nominatim/index.c

   1 /*
   2 */
   3
   4 #include <stdio.h>
   5 #include <unistd.h>
   6 #include <stdlib.h>
   7 #include <string.h>
   8 #include <assert.h>
   9 #include <pthread.h>
  10 #include <time.h>
  11 #include <stdint.h>
  12
  13 #include <libpq-fe.h>
  14
  15 #include "nominatim.h"
  16 #include "index.h"
  17 #include "export.h"
  18 #include "postgresql.h"
  19
  20 extern int verbose;
  21
  22 void nominatim_index(int rank_min, int rank_max, int num_threads, const char *conninfo, const char *structuredoutputfile)
  23 {
  24         struct index_thread_data * thread_data;
  25         pthread_mutex_t count_mutex = PTHREAD_MUTEX_INITIALIZER;
  26         int tuples, count, sleepcount;
  27
  28         time_t rankStartTime;
  29         int rankTotalTuples;
  30         int rankCountTuples;
  31         float rankPerSecond;
  32
  33         PGconn *conn;
  34         PGresult * res;
  35         PGresult * resSectors;
  36         PGresult * resPlaces;
  37
  38         int rank;
  39         int i;
  40         int iSector;
  41
  42     const char *paramValues[2];
  43     int         paramLengths[2];
  44     int         paramFormats[2];
  45     uint32_t    paramRank;
  46     uint32_t    paramSector;
  47     uint32_t    sector;
  48
  49     xmlTextWriterPtr writer;
  50         pthread_mutex_t writer_mutex = PTHREAD_MUTEX_INITIALIZER;
  51
  52     Oid pg_prepare_params[2];
  53
  54     conn = PQconnectdb(conninfo);
  55     if (PQstatus(conn) != CONNECTION_OK) {
  56         fprintf(stderr, "Connection to database failed: %s\n", PQerrorMessage(conn));
  57         exit(EXIT_FAILURE);
  58     }
  59
  60     pg_prepare_params[0] = PG_OID_INT4;
  61     res = PQprepare(conn, "index_sectors",
  62         "select geometry_sector,count(*) from placex where rank_search = $1 and indexed_status > 0 group by geometry_sector order by geometry_sector",
  63         1, pg_prepare_params);
  64     if (PQresultStatus(res) != PGRES_COMMAND_OK)
  65     {
  66         fprintf(stderr, "Failed preparing index_sectors: %s\n", PQerrorMessage(conn));
  67         exit(EXIT_FAILURE);
  68     }
  69     PQclear(res);
  70
  71     pg_prepare_params[0] = PG_OID_INT4;
  72     pg_prepare_params[1] = PG_OID_INT4;
  73     res = PQprepare(conn, "index_sector_places",
  74         "select place_id from placex where rank_search = $1 and geometry_sector = $2 and indexed_status > 0",
  75         2, pg_prepare_params);
  76     if (PQresultStatus(res) != PGRES_COMMAND_OK)
  77     {
  78         fprintf(stderr, "Failed preparing index_sector_places: %s\n", PQerrorMessage(conn));
  79         exit(EXIT_FAILURE);
  80     }
  81     PQclear(res);
  82
  83     // Build the data for each thread
  84     thread_data = (struct index_thread_data *)malloc(sizeof(struct index_thread_data)*num_threads);
  85         for (i = 0; i < num_threads; i++)
  86         {
  87                 thread_data[i].conn = PQconnectdb(conninfo);
  88             if (PQstatus(thread_data[i].conn) != CONNECTION_OK) {
  89                 fprintf(stderr, "Connection to database failed: %s\n", PQerrorMessage(thread_data[i].conn));
  90                 exit(EXIT_FAILURE);
  91             }
  92
  93             pg_prepare_params[0] = PG_OID_INT4;
  94             res = PQprepare(thread_data[i].conn, "index_placex",
  95                 "update placex set indexed_status = 0 where place_id = $1",
  96                 1, pg_prepare_params);
  97             if (PQresultStatus(res) != PGRES_COMMAND_OK)
  98             {
  99                 fprintf(stderr, "Failed preparing index_placex: %s\n", PQerrorMessage(conn));
 100                 exit(EXIT_FAILURE);
 101             }
 102             PQclear(res);
 103
 104             nominatim_exportCreatePreparedQueries(thread_data[i].conn);
 105         }
 106
 107         // Create the output file
 108         writer = NULL;
 109         if (structuredoutputfile)
 110         {
 111                 writer = nominatim_exportXMLStart(structuredoutputfile);
 112         }
 113
 114     fprintf(stderr, "Starting indexing rank (%i > %i ) using %i treads\n", rank_min, rank_max, num_threads);
