/** * Builds an inverted index shard (word --> {docs it appears in}) * for the current batch of SEEN_URLS_BEFORE_UPDATE_SCHEDULER many pages. * This inverted index shard is then merged by a queue_server * into the inverted index of the current generation of the crawl. * The complete inverted index for the whole crawl is built out of these * inverted indexes for generations. The point of computing a partial * inverted index on the fetcher is to reduce some of the computational * burden on the queue server. The resulting mini index computed by * buildMiniInvertedIndex() is stored in * $this->found_sites[self::INVERTED_INDEX] * */ function buildMiniInvertedIndex() { $start_time = microtime(); $keypad = ""; crawlLog(" Start building mini inverted index ... Current Memory:" . memory_get_usage()); $num_seen = count($this->found_sites[self::SEEN_URLS]); $this->num_seen_sites += $num_seen; /* for the fetcher we are not saving the index shards so name doesn't matter. */ if (!isset($this->found_sites[self::INVERTED_INDEX][$this->current_server])) { $this->found_sites[self::INVERTED_INDEX][$this->current_server] = new IndexShard("fetcher_shard_{$this->current_server}"); } for ($i = 0; $i < $num_seen; $i++) { $interim_time = microtime(); $site = $this->found_sites[self::SEEN_URLS][$i]; if (!isset($site[self::HASH]) || isset($site[self::ROBOT_METAS]) && in_array("JUSTFOLLOW", $site[self::ROBOT_METAS])) { continue; } $doc_rank = false; if ($this->crawl_type == self::ARCHIVE_CRAWL && isset($this->archive_iterator)) { $doc_rank = $this->archive_iterator->weight($site); } if (isset($site[self::TYPE]) && $site[self::TYPE] == "link") { $is_link = true; $doc_keys = $site[self::HTTP_CODE]; $site_url = $site[self::TITLE]; $host = UrlParser::getHost($site_url); $link_parts = explode('|', $site[self::HASH]); if (isset($link_parts[5])) { $link_origin = $link_parts[5]; } else { $link_origin = $site_url; } $meta_ids = PhraseParser::calculateLinkMetas($site_url, $host, $site[self::DESCRIPTION], $link_origin); } else { $is_link = false; $site_url = str_replace('|', "%7C", $site[self::URL]); $host = UrlParser::getHost($site_url); $doc_keys = crawlHash($site_url, true) . $site[self::HASH] . "d" . substr(crawlHash($host . "/", true), 1); $meta_ids = PhraseParser::calculateMetas($site, $this->video_sources); } $word_lists = array(); /* self::JUST_METAS check to avoid getting sitemaps in results for popular words */ $lang = NULL; if (!isset($site[self::JUST_METAS])) { $host_words = UrlParser::getWordsIfHostUrl($site_url); $path_words = UrlParser::getWordsLastPathPartUrl($site_url); if ($is_link) { $phrase_string = $site[self::DESCRIPTION]; } else { if (isset($site[self::LANG])) { if (isset($this->programming_language_extension[$site[self::LANG]])) { $phrase_string = $site[self::DESCRIPTION]; } else { $phrase_string = $host_words . " " . $site[self::TITLE] . " " . $path_words . " " . $site[self::DESCRIPTION]; } } else { $phrase_string = $host_words . " " . $site[self::TITLE] . " " . $path_words . " " . $site[self::DESCRIPTION]; } } if (isset($site[self::LANG])) { $lang = guessLocaleFromString(mb_substr($site[self::DESCRIPTION], 0, AD_HOC_TITLE_LENGTH), $site[self::LANG]); } $word_lists = PhraseParser::extractPhrasesInLists($phrase_string, $lang); $len = strlen($phrase_string); if (isset($this->programming_language_extension[$lang]) || PhraseParser::computeSafeSearchScore($word_lists, $len) < 0.012) { $meta_ids[] = "safe:true"; $safe = true; } else { $meta_ids[] = "safe:false"; $safe = false; } } if (!$is_link) { //store inlinks so they can be searched by $num_links = count($site[self::LINKS]); if ($num_links > 0) { $link_rank = false; if ($doc_rank !