9.25. Anytime Queries

Starting with version 6, Virtuoso offers a partial query evaluation feature that guarantees answers to arbitrary queries within a fixed time. This is intended for use in publicly available SPARQL or SQL end points on large databases. This enforces a finite duration to all queries and will strive to return meaningful partial results. Thus this provides the same security as a transaction timeout but will be more user friendly since results will generally be returned, also for aggregate queries. Outside of a public query service, this may also be handy when exploring a large data set with unknown properties.

The feature is activated with the statement

set result_timeout == <expression>;

The expression should be a count of milliseconds. The scope of the setting is the connection, thus it remains in effect past the return of the calling procedure. Setting the timeout to 0 returns to the default state of having no limit on query run time and always returning complete results. The initial value of the timeout is 0. On a web server thread, the timeout is reset to 0 at the start of processing each new request header.

After a query or cursor has exceeded the running time, it is reset. In the event of the reset occurring inside an aggregation, the part of the query that produces rows for the aggregation is terminated, the timeout is reset and the query continues by using the aggregated results that were obtained before the first timeout. If there is again a timeout, the present aggregation is reset and the query gets another lease on life for processing the results accumulated so far. Thus, a query which has a select, a group by and an order by can consume at most three timeouts. One for producing the aggregated rows, another for sorting them and finally a third for returning them to a client or iterating over them in a stored procedure.

If the timeout occurs outside an aggregation, the result set is simply truncated.

The fact of a query returning partial results is indicated by the result set ending with a S1TAT SQL state. If the result set is read to end without this state being signalled, the results are complete and the query was not interrupted. If the result set is not consumed to end, the client cannot be sure of its completeness.

The error message associated with the S1TAT state contains the human readable version of the output of db_activity () for the query. If result_timeout is non-zero, the opening of a cursor or execution of a query on a client connection or with the exec () function will reset the client resource consumption counters automatically. This is done so as to have the resource consumption statistics for the error message scoped to the query. To get the exact counts, the application may call db_activity (1) in the same procedure. If this is called on a client server connection with the timeout on, the counts will be reset before db_activity () is called.

The resolution of the timeout is fairly low, timeouts are checked every two seconds with default settings.

A query cannot set a timeout from inside itself. Only a stored procedure or a top level statement on a client connection can do this.

A request to SPARQL web service endpoint may specify the timeout as an additional &timeout=<milliseconds> parameter. The parameter value should be equal to or greater than 1000, otherwise it will be ignored. If SPARQL Virtuoso configuration file contains MaxQueryExecutionTime parameter and its value is greater than or equal to 1000 then the actual "anytime" timeout is the minimum of the requested value and the value in the configuration file.

The current standard of SPARQL protocol does not provide any support for partial results. When a SPARQL query ends with S1TAT state, the returned HTTP header contains four additional rows, X-SQL-State: , X-SQL-Message: , X-Exec-Milliseconds: and X-Exec-DB-Activity: ; the returned document is formatted according to the requested MIME type as if the result is complete.