Clarification on setTimeout and isTimeout Java API methods

[ppf2]

Consider the following example of intentionally setting the query timeout value in the Java API to a very small number:

  SearchResponse sres = client.prepareSearch("index_name")
                    .setTimeout(TimeValue.timeValueNanos(1))
                    .addAggregation(AggregationBuilders.terms("id").field("id").size(1000).shardSize(1000))
                    .execute()
                    .actionGet();
            System.out.println("Is query timed out? " + sres.isTimedOut());

The query returns in > 1 nanosecond, but the isTimedOut parameter is false. Would like to get a clarification on this. Should isTimeOut return true in this case? Or does it not return true because of the following enhancement request (#4586) to actually terminate queries that exceed the query timeout?

[clintongormley]

@markharwood could you shed some light on this?

[markharwood]

The existing timeout checks are based on Lucene's TimeLimitingCollector which has settings measured in milliseconds so the granularity of your timer (1 nanosecond) may well be the issue here - it probably doesn't make sense to expect timeout logic to behave with settings of sub-millisecond.

Generally speaking, the accuracy of timeout checks is not as good as it could be and #4586 is all about introducing more frequent checks for overruns than the one provided by Lucene's TimeLimitingCollector. This requires a careful balance of improving early termination of bad queries without slowing search speeds of good queries with additional timer checks.

There will be no change to the client interface - just improved detection of overruns.

[clintongormley]

Closing in favour of #4586