* progress in moving all the fileaccess over to a wrapper system that can do fd based...
[rrdtool.git] / src / rrd_fetch.c
index 3bb9708..4b0f2b5 100644 (file)
@@ -1,10 +1,49 @@
 /*****************************************************************************
- * RRDtool 1.0.33  Copyright Tobias Oetiker, 1997 - 2000
+ * RRDtool 1.2.23  Copyright by Tobi Oetiker, 1997-2007
  *****************************************************************************
  * rrd_fetch.c  read date from an rrd to use for further processing
  *****************************************************************************
  * $Id$
  * $Log$
+ * Revision 1.8  2004/05/18 18:53:03  oetiker
+ * big spell checking patch -- slif@bellsouth.net
+ *
+ * Revision 1.7  2003/11/11 19:46:21  oetiker
+ * replaced time_value with rrd_time_value as MacOS X introduced a struct of that name in their standard headers
+ *
+ * Revision 1.6  2003/01/16 23:27:54  oetiker
+ * fix border condition in rra selection of rrd_fetch
+ * -- Stanislav Sinyagin <ssinyagin@yahoo.com>
+ *
+ * Revision 1.5  2002/06/23 22:29:40  alex
+ * Added "step=1800" and such to "DEF"
+ * Cleaned some of the signed vs. unsigned problems
+ *
+ * Revision 1.4  2002/02/01 20:34:49  oetiker
+ * fixed version number and date/time
+ *
+ * Revision 1.3  2001/12/24 06:51:49  alex
+ * A patch of size 44Kbytes... in short:
+ *
+ * Found and repaired the off-by-one error in rrd_fetch_fn().
+ * As a result I had to remove the hacks in rrd_fetch_fn(),
+ * rrd_tool.c, vdef_calc(), data_calc(), data_proc() and
+ * reduce_data().  There may be other places which I didn't
+ * find so be careful.
+ *
+ * Enhanced debugging in rrd_fetch_fn(), it shows the RRA selection
+ * process.
+ *
+ * Added the ability to print VDEF timestamps.  At the moment it
+ * is a hack, I needed it now to fix the off-by-one error.
+ * If the format string is "%c" (and nothing else!), the time
+ * will be printed by both ctime() and as a long int.
+ *
+ * Moved some code around (slightly altering it) from rrd_graph()
+ *   initializing     now in rrd_graph_init()
+ *   options parsing  now in rrd_graph_options()
+ *   script parsing   now in rrd_graph_script()
+ *
  * Revision 1.2  2001/12/17 12:48:43  oetiker
  * fix overflow error ...
  *
@@ -14,6 +53,8 @@
  *****************************************************************************/
 
 #include "rrd_tool.h"
+
+#include "rrd_is_thread_safe.h"
 /*#define DEBUG*/
 
 int
@@ -32,10 +73,11 @@ rrd_fetch(int argc,
 
     long     step_tmp =1;
     time_t   start_tmp=0, end_tmp=0;
-    enum     cf_en cf_idx;
+    const char *cf;
 
-    struct time_value start_tv, end_tv;
+    struct rrd_time_value start_tv, end_tv;
     char     *parsetime_error = NULL;
+    optind = 0; opterr = 0;  /* initialize getopt */
 
     /* init start and end time */
     parsetime("end-24h", &start_tv);
@@ -108,19 +150,39 @@ rrd_fetch(int argc,
        rrd_set_error("not enough arguments");
        return -1;
     }
-    
-    if ((cf_idx=cf_conv(argv[optind+1])) == -1 ){
-       return -1;
-    }
 
-    if (rrd_fetch_fn(argv[optind],cf_idx,start,end,step,ds_cnt,ds_namv,data) == -1)
+    cf = argv[optind+1];
+
+    if (rrd_fetch_r(argv[optind],cf,start,end,step,ds_cnt,ds_namv,data) == -1)
        return(-1);
     return (0);
 }
 
