Modified filterResults() to only return a a single hit per keyword per object where a hit for that keyword on that object has not been seen before.

This commit is contained in:
esaunders 2017-01-18 17:31:50 -05:00
parent 401ac636b9
commit 4bf5988c40

View File

@ -264,7 +264,9 @@ public final class SearchRunner {
// mutable state: // mutable state:
private volatile boolean workerRunning; private volatile boolean workerRunning;
private List<String> keywordListNames; //guarded by SearchJobInfo.this private List<String> keywordListNames; //guarded by SearchJobInfo.this
private Map<Keyword, Set<String>> currentResults; //guarded by SearchJobInfo.this
// Map of keyword to the object ids that contain a hit
private Map<Keyword, Set<Long>> currentResults; //guarded by SearchJobInfo.this
private SearchRunner.Searcher currentSearcher; private SearchRunner.Searcher currentSearcher;
private AtomicLong moduleReferenceCount = new AtomicLong(0); private AtomicLong moduleReferenceCount = new AtomicLong(0);
private final Object finalSearchLock = new Object(); //used for a condition wait private final Object finalSearchLock = new Object(); //used for a condition wait
@ -296,11 +298,11 @@ public final class SearchRunner {
} }
} }
private synchronized Set<String> currentKeywordResults(Keyword k) { private synchronized Set<Long> currentKeywordResults(Keyword k) {
return currentResults.get(k); return currentResults.get(k);
} }
private synchronized void addKeywordResults(Keyword k, Set<String> resultsIDs) { private synchronized void addKeywordResults(Keyword k, Set<Long> resultsIDs) {
currentResults.put(k, resultsIDs); currentResults.put(k, resultsIDs);
} }
@ -470,8 +472,8 @@ public final class SearchRunner {
return null; return null;
} }
// calculate new results by substracting results already obtained in this ingest // Reduce the results of the query to only those hits we
// this creates a map of each keyword to the list of unique files that have that hit. // have not already seen.
QueryResults newResults = filterResults(queryResults); QueryResults newResults = filterResults(queryResults);
if (!newResults.getKeywords().isEmpty()) { if (!newResults.getKeywords().isEmpty()) {
@ -570,61 +572,67 @@ public final class SearchRunner {
} }
/** /**
* Over time, periodic searches can produce keyword hits that were seen
* in earlier searches (in addition to new hits).
* This method filters out all of the hits found in earlier * This method filters out all of the hits found in earlier
* periodic searches and returns only the results found by the most * periodic searches and returns only the results found by the most
* recent search. * recent search.
* *
* This method will only return hits for objects for which we haven't
* previously seen a hit for the keyword.
*
* @param queryResult The results returned by a keyword search. * @param queryResult The results returned by a keyword search.
* @return The set of hits found by the most recent search. * @return The set of hits found by the most recent search for objects
* that have not previously had a hit.
* *
*/ */
private QueryResults filterResults(QueryResults queryResult) { private QueryResults filterResults(QueryResults queryResult) {
// Create a new (empty) QueryResults object to hold the most recently found hits. // Create a new (empty) QueryResults object to hold the most recently
// found hits.
QueryResults newResults = new QueryResults(queryResult.getQuery(), queryResult.getKeywordList()); QueryResults newResults = new QueryResults(queryResult.getQuery(), queryResult.getKeywordList());
// For each keyword represented in the results.
for (Keyword keyword : queryResult.getKeywords()) { for (Keyword keyword : queryResult.getKeywords()) {
// These are all of the hits across all objects for the most recent search.
// This may well include duplicates of hits we've seen in earlier periodic searches.
List<KeywordHit> queryTermResults = queryResult.getResults(keyword); List<KeywordHit> queryTermResults = queryResult.getResults(keyword);
// Grab the set of solr document ids that have a hit for this // This will be used to build up the hits we haven't seen before
// keyword. // for this keyword.
Set<String> queryTermResultsIDs = new HashSet<>(); List<KeywordHit> newUniqueHits = new ArrayList<>();
for (KeywordHit hit : queryTermResults) {
queryTermResultsIDs.add(hit.getSolrDocumentId()); // Get the set of object ids seen in the past by this searcher
// for the given keyword.
Set<Long> curTermResults = job.currentKeywordResults(keyword);
if (curTermResults == null) {
// We create a new empty set if we haven't seen results for
// this keyword before.
curTermResults = new HashSet<>();
} }
// Get the set of document ids seen in the past by this searcher // For each hit for this keyword.
// for the given keyword. for (KeywordHit hit : queryTermResults) {
Set<String> curTermResults = job.currentKeywordResults(keyword); if (curTermResults.contains(hit.getSolrObjectId())) {
if (curTermResults == null) { // Skip the hit if we've already seen a hit for
// If we haven't seen any results in the past, we simply // this keyword in the object.
// add all of the incoming results to the outgoing collection. continue;
job.addKeywordResults(keyword, queryTermResultsIDs);
newResults.addResult(keyword, queryTermResults);
} else {
// Otherwise, we have seen results for this keyword in the past.
// For each of the keyword hits in the incoming results we check to
// see if we have seen the document id previously.
for (KeywordHit res : queryTermResults) {
if (!curTermResults.contains(res.getSolrDocumentId())) {
// We have not seen the document id before so we add
// the keyword hit to the results
List<KeywordHit> newKeywordHits = newResults.getResults(keyword);
if (newKeywordHits == null) {
// Create an empty list to hold the new hits.
newKeywordHits = new ArrayList<>();
newResults.addResult(keyword, newKeywordHits);
}
// Add the new hit to the list.
newKeywordHits.add(res);
// Add the document id to the set of document ids
// that have been seen for this keyword.
curTermResults.add(res.getSolrDocumentId());
}
} }
// We haven't seen the hit before so add it to list of new
// unique hits.
newUniqueHits.add(hit);
// Add the object id to the results we've seen for this
// keyword.
curTermResults.add(hit.getSolrObjectId());
} }
// Update the job with the list of objects for which we have
// seen hits for the current keyword.
job.addKeywordResults(keyword, curTermResults);
// Add the new hits for the current keyword into the results
// to be returned.
newResults.addResult(keyword, newUniqueHits);
} }
return newResults; return newResults;