Re: Amanda 3.5.3-1.fc38 aborts dump on EOF error to planner

Nuno,

     Thanks for the reply! And apologies for being not quite clear.

     I'm quite sure the offending hosts are powered down, so no chance 
of partial response. When I look at the planner.<timestamp>.debug log, I 
can see sendsize requests going out to the hosts that are powered up and 
responsive, and I can see their responses arrive. There are two hosts 
powered down, gallifrey and jpt. The requests go out to gallifrey, then 
jpt. When the request to gallifrey times out, planner sees 255 status 
from SSH and aborts with the EOF error. Doesn't even wait around for the 
timeout on jpt.

     If I go back and look at some old logs, I can see planner continue 
past the `EOF on read' error. So I'm really starting to think this is a 
new bug in 3.5.3.

     For what it's worth, I'd interpret your error,

ERROR Request to MACHINE failed: Connection refused

as the machine was powered up and responsive but actively refused the 
connection for some reason.

     I'm puzzled by another thing: we're using the same version of 
amanda (3.5.3) and I run backups to disk, no tape drive involved, but 
I've never seen the error you mention in April: backup aborts after 
first machine/disk in disklist. The obvious difference is Fedora 37 
versus Fedora 38, but really that shouldn't cause this much difference 
in behaviour.

     Bah! Sometimes staying up-to-date is a bit painful. I'll see if 
anyone else chimes in before I report this as a bug.

     Spent two weeks in Porto and the Douro Valley in Fall 2022. Loved 
the country!

         Lou


On 2023-07-19 2:17 a.m., Nuno Dias wrote:
>   Hi Lou,
> 
>   I'm using the same version as you, although in Fedora 37
> amanda-3.5.3-1.fc37.x86_64 and I don't see that behaviour, I have some
> machines that are down and the rest of the backups were made.
> 
>   In my case I have this
> 
>    planner: ERROR Request to MACHINE failed: Connection refused
> 
>   From what you wrote, it seems gallifrey.ivriel is not down is
> responding, but has some problems reporting the size.
> 
>   Maybe this page will help
> 
> https://www.zmanda.com/knowledge-base/eof-on-read-error-from-a-client/
> 
>   Although if is aborting all the planner it seems a bug, or there are
> other reasons for aborting all the planner, maybe checking if the
> etimeout  is not very low.
> 
> Cheers,
> Nuno
> 
> On Tue, 2023-07-18 at 13:50 -0700, Lou Hafer wrote:
>> Folks,
>>
>>       I've been using amanda for several years on a simple home
>> network.
>> Hosts are often powered down. Up through amanda 3.5.2, this worked
>> like
>> a charm. If the host didn't respond, it was simply skipped. Hosts
>> that
>> responded were properly backed up.
>>
>>       With amanda 3.5.3, the behaviour has changed. If a host doesn't
>> respond to the planner size request, the planner aborts the entire
>> backup with the error
>>
>>     planner: ERROR Request to gallifrey.ivriel failed:
>>     EOF on read from gallifrey.ivriel
>>
>> I've confirmed that my configuration is generally correct --- as long
>> as
>> all hosts in the disklist respond to the size request, the backup
>> succeeds.
>>
>> Is this a bug? Do I need to change some parameter in my configuration
>> to
>> persuade planner to soldier on? Any thoughts would be appreciated.
>>
>>       As context, this problem came about with an upgrade from Fedora
>> 37
>> to Fedora 38, with a matching upgrade from amanda 3.5.2 to amanda
>> 3.5.3.
>>
>>               Thanks,
>>                       Lou
>>
>