 int
+rrd_fetch_r(
+    const char           *filename,  /* name of the rrd */
+    const char           *cf,        /* which consolidation function ?*/
+    time_t         *start,
+    time_t         *end,       /* which time frame do you want ?
+                                * will be changed to represent reality */
+    unsigned long  *step,      /* which stepsize do you want? 
+                                * will be changed to represent reality */
+    unsigned long  *ds_cnt,    /* number of data sources in file */
+    char           ***ds_namv, /* names of data_sources */
+    rrd_value_t    **data)     /* two dimensional array containing the data */
+{
+    enum     cf_en cf_idx;
+
+    if ((int)(cf_idx=cf_conv(cf)) == -1 ){
+        return -1;
+    }
+
+    return (rrd_fetch_fn(filename,cf_idx,start,end,step,ds_cnt,ds_namv,data));
+}
+
+int
 rrd_fetch_fn(
-    char           *filename,  /* name of the rrd */
+    const char     *filename,  /* name of the rrd */
     enum cf_en     cf_idx,         /* which consolidation function ?*/
     time_t         *start,
     time_t         *end,       /* which time frame do you want ?
@@ -132,36 +194,49 @@ rrd_fetch_fn(
     rrd_value_t    **data)     /* two dimensional array containing the data */
 {
     long           i,ii;
-    FILE           *in_file;
     time_t         cal_start,cal_end, rra_start_time,rra_end_time;
     long  best_full_rra=0, best_part_rra=0, chosen_rra=0, rra_pointer=0;
-    long  best_step_diff=0, tmp_step_diff=0, tmp_match=0, best_match=0;
+    long  best_full_step_diff=0, best_part_step_diff=0, tmp_step_diff=0, tmp_match=0, best_match=0;
     long  full_match, rra_base;
     long           start_offset, end_offset;
     int            first_full = 1;
     int            first_part = 1;
     rrd_t     rrd;
+    rrd_file_t     *rrd_file;
     rrd_value_t    *data_ptr;
     unsigned long  rows;
+#ifdef HAVE_POSIX_FADVISE
+    long  rrd_head_size;
+#endif
+
+#ifdef DEBUG
+fprintf(stderr,"Entered rrd_fetch_fn() searching for the best match\n");
+fprintf(stderr,"Looking for: start %10lu end %10lu step %5lu\n",
+                                               *start,*end,*step);
+#endif
 
-    if(rrd_open(filename,&in_file,&rrd, RRD_READONLY)==-1)
+    rrd_file = rrd_open(filename,&rrd, RRD_READONLY);
+    if (rrd_file == NULL)
        return(-1);
-    
-    /* when was the realy last update of this file ? */
+
+#ifdef HAVE_POSIX_FADVISE
+    rrd_head_size = rrd_file->header_len;
+#endif
+    /* when was the really last update of this file ? */
 
     if (((*ds_namv) = (char **) malloc(rrd.stat_head->ds_cnt * sizeof(char*)))==NULL){
        rrd_set_error("malloc fetch ds_namv array");
        rrd_free(&rrd);
-       fclose(in_file);
+       close(rrd_file->fd);
        return(-1);
     }
     
-    for(i=0;i<rrd.stat_head->ds_cnt;i++){
+    for(i=0;(unsigned long)i<rrd.stat_head->ds_cnt;i++){
        if ((((*ds_namv)[i]) = malloc(sizeof(char) * DS_NAM_SIZE))==NULL){
            rrd_set_error("malloc fetch ds_namv entry");
            rrd_free(&rrd);
            free(*ds_namv);
-           fclose(in_file);
+           close(rrd_file->fd);
            return(-1);
        }
        strncpy((*ds_namv)[i],rrd.ds_def[i].ds_nam,DS_NAM_SIZE-1);
@@ -170,10 +245,9 @@ rrd_fetch_fn(
     }
     
     /* find the rra which best matches the requirements */
-    for(i=0;i<rrd.stat_head->rra_cnt;i++){
+    for(i=0;(unsigned)i<rrd.stat_head->rra_cnt;i++){
        if(cf_conv(rrd.rra_def[i].cf_nam) == cf_idx){
            
-              
            cal_end = (rrd.live_head->last_up - (rrd.live_head->last_up 
                          % (rrd.rra_def[i].pdp_cnt 
                             * rrd.stat_head->pdp_step)));
@@ -183,16 +257,29 @@ rrd_fetch_fn(
                            * rrd.stat_head->pdp_step));
 
            full_match = *end -*start;
+#ifdef DEBUG
+fprintf(stderr,"Considering: start %10lu end %10lu step %5lu ",
+                                                       cal_start,cal_end,
+                       rrd.stat_head->pdp_step * rrd.rra_def[i].pdp_cnt);
+#endif
+           /* we need step difference in either full or partial case */
+           tmp_step_diff = labs(*step - (rrd.stat_head->pdp_step
+                                          * rrd.rra_def[i].pdp_cnt));
            /* best full match */
            if(cal_end >= *end 
               && cal_start <= *start){
-               tmp_step_diff = labs(*step - (rrd.stat_head->pdp_step
-                                        * rrd.rra_def[i].pdp_cnt));
-               if (first_full || (tmp_step_diff < best_step_diff)){
+               if (first_full || (tmp_step_diff < best_full_step_diff)){
                    first_full=0;
-                   best_step_diff = tmp_step_diff;
+                   best_full_step_diff = tmp_step_diff;
                    best_full_rra=i;
-               } 
+#ifdef DEBUG
+fprintf(stderr,"best full match so far\n");
+#endif
+               } else {
+#ifdef DEBUG
+fprintf(stderr,"full match, not best\n");
+#endif
+               }
                
