index on geometry of interpolation lines, and more improvements.

This commit is contained in:
Markus Gail 2016-04-27 17:42:59 +02:00
parent a10f537131
commit db719d489f
8 changed files with 262 additions and 386 deletions

View File

@ -466,10 +466,10 @@
$sSQL .= " group by place_id, housenumber_for_place"; //is this group by really needed?, place_id + housenumber (in combination) are unique
if (!$this->bDeDupe) $sSQL .= ", place_id ";
}
// osmline, osm_type is 'I' for Interpolation Line
// osmline
// interpolation line search only if a housenumber was searched and if it was found (i.e. aPlaceIDs[placeID] = housenumber != -1) (realized through a join)
$sSQL .= " union ";
$sSQL .= "select 'I' as osm_type, place_id as osm_id, 'place' as class, 'house' as type, null as admin_level, 30 as rank_search, 30 as rank_address, min(place_id) as place_id, min(parent_place_id) as parent_place_id, calculated_country_code as country_code, ";
$sSQL .= "select 'W' as osm_type, place_id as osm_id, 'place' as class, 'house' as type, null as admin_level, 30 as rank_search, 30 as rank_address, min(place_id) as place_id, min(parent_place_id) as parent_place_id, calculated_country_code as country_code, ";
$sSQL .= "get_address_by_language(place_id, housenumber_for_place, $sLanguagePrefArraySQL) as langaddress, ";
$sSQL .= "null as placename, ";
$sSQL .= "null as ref, ";
@ -477,7 +477,8 @@
if ($this->bIncludeNameDetails) $sSQL .= "null as names, ";
$sSQL .= " avg(st_x(centroid)) as lon, avg(st_y(centroid)) as lat,";
$sSQL .= $sImportanceSQL."-0.1 as importance, "; // slightly smaller than the importance for normal houses with rank 30, which is 0
$sSQL .= " (select max(p.importance*(p.rank_address+2)) from place_addressline s, placex p where s.place_id = min(blub.parent_place_id) and p.place_id = s.address_place_id and s.isaddress and p.importance is not null) as addressimportance, ";
$sSQL .= " (select max(p.importance*(p.rank_address+2)) from place_addressline s, placex p";
$sSQL .= " where s.place_id = min(blub.parent_place_id) and p.place_id = s.address_place_id and s.isaddress and p.importance is not null) as addressimportance,";
$sSQL .= " null as extra_place ";
$sSQL .= " from (select place_id, calculated_country_code ";
//interpolate the housenumbers here
@ -1454,7 +1455,7 @@
$aRoadPlaceIDs = $aPlaceIDs;
$sPlaceIDs = join(',',$aPlaceIDs);
// Now they are indexed look for a house attached to a street we found
// Now they are indexed, look for a house attached to a street we found
$sHouseNumberRegex = '\\\\m'.$aSearch['sHouseNumber'].'\\\\M';
$sSQL = "select place_id from placex where parent_place_id in (".$sPlaceIDs.") and transliteration(housenumber) ~* E'".$sHouseNumberRegex."'";
if (sizeof($this->aExcludePlaceIDs))

View File

@ -142,9 +142,9 @@
}
else if ($this->sType == 'interpolation')
{
$sSQL = "select place_id, partition, 'I' as osm_type, osm_id, 'place' as class, 'house' as type, null admin_level, housenumber, null as street, null as isin, postcode,";
$sSQL = "select place_id, partition, 'W' as osm_type, osm_id, 'place' as class, 'house' as type, null admin_level, housenumber, null as street, null as isin, postcode,";
$sSQL .= " calculated_country_code as country_code, parent_place_id, null as linked_place_id, 30 as rank_address, 30 as rank_search,";
$sSQL .= " coalesce(null,0.75-(30::float/40)) as importance, null as indexed_status, null as indexed_date, null as wikipedia, calculated_country_code, ";
$sSQL .= " (0.75-(30::float/40)) as importance, null as indexed_status, null as indexed_date, null as wikipedia, calculated_country_code, ";
$sSQL .= " get_address_by_language(place_id, housenumber, $sLanguagePrefArraySQL) as langaddress,";
$sSQL .= " null as placename,";
$sSQL .= " null as ref,";
@ -187,7 +187,7 @@
if ($this->bAddressDetails)
{
if(CONST_Use_US_Tiger_Data && $this->sType == 'tiger') // to get addressdetails for tiger data, the housenumber is needed
if(CONST_Use_US_Tiger_Data && $this->sType == 'tiger' || $this->sType == 'interpolation') // to get addressdetails for tiger data, the housenumber is needed
$aAddress = $this->getAddressNames($aPlace['housenumber']);
else
$aAddress = $this->getAddressNames();

