341 - rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnameid, rs->hash_fullnameid, rs->type, rs->value);

342 - rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnamename, rs->hash_fullnamename, rs->type, rs->value);

312 + rs->host_fullidid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidid, rs->type, rs->value);

313 + rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidname, rs->type, rs->value);

314 + rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnameid, rs->type, rs->value);

315 + rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnamename, rs->type, rs->value);

316

317 rs->next = rd->variables;

318 rd->variables = rs;

@@ -360,69 +333,34 @@ void rrddimvar_rename_all(RRDDIM *rd) {

333 // name changed

334

335 // name

363 - if (rs->local_name) {

364 - rrdvar_index_del(&st->variables_root_index, rs->local_name);

365 - rrdvar_free(st->rrdhost, rs->local_name);

366 - }

336 + rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);

337 freez(rs->name);

338 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);

339 rs->name = strdupz(buffer);

370 - rs->hash_name = simple_hash(rs->name);

371 - rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->hash_name, rs->type, rs->value);

340 + rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);

341

373 - // fullidname

374 - if (rs->context_fullidname) {

375 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context_fullidname);

376 - rrdvar_free(st->rrdhost, rs->context_fullidname);

377 - }

378 - if (rs->host_fullidname) {

379 - rrdvar_index_del(&st->rrdhost->variables_root_index, rs->context_fullidname);

380 - rrdvar_free(st->rrdhost, rs->host_fullidname);

381 - }

342 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);

343 freez(rs->fullidname);

344 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->id, rs->name);

345 rs->fullidname = strdupz(buffer);

385 - rs->hash_fullidname = simple_hash(rs->fullidname);

386 - rs->context_fullidname = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index,

387 - rs->fullidname, rs->hash_fullidname, rs->type, rs->value);

346 rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,

389 - rs->fullidname, rs->hash_fullidname, rs->type, rs->value);

347 + rs->fullidname, rs->type, rs->value);

348

349 // fullnameid

392 - if (rs->context_fullnameid) {

393 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context_fullnameid);

394 - rrdvar_free(st->rrdhost, rs->context_fullnameid);

395 - }

396 - if (rs->host_fullnameid) {

397 - rrdvar_index_del(&st->rrdhost->variables_root_index, rs->context_fullnameid);

398 - rrdvar_free(st->rrdhost, rs->host_fullnameid);

399 - }

350 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);

351 freez(rs->fullnameid);

352 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->id);

353 rs->fullnameid = strdupz(buffer);

403 - rs->hash_fullnameid = simple_hash(rs->fullnameid);

404 - rs->context_fullnameid = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index,

405 - rs->fullnameid, rs->hash_fullnameid, rs->type, rs->value);

354 rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,

407 - rs->fullnameid, rs->hash_fullnameid, rs->type, rs->value);

355 + rs->fullnameid, rs->type, rs->value);

356

357 // fullnamename

410 - if (rs->context_fullnamename) {

411 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context_fullnamename);

412 - rrdvar_free(st->rrdhost, rs->context_fullnamename);

413 - }

414 - if (rs->host_fullnamename) {

415 - rrdvar_index_del(&st->rrdhost->variables_root_index, rs->context_fullnamename);

416 - rrdvar_free(st->rrdhost, rs->host_fullnamename);

417 - }

358 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);

359 freez(rs->fullnamename);

360 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->name);

361 rs->fullnamename = strdupz(buffer);

421 - rs->hash_fullnamename = simple_hash(rs->fullnamename);

422 - rs->context_fullnamename = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index,

423 - rs->fullnamename, rs->hash_fullnamename, rs->type, rs->value);

362 rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,

425 - rs->fullnamename, rs->hash_fullnamename, rs->type, rs->value);

363 + rs->fullnamename, rs->type, rs->value);

364 }

365 }

366 }

@@ -432,56 +370,16 @@ void rrddimvar_free(RRDDIMVAR *rs) {

370 RRDSET *st = rd->rrdset;

371 debug(D_VARIABLES, "RRDDIMSET free for chart id '%s' name '%s', dimension id '%s', name '%s', prefix='%s', suffix='%s'", st->id, st->name, rd->id, rd->name, rs->prefix, rs->suffix);

372

435 - if(rs->local_id) {

436 - rrdvar_index_del(&st->variables_root_index, rs->local_id);

437 - rrdvar_free(st->rrdhost, rs->local_id);

438 - }

439 - if(rs->local_name) {

440 - rrdvar_index_del(&st->variables_root_index, rs->local_name);

441 - rrdvar_free(st->rrdhost, rs->local_name);

442 - }

373 + rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_id);

374 + rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);

375

444 - if(rs->context_id) {

445 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context_id);

446 - rrdvar_free(st->rrdhost, rs->context_id);

447 - }

448 - if(rs->context_name) {

449 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context_name);

450 - rrdvar_free(st->rrdhost, rs->context_name);

451 - }

452 - if(rs->context_fullidid) {

453 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context_fullidid);

454 - rrdvar_free(st->rrdhost, rs->context_fullidid);

455 - }

456 - if(rs->context_fullidname) {

457 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context_fullidname);

458 - rrdvar_free(st->rrdhost, rs->context_fullidname);

459 - }

460 - if(rs->context_fullnameid) {

461 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context_fullnameid);

462 - rrdvar_free(st->rrdhost, rs->context_fullnameid);

463 - }

464 - if(rs->context_fullnamename) {

465 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rs->context_fullnamename);

466 - rrdvar_free(st->rrdhost, rs->context_fullnamename);

