Examples with SearchHits nl.inl.blacklab.searches.SearchHits used on opensource projects

Example 1 with SearchHits

use of nl.inl.blacklab.searches.SearchHits in project BlackLab by INL.

the class HitGroupsTokenFrequencies method get.

/**
 * Get the token frequencies for the given query and hit property.
 *
 * @param source query to find token frequencies for
 * @param requestedGroupingProperty
 * @return token frequencies
 */
public static HitGroups get(SearchHits source, HitProperty requestedGroupingProperty) {
    QueryInfo queryInfo = source.queryInfo();
    Query filterQuery = source.getFilterQuery();
    SearchSettings searchSettings = source.searchSettings();
    try {
        /**
         * This is where we store our groups while we're computing/gathering them. Maps from group Id to number of hits (left) and number of docs (right)
         */
        final ConcurrentHashMap<GroupIdHash, MutablePair<Integer, Integer>> occurances = new ConcurrentHashMap<>();
        final BlackLabIndex index = queryInfo.index();
        /**
         * Document properties that are used in the grouping. (e.g. for query "all tokens, grouped by lemma + document year", will contain DocProperty("document year")
         * This is not necessarily limited to just metadata, can also contain any other DocProperties such as document ID, document length, etc.
         */
        final List<DocProperty> docProperties = new ArrayList<>();
        /**
         * Token properties that need to be grouped on, with sensitivity (case-sensitive grouping or not) and Terms
         */
        final List<Triple<AnnotationForwardIndex, MatchSensitivity, Terms>> hitProperties = new ArrayList<>();
        /**
         * Stores the original index every (doc|hit)property has in the original interleaved/intertwined list.
         * The requestedGroupingProperty sometimes represents more than one property (in the form of HitPropertyMultiple) such as 3 properties: [token text, document year, token lemma]
         * The groups always get an id that is (roughly) the concatenation of the properties (in the example case [token text, document year, token lemma]),
         * and it's important this id contains the respective values in the same order.
         * We need to keep this list because otherwise we'd potentially change the order.
         *
         * Integer contains index in the source list (docProperties or hitProperties, from just above)
         * Boolean is true when origin list was docProperties, false for hitProperties.
         */
        final List<Pair<Integer, Boolean>> originalOrderOfUnpackedProperties = new ArrayList<>();
        // Unpack the requestedGroupingProperty into its constituents and sort those into the appropriate categories: hit and doc properties.
        {
            List<HitProperty> props = requestedGroupingProperty.props() != null ? requestedGroupingProperty.props() : Arrays.asList(requestedGroupingProperty);
            for (HitProperty p : props) {
                final DocProperty asDocPropIfApplicable = p.docPropsOnly();
                if (asDocPropIfApplicable != null) {
                    // property can be converted to docProperty (applies to the document instead of the token/hit)
                    if (DEBUG && asDocPropIfApplicable.props() != null) {
                        throw new RuntimeException("Nested PropertyMultiples detected, should never happen (when this code was originally written)");
                    }
                    final int positionInUnpackedList = docProperties.size();
                    docProperties.add(asDocPropIfApplicable);
                    originalOrderOfUnpackedProperties.add(Pair.of(positionInUnpackedList, true));
                } else {
                    // Property couldn't be converted to DocProperty (is null). The current property is an actual HitProperty (applies to annotation/token/hit value)
                    List<Annotation> annot = p.needsContext();
                    if (DEBUG && (annot == null || annot.size() != 1)) {
                        throw new RuntimeException("Grouping property does not apply to singular annotation (nested propertymultiple? non-annotation grouping?) should never happen.");
                    }
                    final int positionInUnpackedList = hitProperties.size();
                    final AnnotationForwardIndex annotationFI = index.annotationForwardIndex(annot.get(0));
                    hitProperties.add(Triple.of(annotationFI, p.getSensitivities().get(0), annotationFI.terms()));
                    originalOrderOfUnpackedProperties.add(Pair.of(positionInUnpackedList, false));
                }
            }
        }
        final int numAnnotations = hitProperties.size();
        long numberOfDocsProcessed;
        final AtomicInteger numberOfHitsProcessed = new AtomicInteger();
        final AtomicBoolean hitMaxHitsToProcess = new AtomicBoolean(false);
        try (final BlockTimer c = BlockTimer.create("Top Level")) {
            final List<Integer> docIds = new ArrayList<>();
            try (BlockTimer d = c.child("Gathering documents")) {
                queryInfo.index().searcher().search(filterQuery == null ? new MatchAllDocsQuery() : filterQuery, new SimpleCollector() {

                    private int docBase;