== false) { $link_rank = max($doc_rank - 1, 1); } } else { $link_rank = false; } } $num_queue_servers = count($this->queue_servers); if (isset($site[self::USER_RANKS]) && count($site[self::USER_RANKS]) > 0) { $score_keys = ""; foreach ($site[self::USER_RANKS] as $label => $score) { $score_keys .= packInt($score); } if (strlen($score_keys) % 8 != 0) { $score_keys .= $keypad; } $doc_keys .= $score_keys; } $this->found_sites[self::INVERTED_INDEX][$this->current_server]->addDocumentWords($doc_keys, self::NEEDS_OFFSET_FLAG, $word_lists, $meta_ids, PhraseParser::$materialized_metas, true, $doc_rank); /* $this->no_process_links is set when doing things like mix recrawls. In this case links likely already will appear in what indexing, so don't index again. $site[self::JUST_META] is set when have a sitemap or robots.txt (this case set later). In this case link info is not particularly useful for indexing and can greatly slow building inverted index. */ if (!$this->no_process_links && !isset($site[self::JUST_METAS]) && !isset($this->programming_language_extension[$lang])) { foreach ($site[self::LINKS] as $url => $link_text) { /* this mysterious check means won't index links from robots.txt. Sitemap will still be in TO_CRAWL, but that's done elsewhere */ if (strlen($url) == 0 || is_numeric($url)) { continue; } $link_host = UrlParser::getHost($url); if (strlen($link_host) == 0) { continue; } $part_num = calculatePartition($link_host, $num_queue_servers); $summary = array(); if (!isset($this->found_sites[self::LINK_SEEN_URLS][$part_num])) { $this->found_sites[self::LINK_SEEN_URLS][$part_num] = array(); } $elink_flag = $link_host != $host ? true : false; $link_text = strip_tags($link_text); $ref = $elink_flag ? "eref" : "iref"; $url = str_replace('|', "%7C", $url); $link_id = "url|" . $url . "|text|" . urlencode($link_text) . "|{$ref}|" . $site_url; $elink_flag_string = $elink_flag ? "e" : "i"; $link_keys = crawlHash($url, true) . crawlHash($link_id, true) . $elink_flag_string . substr(crawlHash($host . "/", true), 1); $summary[self::URL] = $link_id; $summary[self::TITLE] = $url; // stripping html to be on the safe side $summary[self::DESCRIPTION] = $link_text; $summary[self::TIMESTAMP] = $site[self::TIMESTAMP]; $summary[self::ENCODING] = $site[self::ENCODING]; $summary[self::HASH] = $link_id; $summary[self::TYPE] = "link"; $summary[self::HTTP_CODE] = $link_keys; $summary[self::LANG] = $lang; $this->found_sites[self::LINK_SEEN_URLS][$part_num][] = $summary; $link_word_lists = PhraseParser::extractPhrasesInLists($link_text, $lang); $link_meta_ids = PhraseParser::calculateLinkMetas($url, $link_host, $link_text, $site_url); if (!isset($this->found_sites[self::INVERTED_INDEX][$part_num])) { $this->found_sites[self::INVERTED_INDEX][$part_num] = new IndexShard("fetcher_shard_{$part_num}"); } $this->found_sites[self::INVERTED_INDEX][$part_num]->addDocumentWords($link_keys, self::NEEDS_OFFSET_FLAG, $link_word_lists, $link_meta_ids, PhraseParser::$materialized_metas, false, $link_rank); } } $iterim_elapse = changeInMicrotime($interim_time); if ($iterim_elapse > 5) { crawlLog("..Inverting " . $site[self::URL] . "...took > 5s."); } crawlTimeoutLog("..Still building inverted index. Have processed " . "%s of %s documents.\nLast url processed was %s.", $i, $num_seen, $site[self::URL]); } if ($this->crawl_type == self::ARCHIVE_CRAWL) { $this->recrawl_check_scheduler = true; } crawlLog(" Build mini inverted index time " . changeInMicrotime($start_time)); }