View File

@ -158,43 +158,15 @@
$bIsInUnitedStates = ($aPlace['calculated_country_code'] == 'us');
}
// if a street or house was found, look in interpolation lines table
if ($iMaxRank_orig >= 28 && $iPlaceID && ($aPlace['rank_search'] == 26 || $aPlace['rank_search'] == 27 || $aPlace['rank_search'] == 30))
if ($iMaxRank_orig >= 28 && $aPlace && $aPlace['rank_search'] >= 26)
{
$fSearchDiam = 0.001;
if ($aPlace['rank_search'] == 30)
{
// if a house was found, the closest road needs to be searched, to use its place id as parent_place_id for the interpolation line search
// because a road can be closer to the point than the house from above
$iRoadID = null;
while(!$iRoadID && $fSearchDiam < $fMaxAreaDistance)
{
$fSearchDiam = $fSearchDiam * 2;
$sSQL = 'select place_id ';
$sSQL .= ' FROM placex';
$sSQL .= ' WHERE ST_DWithin('.$sPointSQL.', geometry, '.$fSearchDiam.')';
$sSQL .= ' and (rank_search = 26 or rank_search = 27)';
$sSQL .= ' and class not in (\'waterway\',\'railway\',\'tunnel\',\'bridge\',\'man_made\')';
$sSQL .= ' and indexed_status = 0 ';
$sSQL .= ' ORDER BY ST_distance('.$sPointSQL.', geometry) ASC limit 1';
$aPlaceRoad = $this->oDB->getRow($sSQL);
if (PEAR::IsError($aPlace))
{
failInternalError("Could not determine closest place.", $sSQL, $aPlace);
}
$iRoadID = $aPlaceRoad['place_id'];
$iTempPlaceID = $iRoadID;
}
}
else
{
// if a street was found, we can take its place_id as parent_place_id
$iTempPlaceID = $iPlaceID;
}
// if a house was found, search the interpolation line that is at least as close as the house
$sSQL = 'SELECT place_id, parent_place_id, 30 as rank_search, ST_line_locate_point(linegeo,'.$sPointSQL.') as fraction';
//if (CONST_Debug) { $sSQL .= ', housenumber, ST_distance('.$sPointSQL.', centroid) as distance, st_y(centroid) as lat, st_x(centroid) as lon'; }
$sSQL .= ' FROM location_property_osmline WHERE parent_place_id = '.$iTempPlaceID;
$sSQL .= ' AND ST_DWithin('.$sPointSQL.', linegeo, '.$fSearchDiam.') AND indexed_status = 0';
$sSQL .= ' FROM location_property_osmline';
$sSQL .= ' WHERE ST_DWithin('.$sPointSQL.', linegeo, '.$fSearchDiam.')';
$sSQL .= ' and indexed_status = 0 ';
$sSQL .= ' ORDER BY ST_distance('.$sPointSQL.', linegeo) ASC limit 1';
if (CONST_Debug)
{
$sSQL = preg_replace('/limit 1/', 'limit 100', $sSQL);
@ -211,7 +183,6 @@
{
failInternalError("Could not determine closest housenumber on an osm interpolation line.", $sSQL, $aPlaceLine);
}
$iInterpolationLinePlaceID = $aPlaceLine['place_id'];
if ($aPlaceLine)
{
if (CONST_Debug) var_dump('found housenumber in interpolation lines table', $aPlaceLine);
@ -229,7 +200,7 @@
$fDistancePlacex = $aDistancePlacex['distance'];
// distance between point and interpolated house (fraction on interpolation line)
$sSQL = 'SELECT ST_distance('.$sPointSQL.', ST_LineInterpolatePoint(linegeo, '.$aPlaceLine['fraction'].')) as distance';
$sSQL .= ' FROM location_property_osmline WHERE place_id = '.$iInterpolationLinePlaceID;
$sSQL .= ' FROM location_property_osmline WHERE place_id = '.$aPlaceLine['place_id'];
$aDistanceInterpolation = $this->oDB->getRow($sSQL);
if (PEAR::IsError($aDistanceInterpolation))
{
@ -241,7 +212,7 @@
// interpolation is closer to point than placex house
$bPlaceIsLine = true;
$aPlace = $aPlaceLine;
$iPlaceID = $iInterpolationLinePlaceID;
$iPlaceID = $aPlaceLine['place_id'];
$iParentPlaceID = $aPlaceLine['parent_place_id']; // the street
$fFraction = $aPlaceLine['fraction'];
}

