From d1a224bad4b5a3b8fdf077b1b67d529abe9d9278 Mon Sep 17 00:00:00 2001 From: Brian Quinion Date: Thu, 20 Dec 2012 17:47:57 +0000 Subject: [PATCH] performance: for low frequency name terms disable use of postgresql address index. Theshold is probably about 1000 but optimal value will probably depend on hardware. --- settings/settings.php | 1 + website/search.php | 25 +++++++++++++++++++------ 2 files changed, 20 insertions(+), 6 deletions(-) diff --git a/settings/settings.php b/settings/settings.php index afa20d20..6f6f843b 100644 --- a/settings/settings.php +++ b/settings/settings.php @@ -60,6 +60,7 @@ @define('CONST_Suggestions_Enabled', false); @define('CONST_Search_TryDroppedAddressTerms', false); + @define('CONST_Search_NameOnlySearchFrequencyThreshold', false); // Set to zero to disable polygon output @define('CONST_PolygonOutput_MaximumTypes', 1); diff --git a/website/search.php b/website/search.php index 265f2011..5947fbc7 100755 --- a/website/search.php +++ b/website/search.php @@ -324,7 +324,7 @@ { foreach($aSearchWords as $aSearchTerm) { - $aNewSearch = $aSearch; + $aNewSearch = $aSearch; if ($aSearchTerm['country_code']) { $aNewSearch['sCountryCode'] = strtolower($aSearchTerm['country_code']); @@ -392,7 +392,7 @@ { // Check which tokens we have, get the ID numbers - $sSQL = 'select word_id,word_token, word, class, type, location, country_code, operator'; + $sSQL = 'select word_id,word_token, word, class, type, location, country_code, operator, search_name_count'; $sSQL .= ' from word where word_token in ('.join(',',array_map("getDBQuoted",$aTokens)).')'; $sSQL .= ' and search_name_count < '.CONST_Max_Word_Frequency; // $sSQL .= ' group by word_token, word, class, type, location, country_code'; @@ -419,7 +419,7 @@ { $aValidTokens[$aToken['word_token']] = array($aToken); } - if ($aToken['word_token'][0]==' ' && !$aToken['class'] && !$aToken['country_code']) $aPossibleMainWordIDs[$aToken['word_id']] = 1; + if ($aToken['word_token'][0]==' ' && !$aToken['class'] && !$aToken['country_code']) $aPossibleMainWordIDs[$aToken['word_id']] = 1 + $aToken['search_name_count']; } if (CONST_Debug) var_Dump($aPhrases, $aValidTokens); @@ -527,7 +527,7 @@ { $aSearch = $aCurrentSearch; $aSearch['iSearchRank']++; - if (($sPhraseType == '' || $sPhraseType == 'country') && $aSearchTerm['country_code'] !== null && $aSearchTerm['country_code'] != '0') + if (($sPhraseType == '' || $sPhraseType == 'country') && !empty($aSearchTerm['country_code']) && $aSearchTerm['country_code'] != '0') { if ($aSearch['sCountryCode'] === false) { @@ -860,7 +860,7 @@ if (CONST_Debug) var_dump('
',$aSearch); if (CONST_Debug) _debugDumpGroupedSearches(array($iGroupedRank => array($aSearch)), $aValidTokens); $aPlaceIDs = array(); - + // First we need a position, either aName or fLat or both $aTerms = array(); $aOrder = array(); @@ -868,7 +868,20 @@ // TODO: filter out the pointless search terms (2 letter name tokens and less) // they might be right - but they are just too darned expensive to run if (sizeof($aSearch['aName'])) $aTerms[] = "name_vector @> ARRAY[".join($aSearch['aName'],",")."]"; - if (sizeof($aSearch['aAddress']) && $aSearch['aName'] != $aSearch['aAddress']) $aTerms[] = "nameaddress_vector @> ARRAY[".join($aSearch['aAddress'],",")."]"; + if (sizeof($aSearch['aAddress']) && $aSearch['aName'] != $aSearch['aAddress']) + { + // For infrequent name terms disable index usage for address + if (CONST_Search_NameOnlySearchFrequencyThreshold && + sizeof($aSearch['aName']) == 1 && + $aPossibleMainWordIDs[$aSearch['aName'][reset($aSearch['aName'])]] < CONST_Search_NameOnlySearchFrequencyThreshold) + { + $aTerms[] = "array_cat(nameaddress_vector,ARRAY[]::integer[]) @> ARRAY[".join($aSearch['aAddress'],",")."]"; + } + else + { + $aTerms[] = "nameaddress_vector @> ARRAY[".join($aSearch['aAddress'],",")."]"; + } + } if ($aSearch['sCountryCode']) $aTerms[] = "country_code = '".pg_escape_string($aSearch['sCountryCode'])."'"; if ($aSearch['sHouseNumber']) $aTerms[] = "address_rank in (26,27)"; if ($aSearch['fLon'] && $aSearch['fLat']) -- 2.39.5