[Info-vax] RDB Question

[Arne Vajhøj]

On 12/4/2018 11:43 AM, Dave Froble wrote:
> On 12/4/2018 10:16 AM, Jan-Erik Söderholm wrote:
>> Note also that, even if Rdb supports record locking (has done from
>> V1.0, if I'm not wrong), it also has something called "Adjustable
>> record locking". So Rdb doesn't move locks down to the record level
>> if it doesn't have to (row contention between users).
>>
>> The default is what is shown from this database dump below.
>> So one single process doing a large query on its own will put
>> locks at blocks of 1000 database pages at a time (to start with).
> 
> Jan-Erik, I know you didn't design and implement Rdb, so I'm not 
> directing this at you.
> 
> I have to ask, why would Rdb be taking out locks for a query (if I'm 
> understanding the definition of query)?  If one is going to read, but 
> never write, then why lock?  While there just may be some reason to not 
> read locked data, that is not normally the case.  There is great 
> efficiency in skipping the locking when it is not needed.

RDBMS'es has a very specific definition of transaction isolation
level.

As an example with REPEATABLE READ and SERIALIZEABLE then:

BEGIN TRANSACTION
read some data
do some stuff
read same data again
COMMIT or ROLLBACK transaction

then the two reads has to return the same data.

Arne