aholyoke / traildb-java Goto Github PK

Currently, TrailDBCursor.next creates a new TrailDBEvent and sets the fields of it with the internal values of tdb_event. TrailDBCursor.next will probably one of the hottest method call in codebase since we need to call it for each event.

Therefore, I believe that we should optimize it in a way that it should use minimum amount of memory and CPU. Specifically, I think we can provide two set of optimizations:

1. Add an extra TrailDBEvent parameter to TrailDBCursor.next and if it's not null, reuse that object instead of creating one. Avro is already using this method in order to minimize GC pressure: https://github.com/apache/avro/blob/17f2d75132021fafeca29edbdcade40df960fdc9/lang/java/avro/src/main/java/org/apache/avro/io/DatumReader.java#L35

2. Change the signature of TrailDBCursor.next to return void instead of TrailDBEvent and add getCurrentItem(), getCurrentTimesamp(), getCurrentNumberOfItems() to TrailDBCursor. We can store the reference of native tdb_event object inside TrailDBCursor and use the proxy methods instead of TrailDBEvent. The overhead of Java will be minimum in this case.

TrailDB tdb = new TrailDB("tiny.tdb");
TrailDBCursor cur = new TrailDBCursor(tdb);

cur.getTrail(trailId);
cur.getTrailLength(); // returns the correct value
cur.getTrailLength() // always returns 0

TrailDBCursor.getTrailLength should always return the same value for a specific trail.