467 - }

376 + rrdvar_free(st->rrdhost, &st->rrdcontext->variables_root_index, rs->context_id);

377 + rrdvar_free(st->rrdhost, &st->rrdcontext->variables_root_index, rs->context_name);

378

469 - if(rs->host_fullidid) {

470 - rrdvar_index_del(&st->rrdhost->variables_root_index, rs->host_fullidid);

471 - rrdvar_free(st->rrdhost, rs->host_fullidid);

472 - }

473 - if(rs->host_fullidname) {

474 - rrdvar_index_del(&st->rrdhost->variables_root_index, rs->host_fullidname);

475 - rrdvar_free(st->rrdhost, rs->host_fullidname);

476 - }

477 - if(rs->host_fullnameid) {

478 - rrdvar_index_del(&st->rrdhost->variables_root_index, rs->host_fullnameid);

479 - rrdvar_free(st->rrdhost, rs->host_fullnameid);

480 - }

481 - if(rs->host_fullnamename) {

482 - rrdvar_index_del(&st->rrdhost->variables_root_index, rs->host_fullnamename);

483 - rrdvar_free(st->rrdhost, rs->host_fullnamename);

484 - }

379 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidid);

380 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);

381 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);

382 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);

383

384 if(rd->variables == rs) {

385 debug(D_VARIABLES, "RRDDIMSET removing first entry for chart id '%s' name '%s', dimension id '%s', name '%s'", st->id, st->name, rd->id, rd->name);

@@ -508,75 +406,9 @@ void rrddimvar_free(RRDDIMVAR *rs) {

406

407 // ----------------------------------------------------------------------------

408 // RRDCALC management

511 -/*

512 -// this has to be called while the caller has locked

513 -// the RRDHOST

514 -static inline void rrdset_linked_optimize_rrdhost(RRDHOST *host, RRDCALC *rc) {

515 - rrdhost_check_wrlock(host);

516 -

517 - // move it to be last

518 -

519 - if(!rc->next)

520 - // we are last already

521 - return;

522 -

523 - RRDCALC *t, *last = NULL, *prev = NULL;

524 - for (t = host->calculations; t ; t = t->next) {

525 - if(t->next == rc)

526 - prev = t;

527 -

528 - if(!t->next)

529 - last = t;

530 - }

531 -

532 - if(!last) {

533 - error("RRDCALC '%s' cannot be linked to the end of host '%s' list", rc->name, host->hostname);

534 - return;

535 - }

536 -

537 - if(prev)

538 - prev->next = rc->next;

539 - else {

540 - if(host->calculations == rc)

541 - host->calculations = rc->next;

542 - else {

543 - error("RRDCALC '%s' is not found in host '%s' list", rc->name, host->hostname);

544 - return;

545 - }

546 - }

547 -

548 - last->next = rc;

549 - rc->next = NULL;

550 -}

551 -

552 -// this has to be called while the caller has locked

553 -// the RRDHOST

554 -static inline void rrdcalc_unlinked_optimize_rrdhost(RRDHOST *host, RRDCALC *rc) {

555 - rrdhost_check_wrlock(host);

556 -

557 - // move it to be first

558 -

559 - if(host->calculations == rc) {

560 - // ok, we are the first

561 - return;

562 - }

563 - else {

564 - // find the previous one

565 - RRDCALC *t;

566 - for (t = host->calculations; t && t->next != rc; rc = rc->next) ;

567 - if(unlikely(!t)) {

568 - error("RRDCALC '%s' is not linked to host '%s'.", rc->name, host->hostname);

569 - return;

570 - }

571 - t->next = rc->next;

572 - rc->next = host->calculations;

573 - host->calculations = rc;

574 - }

575 -}

576 -*/

409

410 static void rrdsetcalc_link(RRDSET *st, RRDCALC *rc) {

579 - debug(D_HEALTH, "Health linking alarm '%s' from chart '%s' of host '%s'", rc->name, st->id, st->rrdhost->hostname);

411 + debug(D_HEALTH, "Health linking alarm '%s.%s' from chart '%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, st->id, st->rrdhost->hostname);

412

413 rc->rrdset = st;

414

@@ -586,11 +418,15 @@ static void rrdsetcalc_link(RRDSET *st, RRDCALC *rc) {

418 if(rc->red && !st->red)

419 st->red = rc->red;

420

589 - rc->local = rrdvar_create_and_index("local", &st->variables_root_index, rc->name, rc->hash, RRDVAR_TYPE_CALCULATED, &rc->value);

590 - rc->context = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rc->name, rc->hash, RRDVAR_TYPE_CALCULATED, &rc->value);

591 - rc->host = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rc->name, rc->hash, RRDVAR_TYPE_CALCULATED, &rc->value);

421 + rc->local = rrdvar_create_and_index("local", &st->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);

422 + rc->context = rrdvar_create_and_index("context", &st->rrdcontext->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);

423 +

424 + char fullname[RRDVAR_MAX_LENGTH + 1];

425 + snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->id, rc->name);

426 + rc->hostid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);

427

593 - // rrdset_linked_optimize_rrdhost(st->rrdhost, rc);

428 + snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rc->name);

429 + rc->hostname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);

430 }

431

432 static inline int rrdcalc_is_matching_this_rrdset(RRDCALC *rc, RRDSET *st) {

@@ -602,7 +438,7 @@ static inline int rrdcalc_is_matching_this_rrdset(RRDCALC *rc, RRDSET *st) {

