FTS Search

FTS3 and FTS4 are SQLite virtual table modules that allows users to perform full-text searches on a set of documents.

The basic idea is to have a virtual FTS table for each register table.

The virtual table holds the values that are used for searching and sorting as well as an identifier to link the register table.

Prerequisite

Always consider explain query plan to determine whether your queries will be slow.

For Example:

explain query plan select * from household where FWHOHFNAME LIKE '%abc%'

Returns

Full table scans are really slow, especially when querying a large data set. Consider optimizing such queries.

Adopting FTS search

1. *Application.java →  Add the register table names

   private String[] getFtsTables(){
       String[] ftsTables = { "table1" ... };
       return ftsTables;
   }

2. Add search and sort fields

   private String[] getFtsSearchFields(String tableName){
       if(tableName.equals("table1")){
           String[] ftsSearchFields =  { "SearchField1" ... };
           return ftsSearchFields;
       } ...
   }
   
   private String[] getFtsSortFields(String tableName){
       if(tableName.equals("table1")) {
           String[] sortFields = {"SortField1" ... };
           return sortFields;
       } ...
       return null;
   }

3. Add custom relational Ids if they exist. These are foreign keys apart from relationalId or relational_id fields that store a unique identifier for the related table e.g. childId or parentId

   private String getFtsCustomRelationalId(String tableName){
       if(tableName.equals("table1")){
           String customRelationalId = "parentId1";
           return customRelationalId;
       } ...
       return null;
   }

4. Add condition fields if the query has a main condition

   For Example:

   queryBUilder.mainCondition(" Field1 is not null ");

   private String[] getFtsMainConditions(String tableName){
       if(tableName.equals("table1")) {
           String[] mainConditions = {"Field1"};
           return mainConditions;
       } ...
   }

5. In *Application.onCreate(), update the global context instance i.e. update context.commonFtsObject

   context.updateCommonFtsObject(createCommonFtsObject());

   private CommonFtsObject createCommonFtsObject(){
       CommonFtsObject commonFtsObject = new CommonFtsObject(getFtsTables());
       for(String ftsTable: commonFtsObject.getTables()){
           commonFtsObject.updateSearchFields(ftsTable, getFtsSearchFields(ftsTable));
           commonFtsObject.updateSortFields(ftsTable, getFtsSortFields(ftsTable));
           commonFtsObject.updateMainConditions(ftsTable, getFtsMainConditions(ftsTable));
       }
       return commonFtsObject;
   }

6. In org.ei.opensrp.commonregistry.CommonRepository#onCreate, id and relationalid indexes are created for each common repository table.

   if(StringUtils.isNotBlank(common_ID_INDEX_SQL)) {
       database.execSQL(common_ID_INDEX_SQL);
   }
   if(StringUtils.isNotBlank(common_Relational_ID_INDEX_SQL)) {
       database.execSQL(common_Relational_ID_INDEX_SQL);
   }

7. In org.ei.opensrp.Context.java, commonFtsObject is passed as an argument when initializing the Repository class.

   if(commonFtsObject != null){
       repository = new Repository(this.applicationContext, session(), this.commonFtsObject, drishtireposotoryarray);
   }

   Then the virtual tables are created in org.ei.opensrp.repository.Repository#onCreate just after the database has been created.

   if(this.commonFtsObject != null) {
       for (String ftsTable: commonFtsObject.getTables()) {
           Set<String> searchColumns = new LinkedHashSet<String>();
           searchColumns.add(CommonFtsObject.idColumn);
           searchColumns.add(CommonFtsObject.relationalIdColumn);
           searchColumns.add(CommonFtsObject.phraseColumnName);
   
           String[] mainConditions = this.commonFtsObject.getMainConditions(ftsTable);
           if(mainConditions != null)
               for (String mainCondition : mainConditions) {
                   searchColumns.add(mainCondition);
               }
   
           String[] sortFields = this.commonFtsObject.getSortFields(ftsTable);
           if(sortFields != null)
               for (String sortValue : sortFields) {
                   searchColumns.add(sortValue);
               }
   
           String joinedSearchColumns = StringUtils.join(searchColumns, ",");
   
           String searchSql = "create virtual table " + CommonFtsObject.searchTableName(ftsTable) + " using fts4 (" + joinedSearchColumns + ");";
           database.execSQL(searchSql);
   
       }
   }

8. These virtual tables are updated after a form submission is saved.

   For Example, in FormSubmissionService.java:

   ziggyService.saveForm(getParams(submission), submission.instance());
   
   // Update FTS Tables
   updateFTSsearch(submission);

9. Update filterandSortExecute() and CountExecute() methods to use FTS search

   1. Perform search/sort query in the virtual table(s) and return the reference ids

      Virtual table scans are optimized and therefore faster than normal full table scans.

      Explain query plan

      explain query plan select object_id from household_search where phrase match 'abc*'

      Returns
      

      selectid	order	from	detail
      0	0	0	SCAN TABLE household_search VIRTUAL TABLE INDEX 4:

      
      

   2. Use the ids returned in Step 7(a) above to filter the main query

      Since id fields are indexed, Step 5, full table scans are replaced with search table using index.

      Explain query plan

      explain query plan select * from household where id in (select object_id from household_search where phrase match 'abc*')

      Returns
      

      selectid	order	from	detail
      0	0	0	SEARCH TABLE household USING INDEX sqlite_autoindex_household_1 (id=?)
      0	0	0	EXECUTE LIST SUBQUERY 1
      0	0	0	SCAN TABLE household_search VIRTUAL TABLE INDEX 4:

   
   Step 8 a) and b) in code
   

   Using FTS

   // Get the virtual table query
   String sql = sqb.searchQueryFts(tablename, joinTable, mainCondition, filters, Sortqueries, currentlimit, currentoffset);
   
   // Get the reference ids
   List<String> ids = commonRepository.findSearchIds(sql);
   
   // Update the main query by adding the reference ids in the where clause
   currentquery = sqb.toStringFts(ids, tablename + "." + CommonRepository.ID_COLUMN, Sortqueries);
   query = sqb.Endquery(currentquery);

FTS Table Structure

PRAGMA table_info('household_search');

Returns

dflt_value

object_id → reference id i.e. household.id

object_relational_id → reference relational id i.e. household.relationalid. Useful when query has a join between two register tables.

phrase → search term, contains all the searchable values separated by  ' | ' e.g. " searchFieldValue1 | searchFieldValue2 | searchFieldValue3 "

sort field → sort value

Facts

1. Queries are faster when indexes are used.
2. Indexes are ignored when SQLite's LIKE-keyword based query is used.
3. Virtual tables require more disk space compared to regular tables.
4. It takes more time to insert a record in a virtual table.
5. Always consider explain query plan to determine whether your queries aren’t going to work slowly.
6. FTS does not support suffix search, only prefix.

Links

https://www.sqlite.org/fts3.html

https://lab.getbase.com/killer-search-on-android-part-1

http://stackoverflow.com/questions/29815248/full-text-search-example-in-android