summary | shortlog | log | commit | commitdiff | tree
raw | patch | inline | side by side (parent: caa9411)
raw | patch | inline | side by side (parent: caa9411)
author | oetiker <oetiker@a5681a0c-68f1-0310-ab6d-d61299d08faa> | |
Fri, 7 Nov 2008 14:09:29 +0000 (14:09 +0000) | ||
committer | oetiker <oetiker@a5681a0c-68f1-0310-ab6d-d61299d08faa> | |
Fri, 7 Nov 2008 14:09:29 +0000 (14:09 +0000) |
src/rrd_fetch_libdbi.c | [new file with mode: 0644] | patch | blob |
diff --git a/src/rrd_fetch_libdbi.c b/src/rrd_fetch_libdbi.c
--- /dev/null
+++ b/src/rrd_fetch_libdbi.c
@@ -0,0 +1,533 @@
+#include "rrd_tool.h"
+#include <dbi/dbi.h>
+#include <time.h>
+
+/* the structures */
+struct sql_table_helper {
+ dbi_conn conn;
+ int connected;
+ dbi_result result;
+ char const* filename;
+ char const* dbdriver;
+ char* table_start;
+ char* table_next;
+ char const* where;
+ char * timestamp;
+ char * value;
+};
+
+/* the prototypes */
+int _sql_setparam(struct sql_table_helper* th,char* key, char* value);
+int _sql_fetchrow(struct sql_table_helper* th,time_t *timestamp, rrd_value_t *value);
+char* _find_next_separator(char* start,char separator);
+char* _find_next_separator_twice(char*start,char separator);
+char _hexcharhelper(char c);
+int _inline_unescape (char* string);
+double rrd_fetch_dbi_double(dbi_result *result,int idx);
+long rrd_fetch_dbi_long(dbi_result *result,int idx);
+
+/* the real code */
+
+/* helpers to get correctly converted values from DB*/
+long rrd_fetch_dbi_long(dbi_result *result,int idx) {
+ char *ptmp="";
+ switch (dbi_result_get_field_type_idx(result,idx)) {
+ case DBI_TYPE_STRING:
+ ptmp=(char*)dbi_result_get_string_idx(result,idx);
+ return atoi(ptmp);
+ default:
+ return dbi_result_get_longlong_idx(result,idx);
+ }
+}
+
+double rrd_fetch_dbi_double(dbi_result *result,int idx) {
+ char *ptmp="";
+ /* return NAN if NULL */
+ if(dbi_result_field_is_null_idx(result,idx)) { return DNAN; }
+ /* do some conversions */
+ switch (dbi_result_get_field_type_idx(result,idx)) {
+ case DBI_TYPE_STRING:
+ ptmp=(char*)dbi_result_get_string_idx(result,idx);
+ return strtod(ptmp,NULL);
+ default:
+ return dbi_result_get_double_idx(result,idx);
+ }
+}
+
+int _sql_close(struct sql_table_helper* th) {
+ /* close only if connected */
+ if (th->conn) {
+ /* shutdown dbi */
+ dbi_conn_close(th->conn);
+ dbi_shutdown();
+ /* and assign empty */
+ th->conn=NULL;
+ }
+}
+
+int _sql_setparam(struct sql_table_helper* th,char* key, char* value) {
+ char* dbi_errstr=NULL;
+ /* if not connected */
+ if (! th->conn) {
+ /* initialize some stuff */
+ th->table_next=th->table_start;
+ th->result=NULL;
+ th->connected=0;
+ /* initialize db */
+ dbi_initialize(NULL);
+ th->conn=dbi_conn_new(th->dbdriver);
+ /* and handle errors */
+ if (! th->conn) {
+ dbi_conn_error(th->conn,(const char**)&dbi_errstr);
+ rrd_set_error( "libdbi - no such driver: %s (possibly a dynamic link problem of the driver being linked without -ldbi)",dbi_errstr);
+ dbi_shutdown();
+ return -1;
+ }
+ }
+ if (th->connected) {
+ rrd_set_error( "we are already connected - can not set parameter %s=%s",key,value);
+ _sql_close(th);
+ return -1;
+ }
+ if (dbi_conn_set_option(th->conn,key,value)) {
+ dbi_conn_error(th->conn,(const char**)&dbi_errstr);
+ rrd_set_error( "libdbi: problems setting %s to %s - %s",key,value,dbi_errstr);
+ _sql_close(th);
+ return -1;
+ }
+ return 0;
+}
+
+int _sql_fetchrow(struct sql_table_helper* th,time_t *timestamp, rrd_value_t *value) {
+ char* dbi_errstr=NULL;
+ char sql[10240];
+ time_t startt=0,endt=0;
+ /*connect to the database if needed */
+ if (! th->conn) {
+ rrd_set_error( "libdbi no parameters set for libdbi",th->filename,dbi_errstr);
+ return -1;
+ }
+ if (! th->connected) {
+ /* and now connect */
+ if (dbi_conn_connect(th->conn) <0) {
+ dbi_conn_error(th->conn,(const char**)&dbi_errstr);
+ rrd_set_error( "libdbi: problems connecting to db with connect string %s - error: %s",th->filename,dbi_errstr);