438 }

439

440 // this has to be called while the RRDHOST is locked

605 -void rrdsetcalc_link_matching(RRDSET *st) {

441 +inline void rrdsetcalc_link_matching(RRDSET *st) {

442 // debug(D_HEALTH, "find matching alarms for chart '%s'", st->id);

443

444 RRDCALC *rc;

@@ -615,17 +451,17 @@ void rrdsetcalc_link_matching(RRDSET *st) {

451 }

452

453 // this has to be called while the RRDHOST is locked

618 -void rrdsetcalc_unlink(RRDCALC *rc) {

454 +inline void rrdsetcalc_unlink(RRDCALC *rc) {

455 RRDSET *st = rc->rrdset;

456

457 if(!st) {

622 - error("Requested to unlink RRDCALC '%s' which is not linked to any RRDSET", rc->name);

458 + error("Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);

459 return;

460 }

461

462 RRDHOST *host = st->rrdhost;

463

628 - debug(D_HEALTH, "Health unlinking alarm '%s' from chart '%s' of host '%s'", rc->name, st->id, host->hostname);

464 + debug(D_HEALTH, "Health unlinking alarm '%s.%s' from chart '%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, st->id, host->hostname);

465

466 // unlink it

467 if(rc->rrdset_prev)

@@ -639,29 +475,23 @@ void rrdsetcalc_unlink(RRDCALC *rc) {

475

476 rc->rrdset_prev = rc->rrdset_next = NULL;

477

642 - if(rc->local) {

643 - rrdvar_index_del(&st->variables_root_index, rc->local);

644 - rrdvar_free(st->rrdhost, rc->local);

645 - rc->local = NULL;

646 - }

478 + rrdvar_free(st->rrdhost, &st->variables_root_index, rc->local);

479 + rc->local = NULL;

480

648 - if(rc->context) {

649 - rrdvar_index_del(&st->rrdcontext->variables_root_index, rc->context);

650 - rc->context = NULL;

651 - }

481 + rrdvar_free(st->rrdhost, &st->rrdcontext->variables_root_index, rc->context);

482 + rc->context = NULL;

483

653 - if(rc->host) {

654 - rrdvar_index_del(&st->rrdhost->variables_root_index, rc->host);

655 - rc->host = NULL;

656 - }

484 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostid);

485 + rc->hostid = NULL;

486 +

487 + rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostname);

488 + rc->hostname = NULL;

489

490 rc->rrdset = NULL;

491

492 // RRDCALC will remain in RRDHOST

493 // so that if the matching chart is found in the future

494 // it will be applied automatically

663 -

664 - // rrdcalc_unlinked_optimize_rrdhost(host, rc);

495 }

496

497 static inline int rrdcalc_exists(RRDHOST *host, const char *name, uint32_t hash) {

@@ -670,7 +500,7 @@ static inline int rrdcalc_exists(RRDHOST *host, const char *name, uint32_t hash)

500 // make sure it does not already exist

501 for(rc = host->calculations; rc ; rc = rc->next) {

502 if (rc->hash == hash && !strcmp(name, rc->name)) {

673 - error("alarm '%s' already exists in host '%s'.", name, host->hostname);

503 + error("Health alarm '%s' already exists in host '%s'.", name, host->hostname);

504 return 1;

505 }

506 }

@@ -678,7 +508,7 @@ static inline int rrdcalc_exists(RRDHOST *host, const char *name, uint32_t hash)

508 return 0;

509 }

510

681 -void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {

511 +static inline void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {

512 rrdhost_check_rdlock(host);

513

514 if(rc->calculation) {

@@ -710,14 +540,22 @@ void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {

540 }

541 }

542

713 -RRDCALC *rrdcalc_create(RRDHOST *host, const char *name, const char *chart, const char *dimensions, int group_method,

543 +static inline uint32_t rrdcalc_fullname(char *fullname, size_t len, const char *chart, const char *name) {

544 + snprintfz(fullname, len - 1, "%s%s%s", chart?chart:"", chart?".":"", name);

545 + rrdvar_fix_name(fullname);

546 + return simple_hash(fullname);

547 +}

548 +

549 +static inline RRDCALC *rrdcalc_create(RRDHOST *host, const char *name, const char *chart, const char *dimensions, int group_method,

550 int after, int before, int update_every, uint32_t options,

551 calculated_number green, calculated_number red,

552 const char *exec, const char *source,

553 const char *calc, const char *warn, const char *crit) {

718 - uint32_t hash = simple_hash(name);

554

720 - if(rrdcalc_exists(host, name, hash))

555 + char fullname[RRDVAR_MAX_LENGTH + 1];

556 + uint32_t hash = rrdcalc_fullname(fullname, RRDVAR_MAX_LENGTH + 1, chart, name);

557 +

558 + if(rrdcalc_exists(host, fullname, hash))

559 return NULL;

560

561 RRDCALC *rc = callocz(1, sizeof(RRDCALC));

@@ -745,22 +583,22 @@ RRDCALC *rrdcalc_create(RRDHOST *host, const char *name, const char *chart, cons

583 if(calc) {

584 rc->calculation = expression_parse(calc, NULL, NULL);

585 if(!rc->calculation)

748 - error("Failed to parse calculation expression '%s'", calc);

586 + error("Health alarm '%s.%s': failed to parse calculation expression '%s'", chart, name, calc);

587 }

588 if(warn) {

589 rc->warning = expression_parse(warn, NULL, NULL);

590 if(!rc->warning)

