Re: Final findings: hung jobs

Hello Phil,

Thanks for clarifying the case in which the problem occurs.

Nothing significant was modifed in the catalog code between 9.6.3 and 9.6.5.

If you are building the Director without SMARTALLOC enabled in the 
./configure, then I can imagine that you would have a problem in 9.6.5.  
I do have a patch to fix that, but the simplest for you would be to 
ensure that smartalloc is enabled during your Director build.

Otherwise, I have no ideas what can be broken.

Best regards,

Kern

On 8/14/20 3:25 PM, Phil Stracchino wrote:
> On 2020-08-14 08:49, Kern Sibbald wrote:
>> Hello Phil,
>>
>>   From what I hear from you, the bug involves several different Bacula
>> versions running on different machines, and never happens when
>> everything is running version 9.6.5.
>
> No, that's not correct.  I have a heterogenous environment with clients
> running on three different OS distributions (Fedora Linux, Gentoo Linux,
> Solaris 11.3) and have tried multiple combinations of versions, and I
> have found only a single determinant factor:
>
> If the Director is 9.6.5, jobs intermittently hang.  If it is 9.6.3,
> they don't.
>
> This is completely unaffected by the platform or version of the clients.
>   Either all clients, 9.6.3 or 9.6.5, get randomly hung jobs, or none of
> them, determined *solely* by whether the Director is 9.6.5 or 9.6.3.
>
> Right now the Director is 9.6.3, *everything* else is 9.6.5, and I
> haven't had a hung job in two weeks.
>
>
> My suspicion is that the bug lies somewhere in the MySQL driver code or
> in interactions with the MySQL driver, but it's just a suspicion.  I
> suspect that driver could use an overhaul.
>
>
>


_______________________________________________
Bacula-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/bacula-devel