+ _sql_close(th);
+ return -1;
+ }
+ th->connected=1;
+ }
+ /* now find out regarding an existing result-set */
+ if (! th->result) {
+ /* return if table_next is NULL */
+ if (th->table_next==NULL) {
+ /* but first close connection */
+ _sql_close(th);
+ /* and return with end of data */
+ return 0;
+ }
+ /* calculate the table to use next */
+ th->table_start=th->table_next;
+ th->table_next=_find_next_separator(th->table_start,'+');
+ _inline_unescape(th->table_start);
+ /* and prepare FULL SQL Statement */
+ snprintf(sql,sizeof(sql)-1,"SELECT %s as rrd_time, %s as rrd_value FROM %s WHERE %s GROUP BY rrd_time",
+ th->timestamp,th->value,th->table_start,th->where);
+ /* and execute sql */
+ if (getenv("RRDDEBUGSQL")) { startt=time(NULL); fprintf(stderr,"RRDDEBUGSQL: %li: executing %s\n",startt,sql); }
+ th->result=dbi_conn_query(th->conn,sql);
+ if (startt) { endt=time(NULL);fprintf(stderr,"RRDDEBUGSQL: %li: timing %li\n",endt,endt-startt); }
+ /* handle error case */
+ if (! th->result) {
+ dbi_conn_error(th->conn,(const char**)&dbi_errstr);
+ if (startt) { fprintf(stderr,"RRDDEBUGSQL: %li: error %s\n",endt,dbi_errstr); }
+ rrd_set_error("libdbi: problems with query: %s - errormessage: %s",sql,dbi_errstr);
+ _sql_close(th);
+ return -1;
+ }
+ }
+ /* and now fetch key and value */
+ if (! dbi_result_next_row(th->result)) {
+ /* free result */
+ dbi_result_free(th->result);
+ th->result=NULL;
+ /* and call recursively - this will open the next table or close connection as a whole*/
+ return _sql_fetchrow(th,timestamp,value);
+ }
+ /* and return with flag for one value */
+ *timestamp=rrd_fetch_dbi_long(th->result,1);
+ *value=rrd_fetch_dbi_double(th->result,2);
+ return 1;
+}
+
+char* _find_next_separator(char* start,char separator) {
+ char* found=strchr(start,separator);
+ /* have we found it */
+ if (found) {
+ /* then 0 terminate current string */
+ *found=0;
+ /* and return the pointer past the separator */
+ return (found+1);
+ }
+ /* not found, so return NULL */
+ return NULL;
+}
+
+char* _find_next_separator_twice(char*start,char separator) {
+ char *found=start;
+ /* find next separator in string*/
+ while (found) {
+ /* if found and the next one is also a separator */
+ if (found[1] == separator) {
+ /* then 0 terminate current string */
+ *found=0;
+ /* and return the pointer past the current one*/
+ return (found+2);
+ }
+ /* find next occurance */
+ found=strchr(found+1,separator);
+ }
+ /* not found, so return NULL */
+ return NULL;
+}
+
+char _hexcharhelper(char c) {
+ switch (c) {
+ case '0': return 0 ; break;
+ case '1': return 1 ; break;
+ case '2': return 2 ; break;
+ case '3': return 3 ; break;
+ case '4': return 4 ; break;
+ case '5': return 5 ; break;
+ case '6': return 6 ; break;
+ case '7': return 7 ; break;
+ case '8': return 8 ; break;
+ case '9': return 9 ; break;
+ case 'a': return 10 ; break;
+ case 'b': return 11 ; break;
+ case 'c': return 12 ; break;
+ case 'd': return 13 ; break;
+ case 'e': return 14 ; break;
+ case 'f': return 15 ; break;
+ case 'A': return 10 ; break;
+ case 'B': return 11 ; break;
+ case 'C': return 12 ; break;
+ case 'D': return 13 ; break;
+ case 'E': return 14 ; break;
+ case 'F': return 15 ; break;
+ }
+ return -1;
+}
+
+int _inline_unescape (char* string) {
+ char *src=string;
+ char *dst=string;
+ char c,h1,h2;
+ while((c= *src)) {
+ src++;
+ if (c == '%') {
+ if (*src == '%') {
+ /* increase src pointer by 1 skiping second % */
+ src+=1;
+ } else {
+ /* try to calculate hex value from the next 2 values*/
+ h1=_hexcharhelper(*src);
+ if (h1<0) { rrd_set_error( "string escape error at: %s\n",string);return(1); }
+ h2=_hexcharhelper(*(src+1));
+ if (h2<0) { rrd_set_error( "string escape error at: %s\n",string);return(1); }
+ c=h2+(h1<<4);
+ /* increase src pointer by 2 skiping 2 chars */
+ src+=2;
+ }
+ }
+ *dst=c;
+ dst++;
+ }
+ *dst=0;
+ return 0;
+}
+
+int
+rrd_fetch_fn_libdbi(
+ 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 ?