753 - error("Failed to re-parse warning expression '%s'", warn);

591 + error("Health alarm '%s.%s': failed to re-parse warning expression '%s'", chart, name, warn);

592 }

593 if(crit) {

594 rc->critical = expression_parse(crit, NULL, NULL);

595 if(!rc->critical)

758 - error("Failed to re-parse critical expression '%s'", crit);

596 + error("Health alarm '%s.%s': failed to re-parse critical expression '%s'", chart, name, crit);

597 }

598

761 - debug(D_HEALTH, "Health runtime added alarm '%s': chart '%s', exec '%s', green %Lf, red %Lf, lookup: group %d, after %d, before %d, options %u, dimensions '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s",

599 + debug(D_HEALTH, "Health runtime added alarm '%s.%s': exec '%s', green %Lf, red %Lf, lookup: group %d, after %d, before %d, options %u, dimensions '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s",

600 + (rc->chart)?rc->chart:"NOCHART",

601 rc->name,

763 - (rc->chart)?rc->chart:"NONE",

602 (rc->exec)?rc->exec:"DEFAULT",

603 rc->green,

604 rc->red,

@@ -783,7 +621,7 @@ RRDCALC *rrdcalc_create(RRDHOST *host, const char *name, const char *chart, cons

621 void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {

622 if(!rc) return;

623

786 - debug(D_HEALTH, "Health removing alarm '%s' of host '%s'", rc->name, host->hostname);

624 + debug(D_HEALTH, "Health removing alarm '%s.%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);

625

626 // unlink it from RRDSET

627 if(rc->rrdset) rrdsetcalc_unlink(rc);

@@ -799,10 +637,10 @@ void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {

637 if(last && last->next == rc)

638 last->next = rc->next;

639 else

802 - error("Cannot unlink RRDCALC '%s' from RRDHOST '%s': not found", rc->name, host->hostname);

640 + error("Cannot unlink alarm '%s.%s' from host '%s': not found", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);

641 }

642 else

805 - error("Cannot unlink RRDCALC '%s' from RRDHOST '%s': RRDHOST does not have any calculations", rc->name, host->hostname);

643 + error("Cannot unlink unlink '%s.%s' from host '%s': This host does not have any calculations", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);

644

645 if(rc->warning) expression_free(rc->warning);

646 if(rc->critical) expression_free(rc->critical);

@@ -819,38 +657,28 @@ void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {

657 // ----------------------------------------------------------------------------

658 // RRDCALCTEMPLATE management

659

822 -static inline int variable_fix_name(char *variable) {

823 - int fixed = 0;

824 - while(*variable) {

825 - if (!isalnum(*variable) && *variable != '.' && *variable != '_') {

826 - *variable++ = '_';

827 - fixed++;

828 - }

829 - else

830 - variable++;

831 - }

832 -

833 - return fixed;

834 -}

835 -

660 void rrdcalctemplate_link_matching(RRDSET *st) {

661 RRDCALCTEMPLATE *rt;

662

663 for(rt = st->rrdhost->templates; rt ; rt = rt->next) {

664 if(rt->hash_context == st->hash_context && !strcmp(rt->context, st->context)) {

841 - char buffer[RRDSETVAR_ID_MAX + 1];

842 - snprintfz(buffer, RRDSETVAR_ID_MAX, "%s.%s", st->family, rt->name);

843 - variable_fix_name(buffer);

844 - RRDCALC *rc = rrdcalc_create(st->rrdhost, buffer, st->id,

665 +

666 + RRDCALC *rc = rrdcalc_create(st->rrdhost, rt->name, st->id,

667 rt->dimensions, rt->group, rt->after, rt->before, rt->update_every, rt->options,

668 rt->green, rt->red, rt->exec, rt->source,

669 (rt->calculation)?rt->calculation->source:NULL,

670 (rt->warning)?rt->warning->source:NULL,

671 (rt->critical)?rt->critical->source:NULL);

672

851 - // FIXME

852 - if(rc->rrdset != st)

853 - fatal("RRDCAL '%s' is not linked to chart '%s'", rc->name, st->id);

673 + if(!rc)

674 + error("Health tried to create alarm from template '%s', but it failed", rt->name);

675 +

676 +#ifdef NETDATA_INTERNAL_CHECKS

677 + else if(rc->rrdset != st)

678 + error("Health alarm '%s.%s' should be linked to chart '%s', but it is not", rc->chart?rc->chart:"NOCHART", rc->name, st->id);

679 +#else

680 + (void)rc;

681 +#endif

682 }

683 }

684 }

@@ -902,23 +730,28 @@ static inline void rrdcalctemplate_free(RRDHOST *host, RRDCALCTEMPLATE *rt) {

730 #define HEALTH_CRIT_KEY "crit"

731 #define HEALTH_EXEC_KEY "exec"

732

905 -static inline int rrdcalc_add(RRDHOST *host, RRDCALC *rc) {

906 - if(rrdcalc_exists(host, rc->name, rc->hash))

907 - return 0;

733 +static inline int rrdcalc_add_alarm_from_config(RRDHOST *host, RRDCALC *rc) {

734 + {

735 + char fullname[RRDVAR_MAX_LENGTH + 1];

736 + uint32_t hash = rrdcalc_fullname(fullname, RRDVAR_MAX_LENGTH + 1, rc->chart, rc->name);

737 +

738 + if (rrdcalc_exists(host, fullname, hash))

739 + return 0;

740 + }

741

742 if(!rc->chart) {

743 error("Health configuration for alarm '%s' does not have a chart", rc->name);

