> i dont think, that you are right.
> are you really sure that you do not mean isolationlevel 
> 'serializable'.
> 

Uhm, err, yes - it is only such strict for 'serializable'. I messed it up :-)

> by the way:
> i got an answer some times ago from Elke Zabach:  (perhaps 
> its still up to date?)
> 
>  "Usually they are not locked. But there exist kernel-versions 
>  in which a row is selected, but updated by another transaction.
>  It then has to wait for the release of that 
> update/insert/delete-lock, i.e.
>  has to wait for the commit/rollback of the other transaction.
>  Then it will have a lock afterwards."
> ...

Yes, that is still valid. If you have a row locked in exclusive mode by
some session (row_exclusive), and a select is waiting to read it, it will 
get that lock and keep it, e.g. consider the example

Table ABC with A as integer and B as VARCHAR(20):

| A              | B                    |
| -------------- | -------------------- |
|              1 | AAA                  |
|              2 | AAA                  |
|              3 | AAA                  |

SESSION 1:
UPDATE ABC SET B = 'BBB' WHERE A = 2

SESSION 2:
SELECT * FROM ABC

Session 2 hangs until Session 1 has committed or rolled back, and has afterwards 
a row_share lock on the affected row number 2.

Anyway, I fear there might be something with the python script you use for getting the 
content
of LOCKSTATISTICS, as its conten looks a little bit in-plausible ...

Regards
Alexander Schr�der
SAP DB, SAP Labs Berlin

--
MaxDB Discussion Mailing List
For list archives: http://lists.mysql.com/maxdb
To unsubscribe:    http://lists.mysql.com/[EMAIL PROTECTED]

Reply via email to