1 /*****************************************************************************
2 * RRDtool 1.2.23 Copyright by Tobi Oetiker, 1997-2007
3 *****************************************************************************
4 * rrd_fetch.c read date from an rrd to use for further processing
5 *****************************************************************************
6 * $Id$
7 * $Log$
8 * Revision 1.8 2004/05/18 18:53:03 oetiker
9 * big spell checking patch -- slif@bellsouth.net
10 *
11 * Revision 1.7 2003/11/11 19:46:21 oetiker
12 * replaced time_value with rrd_time_value as MacOS X introduced a struct of that name in their standard headers
13 *
14 * Revision 1.6 2003/01/16 23:27:54 oetiker
15 * fix border condition in rra selection of rrd_fetch
16 * -- Stanislav Sinyagin <ssinyagin@yahoo.com>
17 *
18 * Revision 1.5 2002/06/23 22:29:40 alex
19 * Added "step=1800" and such to "DEF"
20 * Cleaned some of the signed vs. unsigned problems
21 *
22 * Revision 1.4 2002/02/01 20:34:49 oetiker
23 * fixed version number and date/time
24 *
25 * Revision 1.3 2001/12/24 06:51:49 alex
26 * A patch of size 44Kbytes... in short:
27 *
28 * Found and repaired the off-by-one error in rrd_fetch_fn().
29 * As a result I had to remove the hacks in rrd_fetch_fn(),
30 * rrd_tool.c, vdef_calc(), data_calc(), data_proc() and
31 * reduce_data(). There may be other places which I didn't
32 * find so be careful.
33 *
34 * Enhanced debugging in rrd_fetch_fn(), it shows the RRA selection
35 * process.
36 *
37 * Added the ability to print VDEF timestamps. At the moment it
38 * is a hack, I needed it now to fix the off-by-one error.
39 * If the format string is "%c" (and nothing else!), the time
40 * will be printed by both ctime() and as a long int.
41 *
42 * Moved some code around (slightly altering it) from rrd_graph()
43 * initializing now in rrd_graph_init()
44 * options parsing now in rrd_graph_options()
45 * script parsing now in rrd_graph_script()
46 *
47 * Revision 1.2 2001/12/17 12:48:43 oetiker
48 * fix overflow error ...
49 *
50 * Revision 1.1.1.1 2001/02/25 22:25:05 oetiker
51 * checkin
52 *
53 *****************************************************************************/
55 #include "rrd_tool.h"
57 #include "rrd_is_thread_safe.h"
58 /*#define DEBUG*/
60 int rrd_fetch(
61 int argc,
62 char **argv,
63 time_t *start,
64 time_t *end, /* which time frame do you want ?
65 * will be changed to represent reality */
66 unsigned long *step, /* which stepsize do you want?
67 * will be changed to represent reality */
68 unsigned long *ds_cnt, /* number of data sources in file */
69 char ***ds_namv, /* names of data sources */
70 rrd_value_t **data)
71 { /* two dimensional array containing the data */
74 long step_tmp = 1;
75 time_t start_tmp = 0, end_tmp = 0;
76 const char *cf;
78 struct rrd_time_value start_tv, end_tv;
79 char *parsetime_error = NULL;
81 optind = 0;
82 opterr = 0; /* initialize getopt */
84 /* init start and end time */
85 parsetime("end-24h", &start_tv);
86 parsetime("now", &end_tv);
88 while (1) {
89 static struct option long_options[] = {
90 {"resolution", required_argument, 0, 'r'},
91 {"start", required_argument, 0, 's'},
92 {"end", required_argument, 0, 'e'},
93 {0, 0, 0, 0}
94 };
95 int option_index = 0;
96 int opt;
98 opt = getopt_long(argc, argv, "r:s:e:", long_options, &option_index);
100 if (opt == EOF)
101 break;
103 switch (opt) {
104 case 's':
105 if ((parsetime_error = parsetime(optarg, &start_tv))) {
106 rrd_set_error("start time: %s", parsetime_error);
107 return -1;
108 }
109 break;
110 case 'e':
111 if ((parsetime_error = parsetime(optarg, &end_tv))) {
112 rrd_set_error("end time: %s", parsetime_error);
113 return -1;
114 }
115 break;
116 case 'r':
117 step_tmp = atol(optarg);
118 break;
119 case '?':
120 rrd_set_error("unknown option '-%c'", optopt);
121 return (-1);
122 }
123 }
126 if (proc_start_end(&start_tv, &end_tv, &start_tmp, &end_tmp) == -1) {
127 return -1;
128 }
131 if (start_tmp < 3600 * 24 * 365 * 10) {
132 rrd_set_error("the first entry to fetch should be after 1980");
133 return (-1);
134 }
136 if (end_tmp < start_tmp) {
137 rrd_set_error("start (%ld) should be less than end (%ld)", start_tmp,
138 end_tmp);
139 return (-1);
140 }
142 *start = start_tmp;
143 *end = end_tmp;
145 if (step_tmp < 1) {
146 rrd_set_error("step must be >= 1 second");
147 return -1;
148 }
149 *step = step_tmp;
151 if (optind + 1 >= argc) {
152 rrd_set_error("not enough arguments");
153 return -1;
154 }
156 cf = argv[optind + 1];
158 if (rrd_fetch_r(argv[optind], cf, start, end, step, ds_cnt, ds_namv, data)
159 != 0)
160 return (-1);
161 return (0);
162 }
164 int rrd_fetch_r(
165 const char *filename, /* name of the rrd */
166 const char *cf, /* which consolidation function ? */
167 time_t *start,
168 time_t *end, /* which time frame do you want ?