 115
 116     for (rank = rank_min; rank <= rank_max; rank++)
 117     {
 118         printf("Starting rank %d\n", rank);
 119         rankStartTime = time(0);
 120         rankCountTuples = 0;
 121         rankPerSecond = 0;
 122
 123         paramRank = PGint32(rank);
 124         paramValues[0] = (char *)&paramRank;
 125         paramLengths[0] = sizeof(paramRank);
 126         paramFormats[0] = 1;
 127         resSectors = PQexecPrepared(conn, "index_sectors", 1, paramValues, paramLengths, paramFormats, 1);
 128         if (PQresultStatus(resSectors) != PGRES_TUPLES_OK)
 129         {
 130             fprintf(stderr, "index_sectors: SELECT failed: %s", PQerrorMessage(conn));
 131             PQclear(resSectors);
 132             exit(EXIT_FAILURE);
 133         }
 134                 if (PQftype(resSectors, 0) != PG_OID_INT4)
 135                 {
 136             fprintf(stderr, "Sector value has unexpected type\n");
 137             PQclear(resSectors);
 138             exit(EXIT_FAILURE);
 139                 }
 140                 if (PQftype(resSectors, 1) != PG_OID_INT8)
 141                 {
 142             fprintf(stderr, "Sector value has unexpected type\n");
 143             PQclear(resSectors);
 144             exit(EXIT_FAILURE);
 145                 }
 146
 147                 rankTotalTuples = 0;
 148         for (iSector = 0; iSector < PQntuples(resSectors); iSector++)
 149         {
 150                 rankTotalTuples += PGint64(*((uint64_t *)PQgetvalue(resSectors, iSector, 1)));
 151         }
 152
 153         for (iSector = 0; iSector < PQntuples(resSectors); iSector++)
 154         {
 155                         sector = PGint32(*((uint32_t *)PQgetvalue(resSectors, iSector, 0)));
 156                 //printf("\n Starting sector %d size %ld\n", sector, PGint64(*((uint64_t *)PQgetvalue(resSectors, iSector, 1))));
 157
 158                         // Get all the place_id's for this sector
 159                 paramRank = PGint32(rank);
 160                 paramValues[0] = (char *)&paramRank;
 161                 paramLengths[0] = sizeof(paramRank);
 162                 paramFormats[0] = 1;
 163                 paramSector = PGint32(sector);
 164                 paramValues[1] = (char *)&paramSector;
 165                 paramLengths[1] = sizeof(paramSector);
 166                 paramFormats[1] = 1;
 167                 resPlaces = PQexecPrepared(conn, "index_sector_places", 2, paramValues, paramLengths, paramFormats, 1);
 168                 if (PQresultStatus(resPlaces) != PGRES_TUPLES_OK)
 169                 {
 170                     fprintf(stderr, "index_sector_places: SELECT failed: %s", PQerrorMessage(conn));
 171                     PQclear(resPlaces);
 172                     exit(EXIT_FAILURE);
 173                 }
 174                         if (PQftype(resPlaces, 0) != PG_OID_INT4)
 175                         {
 176                     fprintf(stderr, "Place_id value has unexpected type\n");
 177                     PQclear(resPlaces);
 178                     exit(EXIT_FAILURE);
 179                         }
 180
 181                         count = 0;
 182                         rankPerSecond = 0;
 183                         tuples = PQntuples(resPlaces);
 184
 185                         if (tuples > 0)
 186                 {
 187                                 // Spawn threads
 188                                 for (i = 0; i < num_threads; i++)
 189                                 {
 190                                         thread_data[i].res = resPlaces;
 191                                         thread_data[i].tuples = tuples;
 192                                         thread_data[i].count = &count;
 193                                         thread_data[i].count_mutex = &count_mutex;
 194                                         thread_data[i].writer = writer;
 195                                         thread_data[i].writer_mutex = &writer_mutex;
 196                                         pthread_create(&thread_data[i].thread, NULL, &nominatim_indexThread, (void *)&thread_data[i]);
 197                                 }
 198
