Continuous Executions with empty result sets

[markaddleman]

Attached is a test case (three cheers for embedded Teiid!) that illustrates a continuous execution which returns an empty result set.  The problem is that the execution will live forever because the callback is never invoked. 

FWIW, I believe that the responsibility to cancel a CE lies with the client rather than translators since since, individually, the translators have no way of knowing if the overall query has been satisfied.  I think StatementCallback needs to be enhanced with some new method which is called when the planner re-executes the plan.  Personally, I think the signature should look something like:

public void onStart(ResultSetMetadata metadata)

[markaddleman]

Duh, make that

public void onStart(Statement s, ResultSetMetadata metadata)

[shawkins]

Yes, no results would be a soft spot in the continuous logic.  As for an onStart or similar method, I've resisted adding one since that would effectively be a push of information to the client, which doesn't fix with the current request response model and so far the continuous logic has been fit in without many additional considerations in the engine.  Can you log something for this?

[markaddleman]

Done:  TEIID-2257

As a workaround, I was thinking of changing our query generator to produce queries of the form

SELECT a,b,c FROM t

UNION ALL

SELECT null, null, null

With the appropriate null ordering (I'll play around to find that), the row of nulls can be used as sentinel for the beginning or end of the result set.  None of our real data sources would produce a completely null row so I feel pretty safe using this as a temporary work around.  Do you see any problem with this approach?

[shawkins]

You could use "order by a nulls [first/last], ..." to control null ordering.