                    @Override
                    protected void doSetNextReader(LeafReaderContext context) throws IOException {
                        docBase = context.docBase;
                        super.doSetNextReader(context);
                    }

                    @Override
                    public void collect(int docId) throws IOException {
                        int globalDocId = docId + docBase;
                        docIds.add(globalDocId);
                    }

                    @Override
                    public boolean needsScores() {
                        return false;
                    }
                });
            }
            numberOfDocsProcessed = docIds.size();
            final IndexReader reader = queryInfo.index().reader();
            final int[] minusOne = new int[] { -1 };
            // What we do instead is for every document just retrieve how many tokens it contains (from its metadata), and add that count to the appropriate group
            if (hitProperties.isEmpty()) {
                try (BlockTimer f = c.child("Grouping documents (metadata only path)")) {
                    String fieldName = index.mainAnnotatedField().name();
                    DocPropertyAnnotatedFieldLength propTokens = new DocPropertyAnnotatedFieldLength(index, fieldName);
                    final int[] emptyTokenValuesArray = new int[0];
                    docIds.parallelStream().forEach(docId -> {
                        // ignore "extra closing token"
                        final int docLength = (int) propTokens.get(docId) - subtractClosingToken;
                        final DocResult synthesizedDocResult = DocResult.fromDoc(queryInfo, new PropertyValueDoc(new DocImpl(queryInfo.index(), docId)), 0, docLength);
                        final PropertyValue[] metadataValuesForGroup = new PropertyValue[docProperties.size()];
                        for (int i = 0; i < docProperties.size(); ++i) {
                            metadataValuesForGroup[i] = docProperties.get(i).get(synthesizedDocResult);
                        }
                        // precompute, it's the same for all hits in document
                        final int metadataValuesHash = Arrays.hashCode(metadataValuesForGroup);
                        numberOfHitsProcessed.addAndGet(docLength);
                        // Add all tokens in document to the group.
                        final GroupIdHash groupId = new GroupIdHash(emptyTokenValuesArray, emptyTokenValuesArray, metadataValuesForGroup, metadataValuesHash);
                        occurances.compute(groupId, (__, groupSizes) -> {
                            if (groupSizes != null) {
                                groupSizes.left += docLength;
                                groupSizes.right += 1;
                                return groupSizes;
                            } else {
                                return MutablePair.of(docLength, 1);
                            }
                        });
                    });
                }
            } else {
                final int maxHitsToProcess = searchSettings.maxHitsToProcess() > 0 ? searchSettings.maxHitsToProcess() : Integer.MAX_VALUE;
                final IntUnaryOperator incrementUntilMax = (v) -> v < maxHitsToProcess ? v + 1 : v;
                final String fieldName = index.mainAnnotatedField().name();
                final String lengthTokensFieldName = AnnotatedFieldNameUtil.lengthTokensField(fieldName);
                numberOfDocsProcessed = docIds.parallelStream().filter(docId -> {
                    try {
                        // Step 1: read all values for the to-be-grouped annotations for this document
                        // This will create one int[] for every annotation, containing ids that map to the values for this document for this annotation
                        final Document doc = reader.document(docId);
                        final List<int[]> tokenValuesPerAnnotation = new ArrayList<>();
                        try (BlockTimer e = c.child("Read annotations from forward index")) {
                            for (Triple<AnnotationForwardIndex, MatchSensitivity, Terms> annot : hitProperties) {
                                final String annotationFIName = annot.getLeft().annotation().forwardIndexIdField();
                                final int fiid = doc.getField(annotationFIName).numericValue().intValue();
                                final List<int[]> tokenValues = annot.getLeft().retrievePartsInt(fiid, minusOne, minusOne);
                                tokenValuesPerAnnotation.addAll(tokenValues);
                            }
                        }
                        // Step 2: retrieve the to-be-grouped metadata for this document
                        // ignore "extra closing token"
                        int docLength = Integer.parseInt(doc.get(lengthTokensFieldName)) - subtractClosingToken;
                        final DocResult synthesizedDocResult = DocResult.fromDoc(queryInfo, new PropertyValueDoc(new DocImpl(queryInfo.index(), docId)), 0, docLength);
                        final PropertyValue[] metadataValuesForGroup = !docProperties.isEmpty() ? new PropertyValue[docProperties.size()] : null;
                        for (int i = 0; i < docProperties.size(); ++i) {
                            metadataValuesForGroup[i] = docProperties.get(i).get(synthesizedDocResult);
                        }
                        // precompute, it's the same for all hits in document
                        final int metadataValuesHash = Arrays.hashCode(metadataValuesForGroup);
                        // now we have all values for all relevant annotations for this document
                        // iterate again and pair up the nth entries for all annotations, then store that as a group.
                        /**
                         * Bookkeeping: track which groups we've already seen in this document,
                         * so we only count this document once per group
                         */
                        HashSet<GroupIdHash> groupsInThisDocument = new HashSet<>();
                        try (BlockTimer f = c.child("Group tokens")) {
                            for (int tokenIndex = 0; tokenIndex < docLength; ++tokenIndex) {
                                if (numberOfHitsProcessed.getAndUpdate(incrementUntilMax) >= maxHitsToProcess) {
