Page Comparison

...

For Example:

Code Block

language	sql
title	Explain query plan

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

...

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

Adopting FTS search

*Application.java → Add the register table names

Code Block

language	java
theme	Eclipse

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

Add search and sort fields

Code Block

language	java
theme	Eclipse

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;
}

Add condition fields If your if the query has a main condition

For Example:

Code Block

language	java
theme	Eclipse

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

Code Block

language	java
theme	Eclipse

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

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

Code Block

language	java
theme	Eclipse

context.updateCommonFtsObject(createCommonFtsObject());

Code Block

language	java
theme	Eclipse

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;
}

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

Code Block

language	java
theme	Eclipse

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);
}

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

Code Block

language	java
theme	Eclipse

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.

Code Block

language	java
theme	Eclipse

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);

    }
}

These virtual tables are updated after saving a form submission is saved.
For Example, in FormSubmissionService.java:
Code Block
language java
theme Eclipse
ziggyService.saveForm(getParams(submission), submission.instance()); // Update FTS Tables updateFTSsearch(submission);

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

Perform search/sort query in the virtual table(s) and return the reference ids
Full Virtual table scans are optimized and therefore faster than normal full table scans.
Code Block
language sql
title 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:
Use the ids returned in Step 7(a) above to filter the main query
Since id fields are indexed, Step 5, there are no full table scans , only virtual table scansare replaced with search table using index.
Code Block
language sql
title 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

Code Block

language	java
theme	Eclipse
title	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);

...

Code Block

language	sql

PRAGMA table_info('household_search');

Returns

cid	name	type	not null	`dflt_value`	pk
0	object_id		0		0
1	object_relational_id		0		0
2	phrase		0		0
3	SortField1		0		0
4	SortField2		0		0
5	SortField3		0		0

object_id → reference id i.e. household.id

...

sort field → sort value

Facts

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

...

Versions Compared

Old Version 10

New Version 11

Key

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: