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 *****************************************************************************
8 * Revision 1.8 2004/05/18 18:53:03 oetiker
9 * big spell checking patch -- slif@bellsouth.net
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
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>
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
22 * Revision 1.4 2002/02/01 20:34:49 oetiker
23 * fixed version number and date/time
25 * Revision 1.3 2001/12/24 06:51:49 alex
26 * A patch of size 44Kbytes... in short:
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
34 * Enhanced debugging in rrd_fetch_fn(), it shows the RRA selection
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.
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()
47 * Revision 1.2 2001/12/17 12:48:43 oetiker
48 * fix overflow error ...
50 * Revision 1.1.1.1 2001/02/25 22:25:05 oetiker
53 *****************************************************************************/
57 #include "rrd_is_thread_safe.h"
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 */
71 { /* two dimensional array containing the data */
75 time_t start_tmp = 0, end_tmp = 0;
78 struct rrd_time_value start_tv, end_tv;
79 char *parsetime_error = NULL;
82 opterr = 0; /* initialize getopt */
84 /* init start and end time */
85 parsetime("end-24h", &start_tv);
86 parsetime("now", &end_tv);
89 static struct option long_options[] = {
90 {"resolution", required_argument, 0, 'r'},
91 {"start", required_argument, 0, 's'},
92 {"end", required_argument, 0, 'e'},
98 opt = getopt_long(argc, argv, "r:s:e:", long_options, &option_index);
105 if ((parsetime_error = parsetime(optarg, &start_tv))) {
106 rrd_set_error("start time: %s", parsetime_error);
111 if ((parsetime_error = parsetime(optarg, &end_tv))) {
112 rrd_set_error("end time: %s", parsetime_error);
117 step_tmp = atol(optarg);
120 rrd_set_error("unknown option '-%c'", optopt);
126 if (proc_start_end(&start_tv, &end_tv, &start_tmp, &end_tmp) == -1) {
131 if (start_tmp < 3600 * 24 * 365 * 10) {
132 rrd_set_error("the first entry to fetch should be after 1980");
136 if (end_tmp < start_tmp) {
137 rrd_set_error("start (%ld) should be less than end (%ld)", start_tmp,
146 rrd_set_error("step must be >= 1 second");
151 if (optind + 1 >= argc) {
152 rrd_set_error("not enough arguments");
156 cf = argv[optind + 1];
158 if (rrd_fetch_r(argv[optind], cf, start, end, step, ds_cnt, ds_namv, data)
165 const char *filename, /* name of the rrd */
166 const char *cf, /* which consolidation function ? */
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 */
175 { /* two dimensional array containing the data */
178 if ((int) (cf_idx = cf_conv(cf)) == -1) {
183 (filename, cf_idx, start, end, step, ds_cnt, ds_namv, data));
187 const char *filename, /* name of the rrd */
188 enum cf_en cf_idx, /* which consolidation function ? */
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 */
197 { /* two dimensional array containing the data */
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 =
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;
209 rrd_file_t *rrd_file;
210 rrd_value_t *data_ptr;
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);
219 rrd_file = rrd_open(filename, &rrd, RRD_READONLY);
220 if (rrd_file == NULL)
223 /* when was the really last update of this file ? */
226 (char **) malloc(rrd.stat_head->ds_cnt * sizeof(char *))) == NULL) {
227 rrd_set_error("malloc fetch ds_namv array");
233 for (i = 0; (unsigned long) i < rrd.stat_head->ds_cnt; i++) {
234 if ((((*ds_namv)[i]) = malloc(sizeof(char) * DS_NAM_SIZE)) == NULL) {
235 rrd_set_error("malloc fetch ds_namv entry");
241 strncpy((*ds_namv)[i], rrd.ds_def[i].ds_nam, DS_NAM_SIZE - 1);
242 (*ds_namv)[i][DS_NAM_SIZE - 1] = '\0';
246 /* find the rra which best matches the requirements */
247 for (i = 0; (unsigned) i < rrd.stat_head->rra_cnt; i++) {
248 if (cf_conv(rrd.rra_def[i].cf_nam) == cf_idx) {
250 cal_end = (rrd.live_head->last_up - (rrd.live_head->last_up
251 % (rrd.rra_def[i].pdp_cnt
257 (rrd.rra_def[i].pdp_cnt * rrd.rra_def[i].row_cnt *
258 rrd.stat_head->pdp_step));
260 full_match = *end - *start;
262 fprintf(stderr, "Considering: start %10lu end %10lu step %5lu ",
264 rrd.stat_head->pdp_step * rrd.rra_def[i].pdp_cnt);
266 /* we need step difference in either full or partial case */
267 tmp_step_diff = labs(*step - (rrd.stat_head->pdp_step
268 * rrd.rra_def[i].pdp_cnt));
269 /* best full match */
270 if (cal_end >= *end && cal_start <= *start) {
271 if (first_full || (tmp_step_diff < best_full_step_diff)) {
273 best_full_step_diff = tmp_step_diff;
276 fprintf(stderr, "best full match so far\n");
280 fprintf(stderr, "full match, not best\n");
285 /* best partial match */
286 tmp_match = full_match;
287 if (cal_start > *start)
288 tmp_match -= (cal_start - *start);
290 tmp_match -= (*end - cal_end);
292 (best_match < tmp_match) ||
293 (best_match == tmp_match &&
294 tmp_step_diff < best_part_step_diff)) {
296 fprintf(stderr, "best partial so far\n");
299 best_match = tmp_match;
300 best_part_step_diff = tmp_step_diff;
304 fprintf(stderr, "partial match, not best\n");
311 /* lets see how the matching went. */
313 chosen_rra = best_full_rra;
314 else if (first_part == 0)
315 chosen_rra = best_part_rra;
318 ("the RRD does not contain an RRA matching the chosen CF");
324 /* set the wish parameters to their real values */
325 *step = rrd.stat_head->pdp_step * rrd.rra_def[chosen_rra].pdp_cnt;
326 *start -= (*start % *step);
327 *end += (*step - *end % *step);
328 rows = (*end - *start) / *step + 1;
332 "We found: start %10lu end %10lu step %5lu rows %lu\n",
333 *start, *end, *step, rows);
336 /* Start and end are now multiples of the step size. The amount of
337 ** steps we want is (end-start)/step and *not* an extra one.