169 * will be changed to represent reality */
170 unsigned long *step, /* which stepsize do you want?
171 * will be changed to represent reality */
172 unsigned long *ds_cnt, /* number of data sources in file */
173 char ***ds_namv, /* names of data_sources */
174 rrd_value_t **data)
175 { /* two dimensional array containing the data */
176 enum cf_en cf_idx;
178 if ((int) (cf_idx = cf_conv(cf)) == -1) {
179 return -1;
180 }
182 return (rrd_fetch_fn
183 (filename, cf_idx, start, end, step, ds_cnt, ds_namv, data));
184 }
186 int rrd_fetch_fn(
187 const char *filename, /* name of the rrd */
188 enum cf_en cf_idx, /* which consolidation function ? */
189 time_t *start,
190 time_t *end, /* which time frame do you want ?
191 * will be changed to represent reality */
192 unsigned long *step, /* which stepsize do you want?
193 * will be changed to represent reality */
194 unsigned long *ds_cnt, /* number of data sources in file */
195 char ***ds_namv, /* names of data_sources */
196 rrd_value_t **data)
197 { /* two dimensional array containing the data */
198 long i, ii;
199 time_t cal_start, cal_end, rra_start_time, rra_end_time;
200 long best_full_rra = 0, best_part_rra = 0, chosen_rra =
201 0, rra_pointer = 0;
202 long best_full_step_diff = 0, best_part_step_diff =
203 0, tmp_step_diff = 0, tmp_match = 0, best_match = 0;
204 long full_match, rra_base;
205 long start_offset, end_offset;
206 int first_full = 1;
207 int first_part = 1;
208 rrd_t rrd;
209 rrd_file_t *rrd_file;
210 rrd_value_t *data_ptr;
211 unsigned long rows;
213 #ifdef DEBUG
214 fprintf(stderr, "Entered rrd_fetch_fn() searching for the best match\n");
215 fprintf(stderr, "Looking for: start %10lu end %10lu step %5lu\n",
216 *start, *end, *step);
217 #endif
219 rrd_file = rrd_open(filename, &rrd, RRD_READONLY);
220 if (rrd_file == NULL)
221 goto err_free;
223 /* when was the really last update of this file ? */
225 if (((*ds_namv) =
226 (char **) malloc(rrd.stat_head->ds_cnt * sizeof(char *))) == NULL) {
227 rrd_set_error("malloc fetch ds_namv array");
228 goto err_close;
229 }
231 for (i = 0; (unsigned long) i < rrd.stat_head->ds_cnt; i++) {
232 if ((((*ds_namv)[i]) = malloc(sizeof(char) * DS_NAM_SIZE)) == NULL) {
233 rrd_set_error("malloc fetch ds_namv entry");
234 goto err_free_ds_namv;
235 }
236 strncpy((*ds_namv)[i], rrd.ds_def[i].ds_nam, DS_NAM_SIZE - 1);
237 (*ds_namv)[i][DS_NAM_SIZE - 1] = '\0';
239 }
241 /* find the rra which best matches the requirements */
242 for (i = 0; (unsigned) i < rrd.stat_head->rra_cnt; i++) {
243 if (cf_conv(rrd.rra_def[i].cf_nam) == cf_idx) {
245 cal_end = (rrd.live_head->last_up - (rrd.live_head->last_up
246 % (rrd.rra_def[i].pdp_cnt
247 *
248 rrd.stat_head->
249 pdp_step)));
250 cal_start =
251 (cal_end -
252 (rrd.rra_def[i].pdp_cnt * rrd.rra_def[i].row_cnt *
253 rrd.stat_head->pdp_step));
255 full_match = *end - *start;
256 #ifdef DEBUG
257 fprintf(stderr, "Considering: start %10lu end %10lu step %5lu ",
258 cal_start, cal_end,
259 rrd.stat_head->pdp_step * rrd.rra_def[i].pdp_cnt);
260 #endif
261 /* we need step difference in either full or partial case */
262 tmp_step_diff = labs(*step - (rrd.stat_head->pdp_step