                                    hitMaxHitsToProcess.set(true);
                                    // true if any token of this document made the cut, false if we escaped immediately
                                    return tokenIndex > 0;
                                }
                                // Unfortunate fact: token ids are case-sensitive, and in order to group on a token's values case and diacritics insensitively,
                                // we need to actually group by their "sort positions" - which is just the index the term would have if all terms would have been sorted
                                // so in essence it's also an "id", but a case-insensitive one.
                                // we could further optimize to not do this step when grouping sensitively by making a specialized instance of the GroupIdHash class
                                // that hashes the token ids instead of the sortpositions in that case.
                                int[] annotationValuesForThisToken = new int[numAnnotations];
                                int[] sortPositions = new int[annotationValuesForThisToken.length];
                                for (int annotationIndex = 0; annotationIndex < numAnnotations; ++annotationIndex) {
                                    int[] tokenValuesThisAnnotation = tokenValuesPerAnnotation.get(annotationIndex);
                                    final int termId = annotationValuesForThisToken[annotationIndex] = tokenValuesThisAnnotation[tokenIndex];
                                    Triple<AnnotationForwardIndex, MatchSensitivity, Terms> currentHitProp = hitProperties.get(annotationIndex);
                                    MatchSensitivity matchSensitivity = currentHitProp.getMiddle();
                                    Terms terms = currentHitProp.getRight();
                                    sortPositions[annotationIndex] = terms.idToSortPosition(termId, matchSensitivity);
                                }
                                final GroupIdHash groupId = new GroupIdHash(annotationValuesForThisToken, sortPositions, metadataValuesForGroup, metadataValuesHash);
                                occurances.compute(groupId, (__, groupSize) -> {
                                    if (groupSize != null) {
                                        groupSize.left += 1;
                                        // second (or more) occurance of these token values in this document
                                        groupSize.right += groupsInThisDocument.add(groupId) ? 1 : 0;
                                        return groupSize;
                                    } else {
                                        // should always return true, but we need to add this group anyway!
                                        return MutablePair.of(1, groupsInThisDocument.add(groupId) ? 1 : 0);
                                    }
                                });
                            }
                        }
                    } catch (IOException e) {
                        throw BlackLabRuntimeException.wrap(e);
                    }
                    return true;
                }).count();
                logger.trace("Number of processed docs: " + numberOfDocsProcessed);
            }
        }
        Set<PropertyValue> duplicateGroupsDebug = DEBUG ? new HashSet<PropertyValue>() : null;
        List<HitGroup> groups;
        try (final BlockTimer c = BlockTimer.create("Resolve string values for tokens")) {
            final int numMetadataValues = docProperties.size();
            groups = occurances.entrySet().parallelStream().map(e -> {
                final int groupSizeHits = e.getValue().getLeft();
                final int groupSizeDocs = e.getValue().getRight();
                final int[] annotationValues = e.getKey().tokenIds;
                final PropertyValue[] metadataValues = e.getKey().metadataValues;
                // allocate new - is not copied when moving into propertyvaluemultiple
                final PropertyValue[] groupIdAsList = new PropertyValue[numAnnotations + numMetadataValues];
                // Convert all raw values (integers) into their appropriate PropertyValues
                // Taking care to preserve the order of the resultant PropertyValues with the order of the input HitProperties
                int indexInOutput = 0;
                for (Pair<Integer, Boolean> p : originalOrderOfUnpackedProperties) {
                    final int indexInInput = p.getLeft();
                    if (p.getRight()) {
                        // is docprop, add PropertyValue as-is
                        groupIdAsList[indexInOutput++] = metadataValues[indexInInput];
                    } else {
                        // is hitprop, convert value to PropertyValue.
                        Annotation annot = hitProperties.get(indexInInput).getLeft().annotation();
                        MatchSensitivity sens = hitProperties.get(indexInInput).getMiddle();
                        groupIdAsList[indexInOutput++] = new PropertyValueContextWords(index, annot, sens, new int[] { annotationValues[indexInInput] }, false);
                    }
                }
                PropertyValue groupId = groupIdAsList.length > 1 ? new PropertyValueMultiple(groupIdAsList) : groupIdAsList[0];
                if (DEBUG) {
                    synchronized (duplicateGroupsDebug) {
                        if (!duplicateGroupsDebug.add(groupId)) {
                            throw new RuntimeException("Identical groups - should never happen");
                        }
                    }
                }
                return new HitGroupWithoutResults(queryInfo, groupId, groupSizeHits, groupSizeDocs, false, false);
            }).collect(Collectors.toList());
        }
        logger.debug("fast path used for grouping");
        ResultsStats hitsStats = new ResultsStatsStatic(numberOfHitsProcessed.get(), numberOfHitsProcessed.get(), new MaxStats(hitMaxHitsToProcess.get(), hitMaxHitsToProcess.get()));
        ResultsStats docsStats = new ResultsStatsStatic((int) numberOfDocsProcessed, (int) numberOfDocsProcessed, new MaxStats(hitMaxHitsToProcess.get(), hitMaxHitsToProcess.get()));
        return HitGroups.fromList(queryInfo, groups, requestedGroupingProperty, null, null, hitsStats, docsStats);
    } catch (IOException e) {
        throw BlackLabRuntimeException.wrap(e);
    }
}