/** * Used to recompute both the index shards and the dictionary * of an index archive. The first step involves re-extracting the * word into an inverted index from the summaries' web_archives. * Then a reindex is done. * * @param string $archive_path file path to a IndexArchiveBundle */ function rebuildIndexArchive($archive_path) { $archive_type = $this->getArchiveKind($archive_path); if ($archive_type != "IndexArchiveBundle") { $this->badFormatMessageAndExit($archive_path); } $info = $archive_type::getArchiveInfo($archive_path); $tmp = unserialize($info["DESCRIPTION"]); $video_sources = $tmp[self::VIDEO_SOURCES]; $generation_info = unserialize(file_get_contents("{$archive_path}/generation.txt")); $num_generations = $generation_info['ACTIVE'] + 1; $archive = new WebArchiveBundle($archive_path . "/summaries"); $seen = 0; $generation = 0; $keypad = ""; while ($generation < $num_generations) { $partition = $archive->getPartition($generation, false); $shard_name = $archive_path . "/posting_doc_shards/index{$generation}"; crawlLog("Processing partition {$generation}"); if (file_exists($shard_name)) { crawlLog("..Unlinking old shard {$generation}"); @unlink($shard_name); } $shard = new IndexShard($shard_name, $generation, NUM_DOCS_PER_GENERATION, true); $seen_partition = 0; while ($seen_partition < $partition->count) { $num_to_get = min($partition->count - $seen_partition, 8000); $offset = $partition->iterator_pos; $objects = $partition->nextObjects($num_to_get); $cnt = 0; foreach ($objects as $object) { $cnt++; $site = $object[1]; if (isset($site[self::TYPE]) && $site[self::TYPE] == "link") { $is_link = true; $doc_keys = $site[self::HTTP_CODE]; $site_url = $site[self::TITLE]; $host = UrlParser::getHost($site_url); $link_parts = explode('|', $site[self::HASH]); if (isset($link_parts[5])) { $link_origin = $link_parts[5]; } else { $link_origin = $site_url; } $meta_ids = PhraseParser::calculateLinkMetas($site_url, $host, $site[self::DESCRIPTION], $link_origin); $link_to = "LINK TO:"; } else { $is_link = false; $site_url = str_replace('|', "%7C", $site[self::URL]); $host = UrlParser::getHost($site_url); $doc_keys = crawlHash($site_url, true) . $site[self::HASH] . "d" . substr(crawlHash($host . "/", true), 1); $meta_ids = PhraseParser::calculateMetas($site, $video_sources); $link_to = ""; } $so_far_cnt = $seen_partition + $cnt; $time_out_message = "..still processing {$so_far_cnt} " . "of {$partition->count} in partition {$generation}." . "\n..Last processed was: " . ($seen + 1) . ". {$link_to}{$site_url}. "; crawlTimeoutLog($time_out_message); $seen++; $word_lists = array(); /* self::JUST_METAS check to avoid getting sitemaps in results for popular words */ $lang = NULL; if (!isset($site[self::JUST_METAS])) { $host_words = UrlParser::getWordsIfHostUrl($site_url); $path_words = UrlParser::getWordsLastPathPartUrl($site_url); if ($is_link) { $phrase_string = $site[self::DESCRIPTION]; } else { $phrase_string = $host_words . " " . $site[self::TITLE] . " " . $path_words . " " . $site[self::DESCRIPTION]; } if (isset($site[self::LANG])) { $lang = guessLocaleFromString(mb_substr($site[self::DESCRIPTION], 0, AD_HOC_TITLE_LENGTH), $site[self::LANG]); } $word_lists = PhraseParser::extractPhrasesInLists($phrase_string, $lang); $len = strlen($phrase_string); if (PhraseParser::computeSafeSearchScore($word_lists, $len) < 0.012) { $meta_ids[] = "safe:true"; $safe = true; } else { $meta_ids[] = "safe:false"; $safe = false; } } if (isset($site[self::USER_RANKS]) && count($site[self::USER_RANKS]) > 0) { $score_keys = ""; foreach ($site[self::USER_RANKS] as $label => $score) { $score_keys .= packInt($score); } if (strlen($score_keys) % 8 != 0) { $score_keys .= $keypad; } $doc_keys .= $score_keys; } $shard->addDocumentWords($doc_keys, $offset, $word_lists, $meta_ids, PhraseParser::$materialized_metas, true, false); $offset = $object[0]; } $seen_partition += $num_to_get; } $shard->save(false, true); $generation++; } $this->reindexIndexArchive($archive_path); }