+ * 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 */
+{
+ /* the separator used */
+ char separator='/';
+ /* a local copy of the filename - used for copying plus some pointer variables */
+ char filenameworkcopy[10240];
+ char *tmpptr=filenameworkcopy;
+ char *nextptr=NULL;
+ char *libdbiargs=NULL;
+ char *sqlargs=NULL;
+ /* the settings for the "works" of rrd */
+ int fillmissing=0;
+ unsigned long minstepsize=300;
+ /* the result-set */
+ long r_timestamp,l_timestamp,d_timestamp;
+ double r_value,l_value,d_value;
+ int r_status;
+ int rows;
+ long idx;
+ int derive=0;
+ /* the libdbi connection data and the table_help structure */
+ struct sql_table_helper table_help;
+ char where[10240];
+ table_help.conn=NULL;
+ table_help.where=where;
+
+ /* some loop variables */
+ int i=0;
+
+ /* check header */
+ if (strncmp("sql",filename,3)!=0) {
+ rrd_set_error( "formatstring wrong - %s",filename );return -1;
+ }
+ if (filename[3]!=filename[4]) {
+ rrd_set_error( "formatstring wrong - %s",filename );return -1;
+ }
+
+ /* now make this the separator */
+ separator=filename[3];
+
+ /* copy filename for local modifications during parsing */
+ strncpy(filenameworkcopy,filename+5,sizeof(filenameworkcopy));
+
+ /* get the driver */
+ table_help.dbdriver=tmpptr;
+ libdbiargs=_find_next_separator(tmpptr,separator);
+ if (! libdbiargs) {
+ /* error in argument */
+ rrd_set_error( "formatstring wrong as we did not find \"%c\"- %s",separator,table_help.dbdriver);
+ return -1;
+ }
+
+ /* now find the next double separator - this defines the args to the database */
+ sqlargs=_find_next_separator_twice(libdbiargs,separator);
+ if (!sqlargs) {
+ rrd_set_error( "formatstring wrong for db arguments as we did not find \"%c%c\" in \"%s\"",separator,separator,libdbiargs);
+ return 1;
+ }
+
+ /* now we can start with the SQL Statement - best to start with this first,
+ as then the error-handling is easier, as we do not have to handle libdbi shutdown as well */
+
+ /* parse the table(s) */
+ table_help.table_start=sqlargs;
+ nextptr=_find_next_separator(table_help.table_start,separator);
+ if (! nextptr) {
+ /* error in argument */
+ rrd_set_error( "formatstring wrong - %s",tmpptr);
+ return -1;
+ }
+ /* hex-unescape the value */
+ if(_inline_unescape(table_help.table_start)) { return -1; }
+
+ /* parse the unix timestamp column */
+ table_help.timestamp=nextptr;
+ nextptr=_find_next_separator(nextptr,separator);
+ if (! nextptr) {
+ /* error in argument */
+ rrd_set_error( "formatstring wrong - %s",tmpptr);
+ return -1;
+ }
+ /* hex-unescape the value */
+ if(_inline_unescape(table_help.timestamp)) { return -1; }
+
+ /* parse the value column */
+ table_help.value=nextptr;
+ nextptr=_find_next_separator(nextptr,separator);
+ if (! nextptr) {
+ /* error in argument */
+ rrd_set_error( "formatstring wrong - %s",tmpptr);
+ return -1;
+ }
+ /* hex-unescape the value */
+ if(_inline_unescape(table_help.value)) { return -1; }
+
+ /* now prepare WHERE clause as empty string*/
+ where[0]=0;
+
+ /* and the where clause */
+ sqlargs=nextptr;
+ while(sqlargs) {