744 return 0;

745 }

746

914 - if(!RRDCALC_HAS_CALCULATION(rc) && !rc->warning && !rc->critical) {

915 - error("Health configuration for alarm '%s' is useless (no calculation, no warning and no critical evaluation)", rc->name);

747 + if(!RRDCALC_HAS_DB_LOOKUP(rc) && !rc->warning && !rc->critical) {

748 + error("Health configuration for alarm '%s.%s' is useless (no calculation, no warning and no critical evaluation)", rc->chart?rc->chart:"NOCHART", rc->name);

749 return 0;

750 }

751

919 - debug(D_HEALTH, "Health configuration adding alarm '%s': chart '%s', exec '%s', green %Lf, red %Lf, lookup: group %d, after %d, before %d, options %u, dimensions '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s",

752 + debug(D_HEALTH, "Health configuration adding alarm '%s.%s': exec '%s', green %Lf, red %Lf, lookup: group %d, after %d, before %d, options %u, dimensions '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s",

753 + rc->chart?rc->chart:"NOCHART",

754 rc->name,

921 - (rc->chart)?rc->chart:"NONE",

755 (rc->exec)?rc->exec:"DEFAULT",

756 rc->green,

757 rc->red,

@@ -1207,7 +1040,7 @@ int health_readfile(const char *path, const char *filename) {

1040 uint32_t hash = simple_uhash(key);

1041

1042 if(hash == hash_alarm && !strcasecmp(key, HEALTH_ALARM_KEY)) {

1210 - if(rc && !rrdcalc_add(&localhost, rc))

1043 + if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))

1044 rrdcalc_free(&localhost, rc);

1045

1046 if(rt) {

@@ -1221,12 +1054,12 @@ int health_readfile(const char *path, const char *filename) {

1054 rc->hash = simple_hash(rc->name);

1055 rc->source = health_source_file(line, path, filename);

1056

1224 - if(variable_fix_name(rc->name))

1057 + if(rrdvar_fix_name(rc->name))

1058 error("Health configuration renamed alarm '%s' to '%s'", value, rc->name);

1059 }

1060 else if(hash == hash_template && !strcasecmp(key, HEALTH_TEMPLATE_KEY)) {

1061 if(rc) {

1229 - if(!rrdcalc_add(&localhost, rc))

1062 + if(!rrdcalc_add_alarm_from_config(&localhost, rc))

1063 rrdcalc_free(&localhost, rc);

1064 rc = NULL;

1065 }

@@ -1239,7 +1072,7 @@ int health_readfile(const char *path, const char *filename) {

1072 rt->hash_name = simple_hash(rt->name);

1073 rt->source = health_source_file(line, path, filename);

1074

1242 - if(variable_fix_name(rt->name))

1075 + if(rrdvar_fix_name(rt->name))

1076 error("Health configuration renamed template '%s' to '%s'", value, rt->name);

1077 }

1078 else if(rc) {

@@ -1406,7 +1239,7 @@ int health_readfile(const char *path, const char *filename) {

1239 }

1240 }

1241

1409 - if(rc && !rrdcalc_add(&localhost, rc))

1242 + if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))

1243 rrdcalc_free(&localhost, rc);

1244

1245 if(rt && !rrdcalctemplate_add(&localhost, rt))

@@ -1483,12 +1316,12 @@ void health_init(void) {

1316

1317 static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run) {

1318 if (unlikely(!rc->rrdset)) {

1486 - debug(D_HEALTH, "Health not running alarm '%s'. It is not linked to a chart.", rc->name);

1319 + debug(D_HEALTH, "Health not running alarm '%s.%s'. It is not linked to a chart.", rc->chart?rc->chart:"NOCHART", rc->name);

1320 return 0;

1321 }

1322

1323 if (unlikely(!rc->update_every)) {

1491 - debug(D_HEALTH, "Health not running alarm '%s'. It does not have an update frequency", rc->name);

1324 + debug(D_HEALTH, "Health not running alarm '%s.%s'. It does not have an update frequency", rc->chart?rc->chart:"NOCHART", rc->name);

1325 return 0;

1326 }

1327

@@ -1496,14 +1329,72 @@ static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run)

1329 if (*next_run > rc->next_update)

1330 *next_run = rc->next_update;

1331

1499 - debug(D_HEALTH, "Health not examining alarm '%s' yet (will do in %d secs).", rc->name,

1500 - (int) (rc->next_update - now));

1332 + debug(D_HEALTH, "Health not examining alarm '%s.%s' yet (will do in %d secs).", rc->chart?rc->chart:"NOCHART", rc->name, (int) (rc->next_update - now));

1333 return 0;

1334 }

1335

1336 return 1;

1337 }

1338

1339 +static inline int rrdcalc_value2status(calculated_number n) {

1340 + if(isnan(n)) return RRDCALC_STATUS_UNDEFINED;

1341 + if(n) return RRDCALC_STATUS_RAISED;

1342 + return RRDCALC_STATUS_OFF;

1343 +}

1344 +

1345 +static inline const char *rrdcalc_status2string(int status) {

1346 + switch(status) {

1347 + case RRDCALC_STATUS_UNINITIALIZED:

1348 + return "UNINITIALIZED";

1349 +

1350 + case RRDCALC_STATUS_UNDEFINED:

1351 + return "UNDEFINED";

1352 +

1353 + case RRDCALC_STATUS_RAISED:

1354 + return "RAISED";

