Re: how synchronous is Oracle?

From: Howard J. Rogers <howardjr2000_at_yahoo.com.au>
Date: Tue, 11 Mar 2003 05:27:28 +1100
Message-ID: <pan.2003.03.10.18.27.26.545417@yahoo.com.au>

On Mon, 10 Mar 2003 07:33:27 +0000, Uwe Kuechler wrote:

> Andrew Allen <andrew.allen_at_sppaammkiller.handleman.com> wrote in message news:<3E67A02E.4060205_at_sppaammkiller.handleman.com>...
>

>> You DO NOT know what you are talking about.

> Ouch. Maybe I should quit my job, then.
>

>> txn b will NOT wait for txn a to commit or rollback.

> This is not what I intended to say. I hope you're the only one who
> misunderstood me this way.
> What I meant is: The very moment Txn a issues a commit and the data
> gets written to disk, waits will occur when Txn b accesses this very
> data.

I hope you did NOT mean to say this, because as written, it's not true. A commit does not imply the data gets written to disk. A commit only causes the redo for transaction A to be written to the redo logs.

So what waits you are talking about transaction b having to suffer, I have no idea, I'm afraid. If you mean it has to wait because it's got to wait for the flush to disk to take place, not so (because the data block isn't flushed in the first place). It doesn't even have to wait for the redo to be flushed, because the definition of a commit means that it doesn't count as a commit until transaction A gets confirmation back from LGWR that the flush was successful. So only AFTER the redo flush has completed is the transaction considered committed.

Now maybe I've just misunderstood what you wrote, too. But then we need to do something about avoiding all these misunderstandings.

Regards
HJR Received on Mon Mar 10 2003 - 12:27:28 CST