In serializable mode, the transaction will fail if the data it's relying on
is changed out from underneath it.
Read Committed will let you see the table at the state it's in when you
first try to access it in your transaction. Serializable will give you the
state of the table when your transaction starts.
On 7/29/06, Kaloyan Iliev <news1(at)faith(dot)digsys(dot)bg> wrote:
> Thank you for your response.
> My questions about going into Read Commited is because in Serializable
> i wouldn't see the changes on the counter, made by other transactions, if
> I am correct?
> And if one transaction increase the counter (If I am using table locking
> in Serializable),
> what will the next one (in Serializable) do - try to take the same value
> or will rollback?
> Thanks again,
> Kaloyan Iliev
> Andrew Hammond wrote:
> Kaloyan Iliev wrote:
> Hello All,
> I have such a question.
> I want to receive from the database subsequent numbers and I can't
> afford to miss one. There must not be any missing numbers.
> So the sequence is not good for me because if transaction rollback the
> there will be gaps.
> So I make a table with one row and the row contains one int.
> Every time I update the row in Serializable transaction level:
> update foo set lastvalue = lastvalue+1;
> select lastvalue from foo;
> This is my decision of the problem. But here is my next question.
> If two apllications try to take next number at the same time one of both
> transactions will abort.
> The one way is to catch the error and try again, but this is what I
> don't want to do.
> So is there a way to escape transaction error. I read about the locks
> and I think they can solve my problem.
> First I thick I must change my transaction Isolation Level to Read Commited.
> Then If I first lock (in ROW EXCLUSIVE mode) the table, then update and
> then read - will this solve my problem.
> And if two functions try to do this in the same time will the second
> transaction waith until it can lock the table and then without errors to
> take the next number?
> And my questions:
> 1. Should I change the transaction isolation level to Read Commited or
> Serializable transaction level is good enough (I prefer to work in
> Serializable transaction level)?
> Going to Read Committed from Serializable would actually decrease the
> level of isolation for your transaction. I'm not sure that's what you
> want to do.
> 2. Is my algorithm correct and will it give me secure way to get
> subsequent numbers without gaps?
> 3. Can I use SELECT FOR UPDATE instead ot locks in this case?
> Well, if the table has only one row, and that row is only for the
> counter, you could use
> LOCK counter_tbl IN ACCESS EXCLUSIVE MODE;
> UPDATE counter_tbl SET counter = counter + 1;
> SELECT counter FROM counter_tbl;
> Which would cause your transactions to queue up when dealing with the
> counter. You don't get rollbacks that way. :)
> 4. Can I change the transaction level back to Serializable after I get
> the number I want, without commiting the transaction?
> ---------------------------(end of broadcast)---------------------------
> TIP 3: Have you checked our extensive FAQ?
In response to
pgsql-novice by date
|Next:||From: julian howard||Date: 2006-07-29 18:32:24|
|Subject: pgAdmin and XML|
|Previous:||From: Kaloyan Iliev||Date: 2006-07-29 17:51:35|
|Subject: Re: Tables Locks Quetion or Strictlly subsequent numbers|