1355 +

1356 + case RRDCALC_STATUS_OFF:

1357 + return "OFF";

1358 +

1359 + default:

1360 + return "UNKNOWN";

1361 + }

1362 +}

1363 +

1364 +void rrdcalc_check_critical_event(RRDCALC *rc) {

1365 + calculated_number n = rc->critical->result;

1366 +

1367 + int old_status = rc->critical_status;

1368 + int new_status = rrdcalc_value2status(n);

1369 +

1370 + if(new_status != old_status) {

1371 + info("Health alarm '%s.%s' - CRITICAL condition changed status from %s to %s",

1372 + rc->chart?rc->chart:"NOCHART", rc->name,

1373 + rrdcalc_status2string(old_status),

1374 + rrdcalc_status2string(new_status)

1375 + );

1376 +

1377 + rc->critical_status = new_status;

1378 + }

1379 +}

1380 +

1381 +void rrdcalc_check_warning_event(RRDCALC *rc) {

1382 + calculated_number n = rc->warning->result;

1383 +

1384 + int old_status = rc->warning_status;

1385 + int new_status = rrdcalc_value2status(n);

1386 +

1387 + if(new_status != old_status) {

1388 + info("Health alarm '%s.%s' - WARNING condition changed status from %s to %s",

1389 + rc->chart?rc->chart:"NOCHART", rc->name,

1390 + rrdcalc_status2string(old_status),

1391 + rrdcalc_status2string(new_status)

1392 + );

1393 +

1394 + rc->warning_status = new_status;

1395 + }

1396 +}

1397 +

1398 void *health_main(void *ptr) {

1399 (void)ptr;

1400

@@ -1534,50 +1425,93 @@ void *health_main(void *ptr) {

1425 error("Cannot set pthread cancel state to DISABLE.");

1426

1427 rrdhost_rdlock(&localhost);

1428 +

1429 + // the first loop is to lookup values from the db

1430 for (rc = localhost.calculations; rc; rc = rc->next) {

1431 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))

1432 continue;

1433

1434 runnable++;

1542 - debug(D_HEALTH, "Health running alarm '%s'", rc->name);

1435

1436 // 1. if there is database lookup, do it

1545 - // 1b. if the lookup has a calculation expression run it

1546 - // 2. if there is warning expression, do it

1547 - // 3. if there is critical expression, do it

1548 -

1549 - if (rc->after) {

1550 - time_t latest_timestamp;

1551 - int value_is_null;

1552 - int ret = rrd2value(rc->rrdset, wb, &rc->value, rc->dimensions, 1, rc->after, rc->before, rc->group,

1553 - rc->options, &latest_timestamp, &value_is_null);

1554 - if (ret != 200) {

1555 - error("Health for alarm '%s', database lookup returned error %d", rc->name, ret);

1437 + // 2. if there is calculation expression, run it

1438 +

1439 + if (unlikely(RRDCALC_HAS_DB_LOOKUP(rc))) {

1440 + time_t old_db_timestamp = rc->db_timestamp;

1441 + int value_is_null = 0;

1442 +

1443 + int ret = rrd2value(rc->rrdset, wb, &rc->value,

1444 + rc->dimensions, 1, rc->after, rc->before, rc->group,

1445 + rc->options, &rc->db_timestamp, &value_is_null);

1446 +

1447 + if (unlikely(ret != 200)) {

1448 + // database lookup failed

1449 + rc->value = NAN;

1450 +

1451 + if (unlikely(!(rc->rrdcalc_options & RRDCALC_OPTION_DB_ERROR))) {

1452 + rc->rrdcalc_options |= RRDCALC_OPTION_DB_ERROR;

1453 + error("Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);

1454 + }

1455 }

1557 - else {

1558 - if(value_is_null) {

1559 - rc->value = NAN;

1560 - error("Health for alarm '%s', database lookup returned empty value (possibly value is not collected yet)", rc->name);

1456 + else if (unlikely(rc->rrdcalc_options & RRDCALC_OPTION_DB_ERROR))

1457 + rc->rrdcalc_options &= ~RRDCALC_OPTION_DB_ERROR;

1458 +

1459 + if (unlikely(old_db_timestamp == rc->db_timestamp)) {

1460 + // database is stale

1461 +

1462 + if (unlikely(!(rc->rrdcalc_options & RRDCALC_OPTION_DB_STALE))) {

1463 + rc->rrdcalc_options |= RRDCALC_OPTION_DB_STALE;

1464 + error("Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);

1465 }

1562 - else {

1563 - debug(D_HEALTH, "Health for alarm '%s', database lookup gave value "

1564 - CALCULATED_NUMBER_FORMAT, rc->name, rc->value);

1565 -

1566 - if (rc->calculation) {

1567 - if (!expression_evaluate(rc->calculation)) {

1568 - error("Health for alarm '%s', failed to evaluate calculation with error: %s", rc->name,

1569 - buffer_tostring(rc->calculation->error_msg));

1570 - } else {

1571 - debug(D_HEALTH, "Health for alarm '%s', calculation expression gave value "

1572 - CALCULATED_NUMBER_FORMAT ": %s",

1573 - rc->name, rc->calculation->result,

1574 - buffer_tostring(rc->calculation->error_msg)

1575 - );

1576 - rc->value = rc->calculation->result;

1577 - }

1578 - }

1466 + }

1467 + else if (unlikely(rc->rrdcalc_options & RRDCALC_OPTION_DB_STALE))

1468 + rc->rrdcalc_options &= ~RRDCALC_OPTION_DB_STALE;