338 ** Reasoning: if step is s and we want to graph from t to t+s,
339 ** we need exactly ((t+s)-t)/s rows. The row to collect from the
340 ** database is the one with time stamp (t+s) which means t to t+s.
342 *ds_cnt = rrd.stat_head->ds_cnt;
343 if (((*data) = malloc(*ds_cnt * rows * sizeof(rrd_value_t))) == NULL) {
344 rrd_set_error("malloc fetch data area");
345 for (i = 0; (unsigned long) i < *ds_cnt; i++)
355 /* find base address of rra */
356 rra_base = rrd_file->header_len;
357 for (i = 0; i < chosen_rra; i++)
358 rra_base += (*ds_cnt * rrd.rra_def[i].row_cnt * sizeof(rrd_value_t));
360 /* find start and end offset */
361 rra_end_time = (rrd.live_head->last_up
362 - (rrd.live_head->last_up % *step));
363 rra_start_time = (rra_end_time
364 - (*step * (rrd.rra_def[chosen_rra].row_cnt - 1)));
365 /* here's an error by one if we don't be careful */
366 start_offset = (long) (*start + *step - rra_start_time) / (long) *step;
367 end_offset = (long) (rra_end_time - *end) / (long) *step;
370 "rra_start %lu, rra_end %lu, start_off %li, end_off %li\n",
371 rra_start_time, rra_end_time, start_offset, end_offset);
374 /* fill the gap at the start if needs be */
376 if (start_offset <= 0)
377 rra_pointer = rrd.rra_ptr[chosen_rra].cur_row + 1;
379 rra_pointer = rrd.rra_ptr[chosen_rra].cur_row + 1 + start_offset;
381 if (rrd_seek(rrd_file, (rra_base
384 * sizeof(rrd_value_t))), SEEK_SET) != 0) {
385 rrd_set_error("seek error in RRA");
386 for (i = 0; (unsigned) i < *ds_cnt; i++)
397 fprintf(stderr, "First Seek: rra_base %lu rra_pointer %lu\n",
398 rra_base, rra_pointer);
400 /* step trough the array */
402 for (i = start_offset;
403 i < (signed) rrd.rra_def[chosen_rra].row_cnt - end_offset; i++) {
404 /* no valid data yet */
407 fprintf(stderr, "pre fetch %li -- ", i);
409 for (ii = 0; (unsigned) ii < *ds_cnt; ii++) {
410 *(data_ptr++) = DNAN;
412 fprintf(stderr, "%10.2f ", *(data_ptr - 1));
416 /* past the valid data area */
417 else if (i >= (signed) rrd.rra_def[chosen_rra].row_cnt) {
419 fprintf(stderr, "past fetch %li -- ", i);
421 for (ii = 0; (unsigned) ii < *ds_cnt; ii++) {
422 *(data_ptr++) = DNAN;
424 fprintf(stderr, "%10.2f ", *(data_ptr - 1));
428 /* OK we are inside the valid area but the pointer has to
430 if (rra_pointer >= (signed) rrd.rra_def[chosen_rra].row_cnt) {
431 rra_pointer -= rrd.rra_def[chosen_rra].row_cnt;
432 if (rrd_seek(rrd_file, (rra_base + rra_pointer
434 * sizeof(rrd_value_t)),
436 rrd_set_error("wrap seek in RRA did fail");
437 for (ii = 0; (unsigned) ii < *ds_cnt; ii++)
438 free((*ds_namv)[ii]);
447 fprintf(stderr, "wrap seek ...\n");
451 if (rrd_read(rrd_file, data_ptr, sizeof(rrd_value_t) * (*ds_cnt))
452 != (ssize_t) (sizeof(rrd_value_t) * (*ds_cnt))) {
453 rrd_set_error("fetching cdp from rra");
454 for (ii = 0; (unsigned) ii < *ds_cnt; ii++)
455 free((*ds_namv)[ii]);
463 #ifdef HAVE_POSIX_FADVISE
464 /* don't pollute the buffer cache with data read from the file. We do this while reading to
465 keep damage minimal */
467 posix_fadvise(rrd_file->fd, rrd_file->header_len, 0,
468 POSIX_FADV_DONTNEED)) {
469 rrd_set_error("setting POSIX_FADV_DONTNEED on '%s': %s",
470 filename, rrd_strerror(errno));
477 fprintf(stderr, "post fetch %li -- ", i);
478 for (ii = 0; ii < *ds_cnt; ii++)
479 fprintf(stderr, "%10.2f ", *(data_ptr + ii));
485 fprintf(stderr, "\n");
490 #ifdef HAVE_POSIX_FADVISE
491 /* and just to be sure we drop everything except the header at the end */
493 posix_fadvise(rrd_file->fd, rrd_file->header_len, 0,
494 POSIX_FADV_DONTNEED)) {
495 rrd_set_error("setting POSIX_FADV_DONTNEED on '%s': %s", filename,
496 rrd_strerror(errno));