+ /* find next separator */
+ nextptr=_find_next_separator(sqlargs,separator);
+ /* now handle fields */
+ if (strcmp(sqlargs,"derive")==0) { /* the derive option with the default allowed max delta */
+ derive=600;
+ } else if (strcmp(sqlargs,"prediction")==0) {
+ rrd_set_error("argument prediction is no longer supported in a VDEF - use new generic CDEF-functions instead");
+ return -1;
+ } else if (strcmp(sqlargs,"sigma")==0) {
+ rrd_set_error("argument sigma is no longer supported in a VDEF - use new generic CDEF-functions instead");
+ return -1;
+ } else if (*sqlargs==0) { /* ignore empty */
+ } else { /* else add to where string */
+ if (where[0]) {strcat(where," AND ");}
+ strcat(where,sqlargs);
+ }
+ /* and continue loop with next pointer */
+ sqlargs=nextptr;
+ }
+ /* and unescape */
+ if(_inline_unescape(where)) { return -1; }
+
+ /* now parse LIBDBI options - this start initializing libdbi and beyond this point we need to reset the db as well in case of errors*/
+ while (libdbiargs) {
+ /* find separator */
+ nextptr=_find_next_separator(libdbiargs,separator);
+ /* now find =, separating key from value*/
+ tmpptr=_find_next_separator(libdbiargs,'=');
+ if (! tmpptr) {
+ rrd_set_error( "formatstring wrong for db arguments as we did not find \"=\" in \"%s\"",libdbiargs);
+ _sql_close(&table_help);
+ return 1;
+ }
+ /* hex-unescape the value */
+ if(_inline_unescape(tmpptr)) { return -1; }
+ /* now handle the key/value pair */
+ if (strcmp(libdbiargs,"rrdminstepsize")==0) { /* allow override for minstepsize */
+ i=atoi(tmpptr);if (i>0) { minstepsize=i; }
+ } else if (strcmp(libdbiargs,"rrdfillmissing")==0) { /* allow override for minstepsize */
+ i=atoi(tmpptr);if (i>0) { fillmissing=i; }
+ } else if (strcmp(libdbiargs,"rrdderivemaxstep")==0) { /* allow override for minstepsize */
+ i=atoi(tmpptr);if (i>0) { derive=i; }
+ } else { /* store in libdbi, as these are parameters */
+ if (_sql_setparam(&table_help,libdbiargs,tmpptr)) {
+ _sql_close(&table_help);
+ return -1;
+ }
+ }
+ /* and continue loop with next pointer */
+ libdbiargs=nextptr;
+ }
+
+ /* and modify step if given */
+ if (*step<minstepsize) {*step=minstepsize;}
+ *start-=(*start)%(*step);
+ *end-=(*end)%(*step);
+
+ /* and append the SQL WHERE Clause for the timeframe calculated above (adding AND if required) */
+ if (where[0]) {strcat(where," AND ");}
+ i=strlen(where);
+ snprintf(where+i,sizeof(where)-1-i,"%li < %s AND %s < %li",*start,table_help.timestamp,table_help.timestamp,*end);
+
+ /* and now calculate the number of rows in the resultset... */
+ rows=((*end)-(*start))/(*step)+2;
+
+ /* define the result set variables/columns returned */
+ *ds_cnt=5;
+ *ds_namv=(char**)malloc((*ds_cnt)*sizeof(char*));
+ for (i=0;i<(int)(*ds_cnt);i++) {
+ tmpptr=(char*)malloc(sizeof(char) * DS_NAM_SIZE);
+ (*ds_namv)[i]=tmpptr;
+ /* now copy what is required */
+ switch (i) {
+ case 0: strncpy(tmpptr,"min",DS_NAM_SIZE-1); break;
+ case 1: strncpy(tmpptr,"avg",DS_NAM_SIZE-1); break;
+ case 2: strncpy(tmpptr,"max",DS_NAM_SIZE-1); break;
+ case 3: strncpy(tmpptr,"count",DS_NAM_SIZE-1); break;