Example 2 with SearchHits

use of nl.inl.blacklab.searches.SearchHits in project BlackLab by INL.

the class RequestHandlerHits method getQueryForHitsInSpecificGroupOnly.

/**
 * Translate the normal Hits query in the searchparams object into a query yielding only those Hits in the group with the specified PropertyValue
 *
 * @param viewGroupVal
 * @param groupByProp
 * @param hitsGrouped
 *
 * @return the SearchHits that will yield the hits, or null if the search could not be reconstructed.
 * @throws BlsException
 * @throws InvalidQuery
 */
private SearchHits getQueryForHitsInSpecificGroupOnly(PropertyValue viewGroupVal, HitProperty groupByProp, HitGroups hitsGrouped) throws BlsException, InvalidQuery {
    // see if we can enhance this query
    if (hitsGrouped.isSample())
        return null;
    // see if this query matches only singular tokens
    // (we can't enhance multi-token queries such as ngrams yet)
    TextPattern tp = searchParam.getPattern();
    if (!tp.toQuery(QueryInfo.create(blIndex())).producesSingleTokens())
        return null;
    // Alright, the original query for the Hits lends itself to enhancement.
    // Create the Query that will do the metadata filtering portion. (Token filtering is done through the TextPattern above)
    BooleanQuery.Builder fqb = new BooleanQuery.Builder();
    boolean usedFilter = false;
    if (searchParam.getFilterQuery() != null) {
        fqb.add(searchParam.getFilterQuery(), Occur.FILTER);
        usedFilter = true;
    }
    // Decode the grouping properties, and the values for those properties in the requested group.
    // So we can enhance the BooleanQuery and TextPattern with these  criteria
    List<PropertyValue> vals = viewGroupVal instanceof PropertyValueMultiple ? ((PropertyValueMultiple) viewGroupVal).values() : Arrays.asList(viewGroupVal);
    List<HitProperty> props = groupByProp instanceof HitPropertyMultiple ? ((HitPropertyMultiple) groupByProp).props() : Arrays.asList(groupByProp);
    int i = 0;
    for (HitProperty p : props) {
        if (p instanceof HitPropertyHitText) {
            String valueForAnnotation = vals.get(i).toString();
            HitPropertyHitText prop = ((HitPropertyHitText) p);
            Annotation annot = prop.needsContext().get(0);
            MatchSensitivity sensitivity = prop.getSensitivities().get(0);
            tp = new TextPatternAnd(tp, new TextPatternAnnotation(annot.name(), new TextPatternSensitive(sensitivity, new TextPatternTerm(valueForAnnotation))));
        } else if (p instanceof HitPropertyDoc || p instanceof HitPropertyDocumentId) {
            Object value = vals.get(i).value();
            int luceneDocId = value instanceof Doc ? ((Doc) value).id() : BlsUtils.getDocIdFromPid(blIndex(), (String) value);
            fqb.add(new SingleDocIdFilter(luceneDocId), Occur.FILTER);
            usedFilter = true;
        } else if (p instanceof HitPropertyDocumentStoredField) {
            fqb.add(new DocValuesTermsQuery(((HitPropertyDocumentStoredField) p).fieldName(), (String) vals.get(i).value()), Occur.FILTER);
            usedFilter = true;
        } else {
            logger.debug("Cannot merge group specifier into query: {} with value {}", p, vals.get(i));
            return null;
        }
        ++i;
    }
    // All specifiers merged!
    // Construct the query that will get us our hits.
    SearchEmpty search = blIndex().search(blIndex().mainAnnotatedField(), searchParam.getUseCache(), searchLogger);
    QueryInfo queryInfo = QueryInfo.create(blIndex(), blIndex().mainAnnotatedField());
    BLSpanQuery query = usedFilter ? tp.toQuery(queryInfo, fqb.build()) : tp.toQuery(queryInfo);
    SearchHits hits = search.find(query, searchParam.getSearchSettings());
    if (searchParam.hitsSortSettings() != null) {
        hits = hits.sort(searchParam.hitsSortSettings().sortBy());
    }
    if (searchParam.getSampleSettings() != null) {
        hits = hits.sample(searchParam.getSampleSettings());
    }
    return hits;
}