 199                                 // Monitor threads to give user feedback
 200                                 sleepcount = 0;
 201                                 while(count < tuples)
 202                                 {
 203                                         usleep(1000);
 204
 205                                         // Aim for one update per second
 206                                         if (sleepcount++ > 500)
 207                                         {
 208                                                 rankPerSecond = ((float)rankCountTuples + (float)count) / MAX(difftime(time(0), rankStartTime),1);
 209                                                 printf("  Done %i in %i @ %f per second - Rank %i ETA (seconds): %f\n", (rankCountTuples + count), (int)(difftime(time(0), rankStartTime)), rankPerSecond, rank, ((float)(rankTotalTuples - (rankCountTuples + count)))/rankPerSecond);
 210                                                 sleepcount = 0;
 211                                         }
 212                                 }
 213
 214                                 // Wait for everything to finish
 215                                 for (i = 0; i < num_threads; i++)
 216                                 {
 217                                         pthread_join(thread_data[i].thread, NULL);
 218                                 }
 219
 220                                 rankCountTuples += tuples;
 221                 }
 222
 223                         // Finished sector
 224                         rankPerSecond = (float)rankCountTuples / MAX(difftime(time(0), rankStartTime),1);
 225                         printf("  Done %i in %i @ %f per second - ETA (seconds): %f\n", rankCountTuples, (int)(difftime(time(0), rankStartTime)), rankPerSecond, ((float)(rankTotalTuples - rankCountTuples))/rankPerSecond);
 226
 227             PQclear(resPlaces);
 228
 229         }
 230         // Finished rank
 231                 printf("\r  Done %i in %i @ %f per second - FINISHED                      \n\n", rankCountTuples, (int)(difftime(time(0), rankStartTime)), rankPerSecond);
 232
 233         PQclear(resSectors);
 234     }
 235
 236     if (writer)
 237     {
 238         nominatim_exportXMLEnd(writer);
 239     }
 240 }
 241
 242 void *nominatim_indexThread(void * thread_data_in)
 243 {
 244         struct index_thread_data * thread_data = (struct index_thread_data * )thread_data_in;
 245
 246         PGresult * res;
 247
 248     const char *paramValues[1];
 249     int         paramLengths[1];
 250     int         paramFormats[1];
 251     uint32_t    paramPlaceID;
 252     uint32_t place_id;
 253
 254         while(1)
 255         {
 256                 pthread_mutex_lock( thread_data->count_mutex );
 257                 if (*(thread_data->count) >= thread_data->tuples)
 258                 {
 259                         pthread_mutex_unlock( thread_data->count_mutex );
 260                         break;
 261                 }
 262
 263                 place_id = PGint32(*((uint32_t *)PQgetvalue(thread_data->res, *thread_data->count, 0)));
 264                 (*thread_data->count)++;
 265
 266                 pthread_mutex_unlock( thread_data->count_mutex );
 267
 268                 //printf("  Processing place_id %ld\n", place_id);
 269                 paramPlaceID = PGint32(place_id);
 270         paramValues[0] = (char *)&paramPlaceID;
 271         paramLengths[0] = sizeof(paramPlaceID);
 272         paramFormats[0] = 1;
 273         res = PQexecPrepared(thread_data->conn, "index_placex", 1, paramValues, paramLengths, paramFormats, 1);
 274         if (PQresultStatus(res) != PGRES_COMMAND_OK)
 275         {
 276             fprintf(stderr, "index_placex: UPDATE failed: %s", PQerrorMessage(thread_data->conn));
 277             PQclear(res);
 278             exit(EXIT_FAILURE);
 279         }
 280         PQclear(res);
 281
 282         if (thread_data->writer)
 283         {
 284                 nominatim_exportPlace(place_id, thread_data->conn, thread_data->writer, thread_data->writer_mutex);
 285         }
 286         }
 287
 288         return NULL;
 289 }