View File

@ -21,34 +21,243 @@
extern int verbose;
void nominatim_index(int rank_min, int rank_max, int num_threads, const char *conninfo, const char *structuredoutputfile)
void run_indexing(int rank, int interpolation, PGconn *conn, int num_threads,
struct index_thread_data * thread_data, const char *structuredoutputfile)
{
struct index_thread_data * thread_data;
pthread_mutex_t count_mutex = PTHREAD_MUTEX_INITIALIZER;
int tuples, count, sleepcount;
pthread_mutex_t count_mutex = PTHREAD_MUTEX_INITIALIZER;
time_t rankStartTime;
int rankTotalTuples;
int rankCountTuples;
float rankPerSecond;
PGconn *conn;
PGresult * res;
PGresult * resSectors;
PGresult * resPlaces;
PGresult * resNULL;
int rank;
int i;
int iSector;
int iResult;
const char *paramValues[2];
int paramLengths[2];
int paramFormats[2];
uint32_t paramRank;
uint32_t paramSector;
uint32_t sector;
xmlTextWriterPtr writer;
pthread_mutex_t writer_mutex = PTHREAD_MUTEX_INITIALIZER;
// Create the output file
writer = NULL;
if (structuredoutputfile)
{
writer = nominatim_exportXMLStart(structuredoutputfile);
}
if (interpolation)
{
fprintf(stderr, "Starting interpolation lines (location_property_osmline)\n");
}
else
{
fprintf(stderr, "Starting rank %d\n", rank);
}
rankCountTuples = 0;
rankPerSecond = 0;
paramRank = PGint32(rank);
paramValues[0] = (char *)&paramRank;
paramLengths[0] = sizeof(paramRank);
paramFormats[0] = 1;
if (interpolation)
{
resSectors = PQexecPrepared(conn, "index_sectors_osmline", 0, NULL, 0, NULL, 1);
}
else
{
resSectors = PQexecPrepared(conn, "index_sectors", 1, paramValues, paramLengths, paramFormats, 1);
}
if (PQresultStatus(resSectors) != PGRES_TUPLES_OK)
{
fprintf(stderr, "index_sectors: SELECT failed: %s", PQerrorMessage(conn));
PQclear(resSectors);
exit(EXIT_FAILURE);
}
if (PQftype(resSectors, 0) != PG_OID_INT4)
{
fprintf(stderr, "Sector value has unexpected type\n");
PQclear(resSectors);
exit(EXIT_FAILURE);
}
if (PQftype(resSectors, 1) != PG_OID_INT8)
{
fprintf(stderr, "Sector value has unexpected type\n");
PQclear(resSectors);
exit(EXIT_FAILURE);
}
rankTotalTuples = 0;
for (iSector = 0; iSector < PQntuples(resSectors); iSector++)
{
rankTotalTuples += PGint64(*((uint64_t *)PQgetvalue(resSectors, iSector, 1)));
}
rankStartTime = time(0);
for (iSector = 0; iSector <= PQntuples(resSectors); iSector++)
{
if (iSector > 0)
{
resPlaces = PQgetResult(conn);
if (PQresultStatus(resPlaces) != PGRES_TUPLES_OK)
{
fprintf(stderr, "index_sector_places: SELECT failed: %s", PQerrorMessage(conn));
PQclear(resPlaces);
exit(EXIT_FAILURE);
}
if (PQftype(resPlaces, 0) != PG_OID_INT8)
{
fprintf(stderr, "Place_id value has unexpected type\n");
PQclear(resPlaces);
exit(EXIT_FAILURE);
}
resNULL = PQgetResult(conn);
if (resNULL != NULL)
{
fprintf(stderr, "Unexpected non-null response\n");
exit(EXIT_FAILURE);
}
}
if (iSector < PQntuples(resSectors))
{
sector = PGint32(*((uint32_t *)PQgetvalue(resSectors, iSector, 0)));
// fprintf(stderr, "\n Starting sector %d size %ld\n", sector, PGint64(*((uint64_t *)PQgetvalue(resSectors, iSector, 1))));
// Get all the place_id's for this sector
paramRank = PGint32(rank);
paramValues[0] = (char *)&paramRank;
paramLengths[0] = sizeof(paramRank);
paramFormats[0] = 1;
paramSector = PGint32(sector);
paramValues[1] = (char *)&paramSector;
paramLengths[1] = sizeof(paramSector);
paramFormats[1] = 1;
if (rankTotalTuples-rankCountTuples < num_threads*1000)
{
// no sectors
if (interpolation)
{
iResult = PQsendQueryPrepared(conn, "index_nosector_places_osmline", 0, NULL, 0, NULL, 1);
}
else
{
iResult = PQsendQueryPrepared(conn, "index_nosector_places", 1, paramValues, paramLengths, paramFormats, 1);
}
}
else
{
if (interpolation)
{
iResult = PQsendQueryPrepared(conn, "index_sector_places", 2, paramValues, paramLengths, paramFormats, 1);
}
else
{
iResult = PQsendQueryPrepared(conn, "index_sector_places_osmline", 1, paramValues, paramLengths, paramFormats, 1);
