Thank you very much. I am now more clear.

Best regards,
Leon

> But since ‘CONNECT BY WITH FILTERING’ just starts once, how can this operation identify all the levels of data?

The execution is not atomic. During the execution this operation does several things… like getting the data of level “n” and starting the search of level “n+1″.
You can imagine that there is a kind of pipe between operation 1 (CONNECT BY WITH FILTERING) and operation 4 (CONNECT BY PUMP). So, every time operation 1 gets the data for level “n” it sends into it the ID to find the data for level “n+1″.

> And it seems that your answer aslo implies that operation for extracting data will start 4 times.

Also the Starts column of the dbms_xplan output confirms that the extraction of data is executed 4 times.

HTH
Chris

But since ‘CONNECT BY WITH FILTERING’ just starts once, how can this operation identify all the levels of data?

And it seems that your answer aslo implies that operation for extracting data will start 4 times. That’s why I asked whether it’s operation 4 passes n level data to level n+1 instead of operation 1.

Thank you very much.
Leon

> My question is which operation passes data from level 2 to operation 4?

Operation 1, CONNECT BY WITH FILTERING

> Can I understand like this,the first execution has generated the level 2 data from
> level 1 and CONNECT BY PUMP gets these data and then launches the subseque loop?

There are 4 executions:
- The first one extracts the data for level 1 and, therefore, provides the data to access level 2.
- The second execution extract the data for level 2 and provides the data to access level 3
- …

In general: execution “n” extracts data for level “n” and provides the data to access level “n+1″.

HTH
Chris

My question is which operation passes data from level 2 to operation 4?
Can I understand like this,the first execution has generated the level 2 data from level 1 and CONNECT BY PUMP gets these data and then launches the subseque loop?

Best regards,
Leon

Sorry, but I don’t know why this double check is needed. My best guess is the following: since it is an access predicate (not a filter predicate), this probably means that data is placed in a memory structure (kind-of hash table) and, to access it, the key is MGR.

Cheers,
Chris

[...]9. Operation 1 applies the access predicate “MGR”=PRIOR “EMPNO” and sends the 14 rows to the caller.[...]

Why do we need the access predicate in operation 1 ? I think the whole algorithm ensures that only “valid” rows get reported to operation 1. Why is it needed to apply the predicate once more before sending them to the caller?