            } else {
                /* best partial match */
@@ -201,17 +288,27 @@ rrd_fetch_fn(
                    tmp_match -= (cal_start-*start);
                if (cal_end<*end)
                    tmp_match -= (*end-cal_end);                
-               if (first_part || best_match < tmp_match){
+               if (first_part ||
+                    (best_match < tmp_match) ||
+                    (best_match == tmp_match && 
+                     tmp_step_diff < best_part_step_diff)){ 
+#ifdef DEBUG
+fprintf(stderr,"best partial so far\n");
+#endif
                    first_part=0;
                    best_match = tmp_match;
+                   best_part_step_diff = tmp_step_diff;
                    best_part_rra =i;
-               } 
+               } else {
+#ifdef DEBUG
+fprintf(stderr,"partial match, not best\n");
+#endif
+               }
            }
        }
     }
 
     /* lets see how the matching went. */
-    
     if (first_full==0)
        chosen_rra = best_full_rra;
     else if (first_part==0)
@@ -219,37 +316,42 @@ rrd_fetch_fn(
     else {
        rrd_set_error("the RRD does not contain an RRA matching the chosen CF");
        rrd_free(&rrd);
-       fclose(in_file);
+       close(rrd_file->fd);
        return(-1);
     }
        
     /* set the wish parameters to their real values */
-    
     *step = rrd.stat_head->pdp_step * rrd.rra_def[chosen_rra].pdp_cnt;
     *start -= (*start % *step);
-    if (*end % *step) *end += (*step - *end % *step);
-    rows = (*end - *start) / *step +1;
+    *end += (*step - *end % *step);
+    rows = (*end - *start) / *step + 1;
 
 #ifdef DEBUG
-    fprintf(stderr,"start %lu end %lu step %lu rows  %lu\n",
-           *start,*end,*step,rows);
+    fprintf(stderr,"We found:    start %10lu end %10lu step %5lu rows  %lu\n",
+                                               *start,*end,*step,rows);
 #endif
 
+/* Start and end are now multiples of the step size.  The amount of
+** steps we want is (end-start)/step and *not* an extra one.
+** Reasoning:  if step is s and we want to graph from t to t+s,
+** we need exactly ((t+s)-t)/s rows.  The row to collect from the
+** database is the one with time stamp (t+s) which means t to t+s.
+*/
     *ds_cnt =   rrd.stat_head->ds_cnt; 
     if (((*data) = malloc(*ds_cnt * rows * sizeof(rrd_value_t)))==NULL){
        rrd_set_error("malloc fetch data area");
-       for (i=0;i<*ds_cnt;i++)
+       for (i=0;(unsigned long)i<*ds_cnt;i++)
              free((*ds_namv)[i]);
        free(*ds_namv);
        rrd_free(&rrd);
-       fclose(in_file);
+       close(rrd_file->fd);
        return(-1);
     }
     
     data_ptr=(*data);
     
     /* find base address of rra */
-    rra_base=ftell(in_file);
+    rra_base = rrd_file->header_len;
     for(i=0;i<chosen_rra;i++)
        rra_base += ( *ds_cnt
                      * rrd.rra_def[i].row_cnt
@@ -260,7 +362,8 @@ rrd_fetch_fn(
                    - (rrd.live_head->last_up % *step));
     rra_start_time = (rra_end_time
                 - ( *step * (rrd.rra_def[chosen_rra].row_cnt-1)));
-    start_offset =(long)(*start - rra_start_time) / (long)*step;
+    /* here's an error by one if we don't be careful */
+    start_offset =(long)(*start + *step - rra_start_time) / (long)*step;
     end_offset = (long)(rra_end_time - *end ) / (long)*step; 
 #ifdef DEBUG
     fprintf(stderr,"rra_start %lu, rra_end %lu, start_off %li, end_off %li\n",
@@ -274,18 +377,18 @@ rrd_fetch_fn(
     else 
        rra_pointer = rrd.rra_ptr[chosen_rra].cur_row+1+start_offset;
     
-    if(fseek(in_file,(rra_base 
+    if(rrd_seek(rrd_file,(rra_base 
                   + (rra_pointer
                      * *ds_cnt
                      * sizeof(rrd_value_t))),SEEK_SET) != 0){
        rrd_set_error("seek error in RRA");
-       for (i=0;i<*ds_cnt;i++)
+       for (i=0;(unsigned)i<*ds_cnt;i++)
              free((*ds_namv)[i]);
        free(*ds_namv);
        rrd_free(&rrd);
        free(*data);
        *data = NULL;
-       fclose(in_file);
+       close(rrd_file->fd);
        return(-1);
 