}
}
if (!iResult)
{
fprintf(stderr, "index_sector_places: SELECT failed: %s", PQerrorMessage(conn));
PQclear(resPlaces);
exit(EXIT_FAILURE);
}
}
if (iSector > 0)
{
count = 0;
rankPerSecond = 0;
tuples = PQntuples(resPlaces);
if (tuples > 0)
{
// Spawn threads
for (i = 0; i < num_threads; i++)
{
thread_data[i].res = resPlaces;
thread_data[i].tuples = tuples;
thread_data[i].count = &count;
thread_data[i].count_mutex = &count_mutex;
thread_data[i].writer = writer;
thread_data[i].writer_mutex = &writer_mutex;
if (interpolation)
{
thread_data[i].table = 0; // use interpolations table
}
else
{
thread_data[i].table = 1; // use placex table
}
pthread_create(&thread_data[i].thread, NULL, &nominatim_indexThread, (void *)&thread_data[i]);
}
// Monitor threads to give user feedback
sleepcount = 0;
while (count < tuples)
{
usleep(1000);
// Aim for one update per second
if (sleepcount++ > 500)
{
rankPerSecond = ((float)rankCountTuples + (float)count) / MAX(difftime(time(0), rankStartTime),1);
fprintf(stderr, " 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);
sleepcount = 0;
}
}
// Wait for everything to finish
for (i = 0; i < num_threads; i++)
{
pthread_join(thread_data[i].thread, NULL);
}
rankCountTuples += tuples;
}
// Finished sector
rankPerSecond = (float)rankCountTuples / MAX(difftime(time(0), rankStartTime),1);
fprintf(stderr, " Done %i in %i @ %f per second - ETA (seconds): %f\n", rankCountTuples, (int)(difftime(time(0), rankStartTime)), rankPerSecond, ((float)(rankTotalTuples - rankCountTuples))/rankPerSecond);
PQclear(resPlaces);
}
if (rankTotalTuples-rankCountTuples < num_threads*20 && iSector < PQntuples(resSectors))
{
iSector = PQntuples(resSectors) - 1;
}
}
// Finished rank
fprintf(stderr, "\r Done %i in %i @ %f per second - FINISHED\n\n", rankCountTuples, (int)(difftime(time(0), rankStartTime)), rankPerSecond);
PQclear(resSectors);
}
void nominatim_index(int rank_min, int rank_max, int num_threads, const char *conninfo, const char *structuredoutputfile)
{
struct index_thread_data * thread_data;
PGconn *conn;
PGresult * res;
int rank;
int i;
xmlTextWriterPtr writer;
pthread_mutex_t writer_mutex = PTHREAD_MUTEX_INITIALIZER;
@ -182,324 +391,22 @@ void nominatim_index(int rank_min, int rank_max, int num_threads, const char *co
nominatim_exportCreatePreparedQueries(thread_data[i].conn);
}
// Create the output file
writer = NULL;
if (structuredoutputfile)
{
writer = nominatim_exportXMLStart(structuredoutputfile);
}
fprintf(stderr, "Starting indexing rank (%i to %i) using %i threads\n", rank_min, rank_max, num_threads);
// first for the placex table
for (rank = rank_min; rank <= rank_max; rank++)
{
// OSMLINE: do reindexing (=> reparenting) for interpolation lines at rank 30, but before all other objects of rank 30
// reason: houses (rank 30) depend on the updated interpolation line, when reparenting (see placex_update in functions.sql)
if (rank == 30)
{
fprintf(stderr, "Starting indexing interpolation lines (location_property_osmline)\n");
rankCountTuples = 0;
rankTotalTuples = 0;
resSectors = PQexecPrepared(conn, "index_sectors_osmline", 0, NULL, 0, NULL, 1);
if (PQresultStatus(resSectors) != PGRES_TUPLES_OK)
{
fprintf(stderr, "index_sectors_osmline: SELECT failed: %s", PQerrorMessage(conn));
PQclear(resSectors);
exit(EXIT_FAILURE);
}
if (PQftype(resSectors, 0) != PG_OID_INT4)
{
fprintf(stderr, "Sector value has unexpected type\n");
PQclear(resSectors);
exit(EXIT_FAILURE);
}
if (PQftype(resSectors, 1) != PG_OID_INT8)
{
fprintf(stderr, "Sector value has unexpected type\n");
PQclear(resSectors);
exit(EXIT_FAILURE);
}
rankStartTime = time(0);
for (iSector = 0; iSector < PQntuples(resSectors); iSector++)
{
rankTotalTuples += PGint64(*((uint64_t *)PQgetvalue(resSectors, iSector, 1)));
}
// do it only if tuples with indexed_status > 0 were found in osmline