Example 3 with SearchHits

use of nl.inl.blacklab.searches.SearchHits in project BlackLab by INL.

the class QueryTool method parseAndExecuteQuery.

/**
 * Parse and execute a query in the current query format.
 *
 * @param query the query
 */
private void parseAndExecuteQuery(String query) {
    Timer t = new Timer();
    try {
        // See if we want to choose any random words
        if (query.contains("@@")) {
            StringBuffer resultString = new StringBuffer();
            Pattern regex = Pattern.compile("@@[A-Za-z0-9_\\-]+");
            Matcher regexMatcher = regex.matcher(query);
            while (regexMatcher.find()) {
                // You can vary the replacement text for each match on-the-fly
                String wordListName = regexMatcher.group().substring(2);
                List<String> list = wordLists.get(wordListName);
                if (list == null) {
                    errprintln("Word list '" + wordListName + "' not found!");
                    return;
                }
                int randomIndex = (int) (Math.random() * list.size());
                regexMatcher.appendReplacement(resultString, list.get(randomIndex));
            }
            regexMatcher.appendTail(resultString);
            query = resultString.toString();
        }
        Parser parser = parsers.get(currentParserIndex);
        TextPattern pattern = parser.parse(query);
        if (pattern == null) {
            errprintln("No query to execute.");
            return;
        }
        // pattern = pattern.rewrite();
        if (verbose)
            outprintln("TextPattern: " + pattern.toString());
        // If the query included filter clauses, use those. Otherwise use the global filter, if any.
        Query filterForThisQuery = parser.getIncludedFilterQuery();
        if (filterForThisQuery == null)
            filterForThisQuery = filterQuery;
        Query filter = filterForThisQuery == null ? null : filterForThisQuery;
        // Execute search
        BLSpanQuery spanQuery = pattern.toQuery(QueryInfo.create(index, contentsField), filter);
        if (verbose)
            outprintln("SpanQuery: " + spanQuery.toString(contentsField.name()));
        SearchHits search = index.search().find(spanQuery);
        hits = search.execute();
        docs = null;
        groups = null;
        sortedHits = null;
        collocations = null;
        showWhichGroup = -1;
        showSetting = ShowSetting.HITS;
        firstResult = 0;
        showResultsPage();
        reportTime(t.elapsed());
        if (determineTotalNumberOfHits)
            statInfo = Integer.toString(hits.size());
        else
            statInfo = "?";
        commandWasQuery = true;
    } catch (InvalidQuery e) {
        // Parse error
        errprintln(e.getMessage());
        errprintln("(Type 'help' for examples or see accompanying documents)");
    } catch (UnsupportedOperationException e) {
        // Unimplemented part of query language used
        // DEBUG createWeight bug
        e.printStackTrace();
        errprintln("Cannot execute query; " + e.getMessage());
        errprintln("(Type 'help' for examples or see accompanying documents)");
    }
}