     }
@@ -296,14 +399,14 @@ rrd_fetch_fn(
     /* step trough the array */
 
     for (i=start_offset;
-        i<(long)(rrd.rra_def[chosen_rra].row_cnt-end_offset);
+        i< (signed)rrd.rra_def[chosen_rra].row_cnt - end_offset;
         i++){
        /* no valid data yet */
        if (i<0) {
 #ifdef DEBUG
            fprintf(stderr,"pre fetch %li -- ",i);
 #endif
-           for(ii=0;ii<*ds_cnt;ii++){
+           for(ii=0;(unsigned)ii<*ds_cnt;ii++){
                *(data_ptr++) = DNAN;
 #ifdef DEBUG
                fprintf(stderr,"%10.2f ",*(data_ptr-1));
@@ -311,11 +414,11 @@ rrd_fetch_fn(
            }
        } 
        /* past the valid data area */
-       else if (i>=rrd.rra_def[chosen_rra].row_cnt) {
+       else if (i >= (signed)rrd.rra_def[chosen_rra].row_cnt) {
 #ifdef DEBUG
            fprintf(stderr,"post fetch %li -- ",i);
 #endif
-           for(ii=0;ii<*ds_cnt;ii++){
+           for(ii=0;(unsigned)ii<*ds_cnt;ii++){
                *(data_ptr++) = DNAN;
 #ifdef DEBUG
                fprintf(stderr,"%10.2f ",*(data_ptr-1));
@@ -324,39 +427,49 @@ rrd_fetch_fn(
        } else {
            /* OK we are inside the valid area but the pointer has to 
             * be wrapped*/
-           if (rra_pointer >= rrd.rra_def[chosen_rra].row_cnt) {
+           if (rra_pointer >= (signed)rrd.rra_def[chosen_rra].row_cnt) {
                rra_pointer -= rrd.rra_def[chosen_rra].row_cnt;
-               if(fseek(in_file,(rra_base+rra_pointer
+               if(rrd_seek(rrd_file,(rra_base+rra_pointer
                               * *ds_cnt
                               * sizeof(rrd_value_t)),SEEK_SET) != 0){
                    rrd_set_error("wrap seek in RRA did fail");
-                   for (ii=0;ii<*ds_cnt;ii++)
+                   for (ii=0;(unsigned)ii<*ds_cnt;ii++)
                        free((*ds_namv)[ii]);
                    free(*ds_namv);
                    rrd_free(&rrd);
                    free(*data);
                    *data = NULL;
-                   fclose(in_file);
+                   close(rrd_file->fd);
                    return(-1);
                }
 #ifdef DEBUG
                fprintf(stderr,"wrap seek ...\n");
-#endif     
+#endif
            }
-           
-           if(fread(data_ptr,
-                    sizeof(rrd_value_t),
-                    *ds_cnt,in_file) != rrd.stat_head->ds_cnt){
+
+           if(rrd_read(rrd_file,data_ptr,
+                    sizeof(rrd_value_t)* (*ds_cnt))
+                   != (ssize_t)(sizeof(rrd_value_t)*(*ds_cnt)*rrd.stat_head->ds_cnt)){
                rrd_set_error("fetching cdp from rra");
-               for (ii=0;ii<*ds_cnt;ii++)
+               for (ii=0;(unsigned)ii<*ds_cnt;ii++)
                    free((*ds_namv)[ii]);
                free(*ds_namv);
                rrd_free(&rrd);
                free(*data);
                *data = NULL;
-               fclose(in_file);
+               close(rrd_file->fd);
                return(-1);
            }
+#ifdef HAVE_POSIX_FADVISE
+       /* don't pollute the buffer cache with data read from the file. We do this while reading to 
+          keep damage minimal */
+       if (0 != posix_fadvise(rrd_file->fd, rrd_head_size, 0, POSIX_FADV_DONTNEED)) {
+           rrd_set_error("setting POSIX_FADV_DONTNEED on '%s': %s",filename, rrd_strerror(errno));
+           close(rrd_file->fd);
+           return(-1);
+       } 
+#endif
+
 #ifdef DEBUG
            fprintf(stderr,"post fetch %li -- ",i);
            for(ii=0;ii<*ds_cnt;ii++)
@@ -371,6 +484,14 @@ rrd_fetch_fn(
        
     }
     rrd_free(&rrd);
-    fclose(in_file);
+#ifdef HAVE_POSIX_FADVISE
+    /* and just to be sure we drop everything except the header at the end */
+    if (0 != posix_fadvise(rrd_file->fd, rrd_head_size, 0, POSIX_FADV_DONTNEED)) {
+           rrd_set_error("setting POSIX_FADV_DONTNEED on '%s': %s",filename, rrd_strerror(errno));
+           close(rrd_file->fd);
+           return(-1);
+    } 
+#endif     
+    close(rrd_file->fd);
     return(0);
 }