int nTuples = PQntuples(resSectors);
if (nTuples > 0)
{
for (iSector = 0; iSector <= nTuples; iSector++)
{
if (iSector > 0)
{
resPlaces = PQgetResult(conn);
if (PQresultStatus(resPlaces) != PGRES_TUPLES_OK)
{
fprintf(stderr, "index_sector_places: SELECT failed: %s\n", PQerrorMessage(conn));
PQclear(resPlaces);
exit(EXIT_FAILURE);
}
if (PQftype(resPlaces, 0) != PG_OID_INT8)
{
fprintf(stderr, "Place_id value has unexpected type\n");
PQclear(resPlaces);
exit(EXIT_FAILURE);
}
resNULL = PQgetResult(conn);
if (resNULL != NULL)
{
fprintf(stderr, "Unexpected non-null response\n");
exit(EXIT_FAILURE);
}
}
if (iSector < nTuples)
{
sector = PGint32(*((uint32_t *)PQgetvalue(resSectors, iSector, 0)));
// fprintf(stderr, "\n Starting sector %d size %ld\n", sector, PGint64(*((uint64_t *)PQgetvalue(resSectors, iSector, 1))));
// Get all the place_id's for this sector
paramSector = PGint32(sector);
paramValues[0] = (char *)&paramSector;
paramLengths[0] = sizeof(paramSector);
paramFormats[0] = 1;
if (rankTotalTuples-rankCountTuples < num_threads*1000)
{
// no sectors
iResult = PQsendQueryPrepared(conn, "index_nosector_places_osmline", 0, NULL, 0, NULL, 1);
}
else
{
iResult = PQsendQueryPrepared(conn, "index_sector_places_osmline", 1, paramValues, paramLengths, paramFormats, 1);
}
if (!iResult)
{
fprintf(stderr, "index_sector_places_osmline: SELECT failed: %s", PQerrorMessage(conn));
PQclear(resPlaces);
exit(EXIT_FAILURE);
}
}
if (iSector > 0)
{
count = 0;
rankPerSecond = 0;
tuples = PQntuples(resPlaces);
if (tuples > 0)
{
// Spawn threads
for (i = 0; i < num_threads; i++)
{
thread_data[i].res = resPlaces;
thread_data[i].tuples = tuples;
thread_data[i].count = &count;
thread_data[i].count_mutex = &count_mutex;
thread_data[i].writer = writer;
thread_data[i].writer_mutex = &writer_mutex;
thread_data[i].table = 0; // use osmline table
pthread_create(&thread_data[i].thread, NULL, &nominatim_indexThread, (void *)&thread_data[i]);
}
// Monitor threads to give user feedback
sleepcount = 0;
while (count < tuples)
{
usleep(1000);
// Aim for one update per second
if (sleepcount++ > 500)
{
rankPerSecond = ((float)rankCountTuples + (float)count) / MAX(difftime(time(0), rankStartTime),1);
fprintf(stderr, " Done %i in %i @ %f per second - Interpolation Lines ETA (seconds): %f\n", (rankCountTuples + count), (int)(difftime(time(0), rankStartTime)), rankPerSecond, ((float)(rankTotalTuples - (rankCountTuples + count)))/(float)rankPerSecond);
sleepcount = 0;
}
}
// Wait for everything to finish
for (i = 0; i < num_threads; i++)
{
pthread_join(thread_data[i].thread, NULL);
}
rankCountTuples += tuples;
}
// Finished sector
rankPerSecond = (float)rankCountTuples / MAX(difftime(time(0), rankStartTime),1);
fprintf(stderr, " Done %i in %i @ %f per second - ETA (seconds): %f\n", rankCountTuples, (int)(difftime(time(0), rankStartTime)), rankPerSecond, ((float)(rankTotalTuples - rankCountTuples))/rankPerSecond);
PQclear(resPlaces);
}
if (rankTotalTuples-rankCountTuples < num_threads*20 && iSector < nTuples)
{
iSector = nTuples - 1;
}
}
PQclear(resSectors);
}
// Finished rank
fprintf(stderr, "\r Done %i tuples in %i seconds- FINISHED\n", rankCountTuples,(int)(difftime(time(0), rankStartTime)));
if (writer)
{
nominatim_exportXMLEnd(writer);
}
run_indexing(rank, 1, conn, num_threads, thread_data, structuredoutputfile);
}
fprintf(stderr, "Starting rank %d\n", rank);
rankCountTuples = 0;
rankPerSecond = 0;
paramRank = PGint32(rank);
paramValues[0] = (char *)&paramRank;
paramLengths[0] = sizeof(paramRank);
paramFormats[0] = 1;
// if (rank < 16)
// resSectors = PQexecPrepared(conn, "index_nosectors", 1, paramValues, paramLengths, paramFormats, 1);
// else
resSectors = PQexecPrepared(conn, "index_sectors", 1, paramValues, paramLengths, paramFormats, 1);
if (PQresultStatus(resSectors) != PGRES_TUPLES_OK)
{
fprintf(stderr, "index_sectors: SELECT failed: %s", PQerrorMessage(conn));
