Database auto-reconnection: 2 trials needed?

[Alexander Klimetschek <ak...@day.com>]

Hi Thomas,

while backporting the DB auto-reconnect issue I have lots of  
conflicts, so I have to fully understand the code to resolve it ;-).  
So here's my question:

In the ConnectionRecoveryManager there are various methods (eg.  
executeStmt) that have a trial-loop which tries 2 times before giving  
up. IIUC, this is for the case when a connection is broken and needs  
to be reestablished. I assume the first call to the connection results  
in an SQLException indicating the connection is broken and thus we  
need a second loop in which the connection is re-established.

Is this correct?

Are there other subtle things I have to be careful about?

Thanks,
Alex

--
Alexander Klimetschek
alexander.klimetschek@day.com

Re: Database auto-reconnection: 2 trials needed?

[Alexander Klimetschek <ak...@day.com>]

Am 25.02.2008 um 14:23 schrieb Thomas Mueller:

> I don't think it is really required to try twice. It is more like: OK
> we have a loop already, why don't we add one iteration for security...
> There is also an option to repeat forever (this is good for testing; I
> don't think it makes sense in most situations).

Ok, so at least one iteration is vital for the re-connect to work. But  
I will copy over the 2-trials code as exact as possible.

>
>
>> Are there other subtle things I have to be careful about?
>
> I don't think. However you need to test it. It is possible to write a
> unit test (using Derby server), unfortunately we don't have any so
> far.

That's exactly what I have done so far. It obviously fails with the  
old code, I hope after fully merging the test will run through.

Thanks!
Alex

--
Alexander Klimetschek
alexander.klimetschek@day.com

Re: Database auto-reconnection: 2 trials needed?

[Thomas Mueller <th...@gmail.com>]

Hi,

>  In the ConnectionRecoveryManager there are various methods (eg.
>  executeStmt) that have a trial-loop which tries 2 times before giving
>  up. IIUC, this is for the case when a connection is broken and needs
>  to be reestablished.

I don't think it is really required to try twice. It is more like: OK
we have a loop already, why don't we add one iteration for security...
There is also an option to repeat forever (this is good for testing; I
don't think it makes sense in most situations).

>  Are there other subtle things I have to be careful about?

I don't think. However you need to test it. It is possible to write a
unit test (using Derby server), unfortunately we don't have any so
far.

Regards,
Thomas