1469 +

1470 + if (unlikely(value_is_null)) {

1471 + // collected value is null

1472 +

1473 + rc->value = NAN;

1474 +

1475 + if (unlikely(!(rc->rrdcalc_options & RRDCALC_OPTION_DB_NAN))) {

1476 + rc->rrdcalc_options |= RRDCALC_OPTION_DB_NAN;

1477 + error("Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",

1478 + rc->chart?rc->chart:"NOCHART", rc->name);

1479 }

1480 }

1481 + else if (unlikely(rc->rrdcalc_options & RRDCALC_OPTION_DB_NAN))

1482 + rc->rrdcalc_options &= ~RRDCALC_OPTION_DB_NAN;

1483 +

1484 + debug(D_HEALTH, "Health alarm '%s.%s': database lookup gave value "

1485 + CALCULATED_NUMBER_FORMAT, rc->chart?rc->chart:"NOCHART", rc->name, rc->value);

1486 + }

1487 +

1488 + if(unlikely(rc->calculation)) {

1489 + if (unlikely(!expression_evaluate(rc->calculation))) {

1490 + // calculation failed

1491 +

1492 + rc->value = NAN;

1493 +

1494 + if (unlikely(!(rc->rrdcalc_options & RRDCALC_OPTION_CALC_ERROR))) {

1495 + rc->rrdcalc_options |= RRDCALC_OPTION_CALC_ERROR;

1496 + error("Health alarm '%s.%s': failed to evaluate calculation with error: %s",

1497 + rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));

1498 + }

1499 + }

1500 + else {

1501 + if (unlikely(rc->rrdcalc_options & RRDCALC_OPTION_CALC_ERROR))

1502 + rc->rrdcalc_options &= ~RRDCALC_OPTION_CALC_ERROR;

1503 +

1504 + debug(D_HEALTH, "Health alarm '%s.%s': calculation expression gave value "

1505 + CALCULATED_NUMBER_FORMAT

1506 + ": %s (source: %s)",

1507 + rc->chart?rc->chart:"NOCHART", rc->name,

1508 + rc->calculation->result,

1509 + buffer_tostring(rc->calculation->error_msg),

1510 + rc->source

1511 + );

1512 +

1513 + rc->value = rc->calculation->result;

1514 + }

1515 }

1516 }

1517 rrdhost_unlock(&localhost);

@@ -1589,48 +1523,56 @@ void *health_main(void *ptr) {

1523 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))

1524 continue;

1525

1592 - if(rc->warning) {

1593 - if (!expression_evaluate(rc->warning)) {

1594 - error("Health for alarm '%s', failed to evaluate warning expression with error: %s", rc->name,

1595 - buffer_tostring(rc->warning->error_msg));

1526 + if(unlikely(rc->warning)) {

1527 + if(unlikely(!expression_evaluate(rc->warning))) {

1528 + // calculation failed

1529 + if (unlikely(!(rc->rrdcalc_options & RRDCALC_OPTION_WARN_ERROR))) {

1530 + rc->rrdcalc_options |= RRDCALC_OPTION_WARN_ERROR;

1531 + error("Health alarm '%s.%s': warning expression failed with error: %s",

1532 + rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));

1533 + }

1534 }

1535 else {

1598 - debug(D_HEALTH, "Health for alarm '%s', warning expression gave value "

1599 - CALCULATED_NUMBER_FORMAT ": %s",

1600 - rc->name, rc->warning->result,

1601 - buffer_tostring(rc->warning->error_msg)

1536 + if(unlikely(rc->rrdcalc_options & RRDCALC_OPTION_WARN_ERROR))

1537 + rc->rrdcalc_options &= ~RRDCALC_OPTION_WARN_ERROR;

1538 +

1539 + debug(D_HEALTH, "Health alarm '%s.%s': warning expression gave value "

1540 + CALCULATED_NUMBER_FORMAT

1541 + ": %s (source: %s)",

1542 + rc->chart?rc->chart:"NOCHART", rc->name,

1543 + rc->warning->result,

1544 + buffer_tostring(rc->warning->error_msg),

1545 + rc->source

1546 );

1603 -

1604 - if(rc->warning->result)

1605 - info("WARNING ALARM '%s' - expression '%s' evaluated to %Lf, %s (source: %s)",

1606 - rc->name,

1607 - rc->warning->source,

1608 - rc->warning->result,

1609 - buffer_tostring(rc->warning->error_msg),

1610 - rc->source);

1547 }

1548 +

1549 + rrdcalc_check_warning_event(rc);

1550 }

1551

1614 - if(rc->critical) {

1615 - if (!expression_evaluate(rc->critical)) {

1616 - error("Health for alarm '%s', failed to evaluate critical expression with error: %s", rc->name,

1617 - buffer_tostring(rc->critical->error_msg));

1552 + if(unlikely(rc->critical)) {

1553 + if(unlikely(!expression_evaluate(rc->critical))) {

1554 + // calculation failed

1555 + if (unlikely(!(rc->rrdcalc_options & RRDCALC_OPTION_CRIT_ERROR))) {

1556 + rc->rrdcalc_options |= RRDCALC_OPTION_CRIT_ERROR;

1557 + error("Health alarm '%s.%s': critical expression failed with error: %s",

1558 + rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));

1559 + }

1560 }

1561 else {

1620 - debug(D_HEALTH, "Health for alarm '%s', critical expression gave value "