PQclear(resSectors);
exit(EXIT_FAILURE);
}
if (PQftype(resSectors, 0) != PG_OID_INT4)
{
fprintf(stderr, "Sector value has unexpected type\n");
PQclear(resSectors);
exit(EXIT_FAILURE);
}
if (PQftype(resSectors, 1) != PG_OID_INT8)
{
fprintf(stderr, "Sector value has unexpected type\n");
PQclear(resSectors);
exit(EXIT_FAILURE);
}
rankTotalTuples = 0;
for (iSector = 0; iSector < PQntuples(resSectors); iSector++)
{
rankTotalTuples += PGint64(*((uint64_t *)PQgetvalue(resSectors, iSector, 1)));
}
rankStartTime = time(0);
for (iSector = 0; iSector <= PQntuples(resSectors); iSector++)
{
if (iSector > 0)
{
resPlaces = PQgetResult(conn);
if (PQresultStatus(resPlaces) != PGRES_TUPLES_OK)
{
fprintf(stderr, "index_sector_places: SELECT failed: %s", PQerrorMessage(conn));
PQclear(resPlaces);
exit(EXIT_FAILURE);
}
if (PQftype(resPlaces, 0) != PG_OID_INT8)
{
fprintf(stderr, "Place_id value has unexpected type\n");
PQclear(resPlaces);
exit(EXIT_FAILURE);
}
resNULL = PQgetResult(conn);
if (resNULL != NULL)
{
fprintf(stderr, "Unexpected non-null response\n");
exit(EXIT_FAILURE);
}
}
if (iSector < PQntuples(resSectors))
{
sector = PGint32(*((uint32_t *)PQgetvalue(resSectors, iSector, 0)));
// fprintf(stderr, "\n Starting sector %d size %ld\n", sector, PGint64(*((uint64_t *)PQgetvalue(resSectors, iSector, 1))));
// Get all the place_id's for this sector
paramRank = PGint32(rank);
paramValues[0] = (char *)&paramRank;
paramLengths[0] = sizeof(paramRank);
paramFormats[0] = 1;
paramSector = PGint32(sector);
paramValues[1] = (char *)&paramSector;
paramLengths[1] = sizeof(paramSector);
paramFormats[1] = 1;
if (rankTotalTuples-rankCountTuples < num_threads*1000)
{
iResult = PQsendQueryPrepared(conn, "index_nosector_places", 1, paramValues, paramLengths, paramFormats, 1);
}
else
{
iResult = PQsendQueryPrepared(conn, "index_sector_places", 2, paramValues, paramLengths, paramFormats, 1);
}
if (!iResult)
{
fprintf(stderr, "index_sector_places: SELECT failed: %s", PQerrorMessage(conn));
PQclear(resPlaces);
exit(EXIT_FAILURE);
}
}
if (iSector > 0)
{
count = 0;
rankPerSecond = 0;
tuples = PQntuples(resPlaces);
if (tuples > 0)
{
// Spawn threads
for (i = 0; i < num_threads; i++)
{
thread_data[i].res = resPlaces;
thread_data[i].tuples = tuples;
thread_data[i].count = &count;
thread_data[i].count_mutex = &count_mutex;
thread_data[i].writer = writer;
thread_data[i].writer_mutex = &writer_mutex;
thread_data[i].table = 1; // use placex table
pthread_create(&thread_data[i].thread, NULL, &nominatim_indexThread, (void *)&thread_data[i]);
}
// Monitor threads to give user feedback
sleepcount = 0;
while (count < tuples)
{
usleep(1000);
// Aim for one update per second
if (sleepcount++ > 500)
{
rankPerSecond = ((float)rankCountTuples + (float)count) / MAX(difftime(time(0), rankStartTime),1);
fprintf(stderr, " 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);
sleepcount = 0;
}
}
// Wait for everything to finish
for (i = 0; i < num_threads; i++)
{
pthread_join(thread_data[i].thread, NULL);
}
rankCountTuples += tuples;
}
// Finished sector
rankPerSecond = (float)rankCountTuples / MAX(difftime(time(0), rankStartTime),1);
fprintf(stderr, " Done %i in %i @ %f per second - ETA (seconds): %f\n", rankCountTuples, (int)(difftime(time(0), rankStartTime)), rankPerSecond, ((float)(rankTotalTuples - rankCountTuples))/rankPerSecond);
PQclear(resPlaces);
}
if (rankTotalTuples-rankCountTuples < num_threads*20 && iSector < PQntuples(resSectors))
{
iSector = PQntuples(resSectors) - 1;
}
}
// Finished rank
fprintf(stderr, "\r Done %i in %i @ %f per second - FINISHED \n\n", rankCountTuples, (int)(difftime(time(0), rankStartTime)), rankPerSecond);
PQclear(resSectors);
run_indexing(rank, 0, conn, num_threads, thread_data, structuredoutputfile);
}
if (rank_max == 30)
if (rank == 30)
{
// Close all connections
for (i = 0; i < num_threads; i++)