Example 4 with SearchHits

use of nl.inl.blacklab.searches.SearchHits in project BlackLab by INL.

the class RequestHandlerHits method getHitsFromGroup.

private Pair<BlsCacheEntry<?>, Hits> getHitsFromGroup(String groupBy, String viewGroup) throws InterruptedException, ExecutionException, InvalidQuery, BlsException {
    PropertyValue viewGroupVal = PropertyValue.deserialize(blIndex(), blIndex().mainAnnotatedField(), viewGroup);
    if (viewGroupVal == null)
        throw new BadRequest("ERROR_IN_GROUP_VALUE", "Cannot deserialize group value: " + viewGroup);
    BlsCacheEntry<HitGroups> jobHitGroups = (BlsCacheEntry<HitGroups>) searchParam.hitsGroupedStats().executeAsync();
    HitGroups hitGroups = jobHitGroups.get();
    HitGroup group = hitGroups.get(viewGroupVal);
    if (group == null)
        throw new BadRequest("GROUP_NOT_FOUND", "Group not found: " + viewGroup);
    Hits hits = null;
    // Only launch a separate search when there are ZERO hits stored in the group
    if (group.storedResults().size() > 0) {
        // Some hits available: return those.
        hits = group.storedResults();
    }
    // No results were actually stored. Fire a separate query to retrieve them.
    if (group.storedResults().size() == 0) {
        HitProperty groupByProp = HitProperty.deserialize(blIndex(), blIndex().mainAnnotatedField(), groupBy);
        SearchHits findHitsFromOnlyRequestedGroup = getQueryForHitsInSpecificGroupOnly(viewGroupVal, groupByProp, hitGroups);
        if (findHitsFromOnlyRequestedGroup != null) {
            // place the group-contents query in the cache and return the results.
            BlsCacheEntry<ResultCount> cacheEntry = (BlsCacheEntry<ResultCount>) findHitsFromOnlyRequestedGroup.count().executeAsync();
            hits = ((BlsCacheEntry<Hits>) findHitsFromOnlyRequestedGroup.executeAsync()).get();
            return Pair.of(cacheEntry, hits);
        }
        // This is a special case:
        // Since the group we got from the cached results didn't contain the hits, we need to get the hits from their original query
        // and then group them here (using a different code path, since the normal code path  doesn't always store the hits due to performance).
        // And, since retrieving just the hits for one group couldn't be done (findHitsFromOnlyRequestedGroup == null), we need to unfortunately get all hits.
        SearchHitGroupsFromHits searchGroups = (SearchHitGroupsFromHits) searchParam.hitsSample().groupWithStoredHits(groupByProp, Results.NO_LIMIT);
        // now run the separate grouping search, making sure not to actually store the hits.
        // Sorting of the resultant groups is not applied, but is also not required because the groups aren't shown, only their contents.
        // If a later query requests the groups in a sorted order, the cache will ensure these results become the input to that query anyway, so worst case we just deferred the work.
        // place groups with hits in search cache
        jobHitGroups = (BlsCacheEntry<HitGroups>) searchGroups.executeAsync();
        hits = jobHitGroups.get().get(// get group
        viewGroupVal).storedResults();
    }
    // NOTE: sortBy is automatically applied to regular results, but not to results within groups
    // See ResultsGrouper::init (uses hits.getByOriginalOrder(i)) and DocResults::constructor
    // Also see SearchParams (hitsSortSettings, docSortSettings, hitGroupsSortSettings, docGroupsSortSettings)
    // There is probably no reason why we can't just sort/use the sort of the input results, but we need some more testing to see if everything is correct if we change this
    String sortBy = searchParam.getString("sort");
    HitProperty sortProp = (sortBy != null && !sortBy.isEmpty()) ? HitProperty.deserialize(hits, sortBy) : null;
    if (sortProp != null)
        hits = hits.sort(sortProp);
    return Pair.of(jobHitGroups, hits);
}