263 * rrd.rra_def[i].pdp_cnt));
264 /* best full match */
265 if (cal_end >= *end && cal_start <= *start) {
266 if (first_full || (tmp_step_diff < best_full_step_diff)) {
267 first_full = 0;
268 best_full_step_diff = tmp_step_diff;
269 best_full_rra = i;
270 #ifdef DEBUG
271 fprintf(stderr, "best full match so far\n");
272 } else {
273 fprintf(stderr, "full match, not best\n");
274 #endif
275 }
277 } else {
278 /* best partial match */
279 tmp_match = full_match;
280 if (cal_start > *start)
281 tmp_match -= (cal_start - *start);
282 if (cal_end < *end)
283 tmp_match -= (*end - cal_end);
284 if (first_part ||
285 (best_match < tmp_match) ||
286 (best_match == tmp_match &&
287 tmp_step_diff < best_part_step_diff)) {
288 #ifdef DEBUG
289 fprintf(stderr, "best partial so far\n");
290 #endif
291 first_part = 0;
292 best_match = tmp_match;
293 best_part_step_diff = tmp_step_diff;
294 best_part_rra = i;
295 } else {
296 #ifdef DEBUG
297 fprintf(stderr, "partial match, not best\n");
298 #endif
299 }
300 }
301 }
302 }
304 /* lets see how the matching went. */
305 if (first_full == 0)
306 chosen_rra = best_full_rra;
307 else if (first_part == 0)
308 chosen_rra = best_part_rra;
309 else {
310 rrd_set_error
311 ("the RRD does not contain an RRA matching the chosen CF");
312 goto err_free_all_ds_namv;
313 }
315 /* set the wish parameters to their real values */
316 *step = rrd.stat_head->pdp_step * rrd.rra_def[chosen_rra].pdp_cnt;
317 *start -= (*start % *step);
318 *end += (*step - *end % *step);
319 rows = (*end - *start) / *step + 1;
321 #ifdef DEBUG
322 fprintf(stderr,
323 "We found: start %10lu end %10lu step %5lu rows %lu\n",
324 *start, *end, *step, rows);
325 #endif
327 /* Start and end are now multiples of the step size. The amount of
328 ** steps we want is (end-start)/step and *not* an extra one.
329 ** Reasoning: if step is s and we want to graph from t to t+s,
330 ** we need exactly ((t+s)-t)/s rows. The row to collect from the
331 ** database is the one with time stamp (t+s) which means t to t+s.
332 */
333 *ds_cnt = rrd.stat_head->ds_cnt;
334 if (((*data) = malloc(*ds_cnt * rows * sizeof(rrd_value_t))) == NULL) {
335 rrd_set_error("malloc fetch data area");
336 goto err_free_all_ds_namv;
337 }
339 data_ptr = (*data);
341 /* find base address of rra */
342 rra_base = rrd_file->header_len;
343 for (i = 0; i < chosen_rra; i++)
344 rra_base += (*ds_cnt * rrd.rra_def[i].row_cnt * sizeof(rrd_value_t));
346 /* find start and end offset */
347 rra_end_time = (rrd.live_head->last_up
348 - (rrd.live_head->last_up % *step));
349 rra_start_time = (rra_end_time
350 - (*step * (rrd.rra_def[chosen_rra].row_cnt - 1)));
351 /* here's an error by one if we don't be careful */
352 start_offset = (long) (*start + *step - rra_start_time) / (long) *step;
353 end_offset = (long) (rra_end_time - *end) / (long) *step;
354 #ifdef DEBUG
355 fprintf(stderr,
356 "rra_start %lu, rra_end %lu, start_off %li, end_off %li\n",
357 rra_start_time, rra_end_time, start_offset, end_offset);
358 #endif
360 /* fill the gap at the start if needs be */
362 if (start_offset <= 0)
363 rra_pointer = rrd.rra_ptr[chosen_rra].cur_row + 1;
364 else
365 rra_pointer = rrd.rra_ptr[chosen_rra].cur_row + 1 + start_offset;