View File

@ -1090,13 +1090,12 @@ BEGIN
END IF;
IF OLD.indexed_status = 2 and NEW.indexed_status=0 THEN
-- do the reparenting: (finally here, so that ALL places in placex, that are needed for reparenting, are up to date)
-- do the reparenting: (finally here, because ALL places in placex, that are needed for reparenting, need to be up to date)
-- (the osm interpolationline in location_property_osmline was marked for reparenting in placex_insert/placex_delete with index_status = 2
-- => index.c: sets index_status back to 0
-- => triggers this function)
place_centroid := ST_PointOnSurface(NEW.linegeo);
-- mark descendants for reparenting
UPDATE placex SET indexed_status = 2 WHERE parent_place_id = OLD.place_id;
-- marking descendants for reparenting is not needed, because there are actually no descendants for interpolation lines
NEW.parent_place_id = get_interpolation_parent(NEW.osm_id, NEW.street, null, NEW.partition, place_centroid, NEW.linegeo); -- addr_place (3rd param) is not necessarily needed
return NEW;
END IF;
@ -1156,14 +1155,6 @@ BEGIN
END IF;
IF NEW.indexed_status != 0 OR OLD.indexed_status = 0 THEN
-- if a node(=>house), which is part of a interpolation line, changes (e.g. the street attribute) => mark this line for reparenting
-- (already here, because interpolation lines are reindexed before nodes, so in the second call it would be too late)
-- needed for test case features/db/import: Scenario: addr:street added to housenumbers
IF NEW.osm_type='N' and NEW.class='place' and NEW.type='house' THEN
-- Is this node part of an interpolation line? search for it in location_property_osmline and mark the interpolation line for reparenting
update location_property_osmline p set indexed_status = 2 from planet_osm_ways w where p.linegeo && NEW.geometry and p.osm_id = w.id and NEW.osm_id = any(w.nodes);
END IF;
RETURN NEW;
END IF;
@ -1906,7 +1897,7 @@ BEGIN
UPDATE placex set indexed_status = 100 where osm_type = OLD.osm_type and osm_id = OLD.osm_id and class = OLD.class and type = OLD.type;
-- interpolations are special
IF OLD.class = 'place' and OLD.type = 'houses' THEN
IF OLD.osm_type='W' and OLD.class = 'place' and OLD.type = 'houses' THEN
UPDATE location_property_osmline set indexed_status = 100 where osm_id = OLD.osm_id; -- osm_id = wayid (=old.osm_id)
END IF;
@ -1931,7 +1922,6 @@ BEGIN
--DEBUG: RAISE WARNING '-----------------------------------------------------------------------------------';
--DEBUG: RAISE WARNING 'place_insert: % % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type,st_area(NEW.geometry);
RAISE WARNING 'X3366 - place_isnert';
-- filter wrong tupels
IF ST_IsEmpty(NEW.geometry) OR NOT ST_IsValid(NEW.geometry) OR ST_X(ST_Centroid(NEW.geometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEW.geometry))::text in ('NaN','Infinity','-Infinity') THEN
INSERT INTO import_polygon_error values (NEW.osm_type, NEW.osm_id, NEW.class, NEW.type, NEW.name, NEW.country_code,
@ -1960,12 +1950,6 @@ BEGIN
-- To paraphrase, if there isn't an existing item
IF existingline.osm_id IS NULL THEN
IF existing.osm_type IS NOT NULL THEN
-- pathological case caused by the triggerless copy into place during initial import
-- force delete even for large areas, it will be reinserted later
UPDATE place set geometry = ST_SetSRID(ST_Point(0,0), 4326) where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type;
DELETE from place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type;
END IF;
-- insert new line into location_property_osmline, use function insert_osmline
i = insert_osmline(NEW.osm_id, NEW.housenumber, NEW.street, NEW.addr_place, NEW.postcode, NEW.country_code, NEW.geometry);
RETURN NEW;
@ -2002,7 +1986,6 @@ BEGIN
-- for interpolations invalidate all nodes on the line
update placex p set indexed_status = 2 from planet_osm_ways w where w.id = NEW.osm_id and p.osm_type = 'N' and p.osm_id = any(w.nodes);
RAISE WARNING 'X3399 - updated nodes of interpolation line';
RETURN NULL;
ELSE -- insert to placex
@ -2170,6 +2153,7 @@ BEGIN
admin_level = NEW.admin_level,
geometry = NEW.geometry
where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type;
IF NEW.class in ('place','boundary') AND NEW.type in ('postcode','postal_code') THEN
IF NEW.postcode IS NULL THEN
@ -2195,6 +2179,14 @@ BEGIN
indexed_status = 2,
geometry = NEW.geometry
where place_id = existingplacex.place_id;
-- if a node(=>house), which is part of a interpolation line, changes (e.g. the street attribute) => mark this line for reparenting
-- (already here, because interpolation lines are reindexed before nodes, so in the second call it would be too late)
-- needed for test case features/db/import: Scenario: addr:street added to housenumbers
IF NEW.osm_type='N' and NEW.class='place' and NEW.type='house' THEN
-- Is this node part of an interpolation line? search for it in location_property_osmline and mark the interpolation line for reparenting
update location_property_osmline p set indexed_status = 2 from planet_osm_ways w where p.linegeo && NEW.geometry and p.osm_id = w.id and NEW.osm_id = any(w.nodes);
END IF;
END IF;
@ -2202,7 +2194,7 @@ BEGIN
RETURN NULL;
END IF;
END;
END;
$$ LANGUAGE plpgsql;

