operational graphite backend
Costa Tsaousis (ktsaou) committed
Nov 20, 2016 at 04:25 UTC
e17976fb86af30fd250aa352b361073e1807bc81
1 file changed
+140
-22
src/backends.c
+140
-22
@@ -154,32 +154,86 @@ static inline int connect_to_one(const char *definition, int default_port) {
154
return fd;
155
}
156
157
+static inline calculated_number backend_duration_average(RRDSET *st, RRDDIM *rd, time_t after, time_t before) {
158
+ time_t first_t = rrdset_first_entry_t(st);
159
+ time_t last_t = rrdset_last_entry_t(st);
160
+
161
+ if(unlikely(before - after < st->update_every && after != after - after % st->update_every))
162
+ // when st->update_every is bigger than the frequency we send data to backend
163
+ // skip the iterations that are not aligned to the database
164
+ return NAN;
165
+
166
+ // align the time-frame
167
+ // for 'after' also skip the first value by adding st->update_every
168
+ after = after - after % st->update_every + st->update_every;
169
+ before = before - before % st->update_every;
170
+
171
+ if(unlikely(after < first_t))
172
+ after = first_t;
173
+
174
+ if(unlikely(after > before))
175
+ // this can happen when the st->update_every > before - after
176
+ before = after;
177
+
178
+ if(unlikely(before > last_t))
179
+ before = last_t;
180
+
181
+ size_t counter = 0;
182
+ calculated_number sum = 0;
183
+
184
+ long start_at_slot = rrdset_time2slot(st, before),
185
+ stop_at_slot = rrdset_time2slot(st, after),
186
+ slot, stop_now = 0;
187
+
188
+ for(slot = start_at_slot; !stop_now ; slot--) {
189
+ if(unlikely(slot < 0)) slot = st->entries - 1;
190
+ if(unlikely(slot == stop_at_slot)) stop_now = 1;
191
+
192
+ storage_number n = rd->values[slot];
193
+ if(unlikely(!does_storage_number_exist(n))) continue;
194
+
195
+ calculated_number value = unpack_storage_number(n);
196
+ sum += value;
197
+ counter++;
198
+ }
199
+
200
+ if(!counter)
201
+ return NAN;
202
+
203
+ return sum / (calculated_number)counter;
204
+}
205
+
206
static inline void format_dimension_collected_graphite_plaintext(BUFFER *b, const char *prefix, RRDHOST *host, RRDSET *st, RRDDIM *rd, time_t after, time_t before) {
207
+ (void)before;
208
+
209
time_t last = rd->last_collected_time.tv_sec;
159
- if(likely(last >= after && last < before))
210
+ if(likely(last > after))
211
buffer_sprintf(b, "%s.%s.%s.%s " COLLECTED_NUMBER_FORMAT " %u\n", prefix, host->hostname, st->id, rd->id, rd->last_collected_value, (uint32_t)last);
212
}
213
214
static inline void format_dimension_stored_graphite_plaintext(BUFFER *b, const char *prefix, RRDHOST *host, RRDSET *st, RRDDIM *rd, time_t after, time_t before) {
164
- time_t last = rd->last_collected_time.tv_sec;
165
- if(likely(last >= after && last < before))
166
- buffer_sprintf(b, "%s.%s.%s.%s " CALCULATED_NUMBER_FORMAT " %u\n", prefix, host->hostname, st->id, rd->id, rd->last_stored_value, (uint32_t)last);
215
+ calculated_number value = backend_duration_average(st, rd, after, before);
216
+ if(!isnan(value))
217
+ buffer_sprintf(b, "%s.%s.%s.%s " CALCULATED_NUMBER_FORMAT " %u\n", prefix, host->hostname, st->id, rd->id, value, (uint32_t)before);
218
}
219
220
static inline void format_dimension_collected_opentsdb_telnet(BUFFER *b, const char *prefix, RRDHOST *host, RRDSET *st, RRDDIM *rd, time_t after, time_t before) {
221
+ (void)before;
222
+
223
time_t last = rd->last_collected_time.tv_sec;
171
- if(likely(last >= after && last < before))
224
+ if(likely(last > after))
225
buffer_sprintf(b, "put %s.%s.%s %u " COLLECTED_NUMBER_FORMAT " host=%s\n", prefix, st->id, rd->id, (uint32_t)last, rd->last_collected_value, host->hostname);
226
}
227
228
static inline void format_dimension_stored_opentsdb_telnet(BUFFER *b, const char *prefix, RRDHOST *host, RRDSET *st, RRDDIM *rd, time_t after, time_t before) {
176
- time_t last = rd->last_collected_time.tv_sec;
177
- if(likely(last >= after && last < before))
178
- buffer_sprintf(b, "put %s.%s.%s %u " CALCULATED_NUMBER_FORMAT " host=%s\n", prefix, st->id, rd->id, (uint32_t)last, rd->last_stored_value, host->hostname);
229
+ calculated_number value = backend_duration_average(st, rd, after, before);
230
+ if(!isnan(value))
231
+ buffer_sprintf(b, "put %s.%s.%s %u " CALCULATED_NUMBER_FORMAT " host=%s\n", prefix, st->id, rd->id, (uint32_t)before, value, host->hostname);
232
}
233
234
void *backends_main(void *ptr) {
235
(void)ptr;
236
+
237
BUFFER *b = buffer_create(1);
238
void (*formatter)(BUFFER *b, const char *prefix, RRDHOST *host, RRDSET *st, RRDDIM *rd, time_t after, time_t before);
239