1621 - CALCULATED_NUMBER_FORMAT ": %s",

1622 - rc->name, rc->critical->result,

1623 - buffer_tostring(rc->critical->error_msg)

1562 + if(unlikely(rc->rrdcalc_options & RRDCALC_OPTION_CRIT_ERROR))

1563 + rc->rrdcalc_options &= ~RRDCALC_OPTION_CRIT_ERROR;

1564 +

1565 + debug(D_HEALTH, "Health alarm '%s.%s': critical expression gave value "

1566 + CALCULATED_NUMBER_FORMAT

1567 + ": %s (source: %s)",

1568 + rc->chart?rc->chart:"NOCHART", rc->name,

1569 + rc->critical->result,

1570 + buffer_tostring(rc->critical->error_msg),

1571 + rc->source

1572 );

1625 -

1626 - if(rc->critical->result)

1627 - info("CRITICAL ALARM '%s' - expression '%s' evaluated to %Lf, %s (source: %s)",

1628 - rc->name,

1629 - rc->critical->source,

1630 - rc->critical->result,

1631 - buffer_tostring(rc->critical->error_msg),

1632 - rc->source);

1573 }

1574 +

1575 + rrdcalc_check_critical_event(rc);

1576 }

1577

1578 rc->last_updated = now;

src/health.h

+22 -23

@@ -34,13 +34,8 @@ typedef struct rrdvar {

34 // these variables

35 typedef struct rrdsetvar {

36 char *fullid; // chart type.chart id.variable

37 - uint32_t hash_fullid;

38 -

37 char *fullname; // chart type.chart name.variable

40 - uint32_t hash_fullname;

41 -

38 char *variable; // variable

43 - uint32_t hash_variable;

40 int type;

41 void *value;

@@ -69,22 +64,11 @@ typedef struct rrddimvar {

64 char *suffix;

66 char *id; // dimension id

72 - uint32_t hash;

73 -

67 char *name; // dimension name

75 - uint32_t hash_name;

76 -

68 char *fullidid; // chart type.chart id.dimension id

78 - uint32_t hash_fullidid;

79 -

69 char *fullidname; // chart type.chart id.dimension name

81 - uint32_t hash_fullidname;

82 -

70 char *fullnameid; // chart type.chart name.dimension id

84 - uint32_t hash_fullnameid;

85 -

71 char *fullnamename; // chart type.chart name.dimension name

87 - uint32_t hash_fullnamename;

73 int type;

74 void *value;

@@ -97,11 +81,6 @@ typedef struct rrddimvar {

81 RRDVAR *context_id;

82 RRDVAR *context_name;

100 - RRDVAR *context_fullidid;

101 - RRDVAR *context_fullidname;

102 - RRDVAR *context_fullnameid;

103 - RRDVAR *context_fullnamename;

104 -

84 RRDVAR *host_fullidid;

85 RRDVAR *host_fullidname;

86 RRDVAR *host_fullnameid;

@@ -124,6 +103,19 @@ typedef struct rrddimvar {

103 // This double-linked list is maintained sorted at all times

104 // having as RRDSET.calculations the RRDCALC to be processed

105 // next.

106 +

107 +#define RRDCALC_STATUS_UNINITIALIZED 0

108 +#define RRDCALC_STATUS_UNDEFINED -1

109 +#define RRDCALC_STATUS_OFF 1

110 +#define RRDCALC_STATUS_RAISED 2

111 +

112 +#define RRDCALC_OPTION_DB_ERROR 0x00000001

113 +#define RRDCALC_OPTION_DB_NAN 0x00000002

114 +#define RRDCALC_OPTION_DB_STALE 0x00000004

115 +#define RRDCALC_OPTION_CALC_ERROR 0x00000008

116 +#define RRDCALC_OPTION_WARN_ERROR 0x00000010

117 +#define RRDCALC_OPTION_CRIT_ERROR 0x00000020

118 +

119 typedef struct rrdcalc {

120 char *name;

121 uint32_t hash;

@@ -150,6 +142,12 @@ typedef struct rrdcalc {

142 EVAL_EXPRESSION *warning;

143 EVAL_EXPRESSION *critical;

144

145 + uint32_t rrdcalc_options;

146 + int warning_status;

147 + int critical_status;

148 +

149 + time_t db_timestamp;

150 +

151 calculated_number value;

152

153 calculated_number green;

@@ -157,7 +155,8 @@ typedef struct rrdcalc {

155

156 RRDVAR *local;

157 RRDVAR *context;

160 - RRDVAR *host;

158 + RRDVAR *hostid;

159 + RRDVAR *hostname;

160

161 struct rrdset *rrdset;

162 struct rrdcalc *rrdset_next;

@@ -166,7 +165,7 @@ typedef struct rrdcalc {

165 struct rrdcalc *next;

166 } RRDCALC;

167

169 -#define RRDCALC_HAS_CALCULATION(rc) ((rc)->after)

168 +#define RRDCALC_HAS_DB_LOOKUP(rc) ((rc)->after)

169

170 // RRDCALCTEMPLATE

171 // these are to be applied to charts found dynamically

src/rrd.h

-5

@@ -306,11 +306,6 @@ struct rrdhost {

306 RRDCALC *calculations;

307

308 RRDCALCTEMPLATE *templates;

309 -

310 - // all variable references are linked here

311 - // RRDVARs may be free'd, so every time this happens

312 - // we need to find all their references and invalidate them

313 - EVAL_VARIABLE *references;

309 };

310 typedef struct rrdhost RRDHOST;

311 extern RRDHOST localhost;