View File

@ -84,6 +84,7 @@ GRANT SELECT ON location_property_aux TO "{www-user}";
CREATE TABLE location_property_tiger (linegeo GEOMETRY, place_id BIGINT, partition INTEGER, parent_place_id BIGINT, startnumber INTEGER, endnumber INTEGER, interpolationtype TEXT, postcode TEXT);
GRANT SELECT ON location_property_tiger TO "{www-user}";
drop table if exists location_property_osmline;
CREATE TABLE location_property_osmline (
linegeo GEOMETRY,
place_id BIGINT NOT NULL,
@ -99,9 +100,11 @@ CREATE TABLE location_property_osmline (
calculated_country_code VARCHAR(2),
geometry_sector INTEGER,
indexed_status INTEGER,
indexed_date TIMESTAMP);
CREATE UNIQUE INDEX idx_osmline_place_id ON location_property_osmline (place_id) {ts:search-index};
CREATE INDEX idx_osmline_parent_place_id ON location_property_osmline (parent_place_id) {ts:search-index};
indexed_date TIMESTAMP){ts:search-data};
CREATE UNIQUE INDEX idx_osmline_place_id ON location_property_osmline USING BTREE (place_id) {ts:search-index};
CREATE INDEX idx_osmline_parent_place_id ON location_property_osmline USING BTREE (parent_place_id) {ts:search-index};
CREATE INDEX idx_osmline_geometry_sector ON location_property_osmline USING BTREE (geometry_sector) {ts:address-index};
CREATE INDEX idx_osmline_linegeo ON location_property_osmline USING GIST (linegeo) {ts:search-index};
GRANT SELECT ON location_property_osmline TO "{www-user}";
drop table IF EXISTS search_name;

View File

@ -18,7 +18,10 @@ Feature: Reverse geocoding
| xml | 4
When looking up coordinates 53.9788769,13.0830313
And results contain valid boundingboxes
Scenario: Reverse geocoding for odd interpolated housenumber
Scenario: Reverse geocoding for even interpolated housenumber
@Tiger
Scenario: TIGER house number

View File

@ -37,7 +37,6 @@ Feature: Update of address interpolations
| object | parent_place_id | startnumber | endnumber
| W10 | W3 | 2 | 6
@Fail
Scenario: addr:street added to housenumbers
Given the scene parallel-road
And the place nodes