@@ -192,12 +246,14 @@ void *backends_main(void *ptr) {
246
error("Cannot set pthread cancel state to ENABLE.");
247
248
int default_port = 0;
249
+ int sock = -1;
250
int enabled = config_get_boolean("backends", "enable", 0);
251
const char *source = config_get("backends", "data source", "as stored");
252
const char *type = config_get("backends", "type", "graphite");
253
const char *destination = config_get("backends", "destination", "localhost");
254
const char *prefix = config_get("backends", "prefix", "netdata");
200
- int frequency = (int)config_get_number("backends", "update every", 10);
255
+ int frequency = (int)config_get_number("backends", "update every", 30);
256
+ int buffer_on_failures = (int)config_get_number("backends", "buffer on failures", 10);
257
258
if(!enabled)
259
goto cleanup;
@@ -221,23 +277,27 @@ void *backends_main(void *ptr) {
277
goto cleanup;
278
}
279
224
- time_t after, before = (time_t)(time_usec() / 10000000ULL);
280
+ unsigned long long step_ut = frequency * 1000000ULL;
281
+ unsigned long long random_ut = time_usec() % (step_ut / 2);
282
+ time_t before = (time_t)((time_usec() - step_ut) / 10000000ULL);
283
+ time_t after = before;
284
+ int failures = 0;
285
226
- unsigned long long step = frequency * 1000000ULL;
286
for(;;) {
228
- unsigned long long now = time_usec();
229
- unsigned long long next = now - (now % step) + step + (step / 2);
287
+ unsigned long long now_ut = time_usec();
288
+ unsigned long long next_ut = now_ut - (now_ut % step_ut) + step_ut;
289
+ before = (time_t)(next_ut / 1000000ULL);
290
231
- while(now < next) {
232
- sleep_usec(next - now);
233
- now = time_usec();
291
+ // add a little delay (1/4 of the step) plus some randomness
292
+ next_ut += (step_ut / 4) + random_ut;
293
+
294
+ while(now_ut < next_ut) {
295
+ sleep_usec(next_ut - now_ut);
296
+ now_ut = time_usec();
297
}
298
236
- after = before;
237
- before = (time_t)(now / 10000000ULL);
299
RRDSET *st;
300
int pthreadoldcancelstate;
240
- buffer_flush(b);
301
302
if(unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &pthreadoldcancelstate) != 0))
303
error("Cannot set pthread cancel state to DISABLE.");
@@ -245,10 +305,11 @@ void *backends_main(void *ptr) {
305
rrdhost_rdlock(&localhost);
306
for(st = localhost.rrdset_root; st ;st = st->next) {
307
pthread_rwlock_rdlock(&st->rwlock);
308
+
309
RRDDIM *rd;
249
- for(rd = st->dimensions; rd ;rd = rd->next) {
310
+ for(rd = st->dimensions; rd ;rd = rd->next)
311
formatter(b, prefix, &localhost, st, rd, after, before);
251
- }
312
+
313
pthread_rwlock_unlock(&st->rwlock);
314
}
315
rrdhost_unlock(&localhost);
@@ -258,10 +319,67 @@ void *backends_main(void *ptr) {
319
320
if(unlikely(netdata_exit)) break;
321
261
- break;
322
+ //fprintf(stderr, "\nBACKEND BEGIN:\n%s\nBACKEND END\n", buffer_tostring(b));
323
+ //fprintf(stderr, "after = %lu, before = %lu\n", after, before);
324
+
325
+ if(unlikely(sock == -1)) {
326
+ const char *s = destination;
327
+ while(*s) {
328
+ const char *e = s;
329
+
330
+ // skip separators, moving both s(tart) and e(nd)
331
+ while(isspace(*e) || *e == ',') s = ++e;
332
+
333
+ // move e(nd) to the first separator
334
+ while(*e && !isspace(*e) && *e != ',') e++;
335
+
336
+ // is there anything?
337
+ if(!*s || s == e) break;
338
+
339
+ char buf[e - s + 1];
340
+ strncpyz(buf, s, e - s);
341
+ sock = connect_to_one(buf, default_port);
342
+ if(sock != -1) break;
343
+ s = e;
344
+ }
345
+ }
346
+
347
+ if(unlikely(netdata_exit)) break;
348
+
349
+ if(likely(sock != -1)) {
350
+ size_t len = buffer_strlen(b);
351
+ ssize_t written = write(sock, buffer_tostring(b), len);
352
+ if(written != -1 && (size_t)written == len) {
353
+ failures = 0;
354
+ buffer_flush(b);
355
+ }
356
+ else {
357
+ failures++;
358
+ error("Failed to write data to database backend '%s'. Willing to write %zu bytes, wrote %zd bytes. Will re-connect.", destination, len, written);
359
+ close(sock);
360
+ sock = -1;
361
+ }
362
+
363
+ after = before;
364
+ }
365
+ else {
366
+ failures++;
367
+ error("Failed to update database backend '%s'", destination);
368
+ }
369
+
370
+ if(failures > buffer_on_failures) {
371
+ error("Reached %d backend failures. Flushing buffers to protect this host - this results in data loss on back-end server '%s'", failures, destination);
372
+ buffer_flush(b);
373
+ failures = 0;
374
+ }
375
+
376
+ if(unlikely(netdata_exit)) break;
377
}
378
379
cleanup:
380
+ if(sock != -1)
381
+ close(sock);
382
+
383
info("BACKENDs thread exiting");
384
385
pthread_exit(NULL);