+ case 4: strncpy(tmpptr,"sigma",DS_NAM_SIZE-1); break;
+ }
+ }
+
+ /* allocate memory for resultset (with the following columns: min,avg,max,count,sigma) */
+ i=rows * sizeof(rrd_value_t)*(*ds_cnt);
+ if (((*data) = malloc(i))==NULL){
+ /* and return error */
+ rrd_set_error("malloc failed for %i bytes",i);
+ return(-1);
+ }
+ /* and fill with NAN */
+ for(i=0;i<rows;i++) {
+ (*data)[i*(*ds_cnt)+0]=DNAN; /* MIN */
+ (*data)[i*(*ds_cnt)+1]=DNAN; /* AVG */
+ (*data)[i*(*ds_cnt)+2]=DNAN; /* MAX */
+ (*data)[i*(*ds_cnt)+3]=0; /* COUNT */
+ (*data)[i*(*ds_cnt)+4]=DNAN; /* SIGMA */
+ }
+ /* and assign undefined values for last - in case of derived calculation */
+ l_value=DNAN;l_timestamp=0;
+ /* here goes the real work processing all data */
+ while((r_status=_sql_fetchrow(&table_help,&r_timestamp,&r_value))>0) {
+ /* processing of value */
+ /* calculate index for the timestamp */
+ idx=(r_timestamp-(*start))/(*step);
+ /* some out of bounds checks on idx */
+ if (idx<0) { idx=0;}
+ if (idx>rows) { idx=rows;}
+ /* and calculate derivative if necessary */
+ if (derive) {
+ /* calc deltas */
+ d_timestamp=r_timestamp-l_timestamp;
+ d_value=r_value-l_value;
+ /* assign current as last values */
+ l_timestamp=r_timestamp;
+ l_value=r_value;
+ /* assign DNAN by default for value */
+ r_value=DNAN;
+ /* check for timestamp delta to be within an acceptable range */
+ if ((d_timestamp>0)&&(d_timestamp<2*derive)) {
+ /* only handle positive delta - avoid wrap-arrounds/counter resets showing up as spikes */
+ if (d_value>0) {
+ /* and normalize to per second */
+ r_value=d_value/d_timestamp;
+ }
+ }
+ }
+ /* only add value if we have a value that is not NAN */
+ if (! isnan(r_value)) {
+ if ((*data)[idx*(*ds_cnt)+3]==0) { /* count is 0 so assign to overwrite DNAN */
+ (*data)[idx*(*ds_cnt)+0]=r_value; /* MIN */
+ (*data)[idx*(*ds_cnt)+1]=r_value; /* AVG */
+ (*data)[idx*(*ds_cnt)+2]=r_value; /* MAX */
+ (*data)[idx*(*ds_cnt)+3]=1; /* COUNT */
+ (*data)[idx*(*ds_cnt)+4]=r_value; /* SIGMA */
+ } else {
+ /* MIN */
+ if ((*data)[idx*(*ds_cnt)+0]>r_value) { (*data)[idx*(*ds_cnt)+0]=r_value; }
+ /* AVG - at this moment still sum - corrected in post processing */
+ (*data)[idx*(*ds_cnt)+1]+=r_value;
+ /* MAX */
+ if ((*data)[idx*(*ds_cnt)+2]<r_value) { (*data)[idx*(*ds_cnt)+2]=r_value; }
+ /* COUNT */
+ (*data)[idx*(*ds_cnt)+3]++;
+ /* SIGMA - at this moment still sum of squares - corrected in post processing */
+ (*data)[idx*(*ds_cnt)+4]+=r_value*r_value;
+ }
+ }
+ }
+ /* and check for negativ status, pass back immediately */
+ if (r_status==-1) { return -1; }
+
+ /* post processing */
+ for(idx=0;idx<rows;idx++) {
+ long count=(*data)[idx*(*ds_cnt)+3];
+ if (count>0) {
+ /* calc deviation first */
+ if (count>2) {
+ r_value=count*(*data)[idx*(*ds_cnt)+4]-(*data)[idx*(*ds_cnt)+1]*(*data)[idx*(*ds_cnt)+1];
+ if (r_value<0) {
+ r_value=DNAN;
+ } else {
+ r_value=sqrt(r_value/(count*(count-1)));
+ }
+ }
+ (*data)[idx*(*ds_cnt)+4]=r_value;
+ /* now the average */
+ (*data)[idx*(*ds_cnt)+1]/=i;
+ }
+ }
+
+ /* and return OK */
+ return 0;
+}