367 if (rrd_seek(rrd_file, (rra_base + (rra_pointer * (*ds_cnt)
368 * sizeof(rrd_value_t))),
369 SEEK_SET) != 0) {
370 rrd_set_error("seek error in RRA");
371 goto err_free_data;
372 }
373 #ifdef DEBUG
374 fprintf(stderr, "First Seek: rra_base %lu rra_pointer %lu\n",
375 rra_base, rra_pointer);
376 #endif
377 /* step trough the array */
379 for (i = start_offset;
380 i < (signed) rrd.rra_def[chosen_rra].row_cnt - end_offset; i++) {
381 /* no valid data yet */
382 if (i < 0) {
383 #ifdef DEBUG
384 fprintf(stderr, "pre fetch %li -- ", i);
385 #endif
386 for (ii = 0; (unsigned) ii < *ds_cnt; ii++) {
387 *(data_ptr++) = DNAN;
388 #ifdef DEBUG
389 fprintf(stderr, "%10.2f ", *(data_ptr - 1));
390 #endif
391 }
392 }
393 /* past the valid data area */
394 else if (i >= (signed) rrd.rra_def[chosen_rra].row_cnt) {
395 #ifdef DEBUG
396 fprintf(stderr, "past fetch %li -- ", i);
397 #endif
398 for (ii = 0; (unsigned) ii < *ds_cnt; ii++) {
399 *(data_ptr++) = DNAN;
400 #ifdef DEBUG
401 fprintf(stderr, "%10.2f ", *(data_ptr - 1));
402 #endif
403 }
404 } else {
405 /* OK we are inside the valid area but the pointer has to
406 * be wrapped*/
407 if (rra_pointer >= (signed) rrd.rra_def[chosen_rra].row_cnt) {
408 rra_pointer -= rrd.rra_def[chosen_rra].row_cnt;
409 if (rrd_seek(rrd_file, (rra_base + rra_pointer * (*ds_cnt)
410 * sizeof(rrd_value_t)),
411 SEEK_SET) != 0) {
412 rrd_set_error("wrap seek in RRA did fail");
413 goto err_free_data;
414 }
415 #ifdef DEBUG
416 fprintf(stderr, "wrap seek ...\n");
417 #endif
418 }
420 if (rrd_read(rrd_file, data_ptr, sizeof(rrd_value_t) * (*ds_cnt))
421 != (ssize_t) (sizeof(rrd_value_t) * (*ds_cnt))) {
422 rrd_set_error("fetching cdp from rra");
423 goto err_free_data;
424 }
425 #ifdef HAVE_POSIX_FADVISE
426 /* don't pollute the buffer cache with data read from the file. We do this while reading to
427 keep damage minimal */
428 if (0 !=
429 posix_fadvise(rrd_file->fd, rrd_file->header_len, 0,
430 POSIX_FADV_DONTNEED)) {
431 rrd_set_error("setting POSIX_FADV_DONTNEED on '%s': %s",
432 filename, rrd_strerror(errno));
433 goto err_close; /*XXX: should use err_free_all_ds_namv */
434 }
435 #endif
437 #ifdef DEBUG
438 fprintf(stderr, "post fetch %li -- ", i);
439 for (ii = 0; ii < *ds_cnt; ii++)
440 fprintf(stderr, "%10.2f ", *(data_ptr + ii));
441 #endif
442 data_ptr += *ds_cnt;
443 rra_pointer++;
444 }
445 #ifdef DEBUG
446 fprintf(stderr, "\n");
447 #endif
449 }
450 #ifdef HAVE_POSIX_FADVISE
451 /* and just to be sure we drop everything except the header at the end */
452 if (0 !=
453 posix_fadvise(rrd_file->fd, rrd_file->header_len, 0,
454 POSIX_FADV_DONTNEED)) {
455 rrd_set_error("setting POSIX_FADV_DONTNEED on '%s': %s", filename,
456 rrd_strerror(errno));
457 goto err_free; /*XXX: should use err_free_all_ds_namv */
458 }
459 #endif
460 rrd_close(rrd_file);
461 return (0);
462 err_free_data:
463 free(*data);
464 *data = NULL;
465 err_free_all_ds_namv:
466 for (i = 0; (unsigned long) i < rrd.stat_head->ds_cnt; ++i)
467 free((*ds_namv)[i]);
468 err_free_ds_namv:
469 free(*ds_namv);
470 err_close:
471 rrd_close(rrd_file);
472 err_free:
473 rrd_free(&rrd);
474 return (-1);
475 }