Re: [Sks-devel] Recon stalls for one hour

[Pete Stephenson]

On 9/5/2014 8:22 AM, Kim Minh Kaplan wrote:
> Pete Stephenson wrote:
> 
>>> I was perusing my recon.log and db.log files today and noticed odd gaps
>>> in gossip activity that last for an hour and result in error messages.
>>> Logs for the last few days indicate this happens 10-15 times per day.
> 
> Maybe this is one of your peers misbehaving. You should raise the
> recon debuglevel to see the IP of the connected peer. Debuglevel 4
> should do the trick.

Thanks. That did help track down the apparent source: a non-SKS
(hockeypuck) server that I was peering with is missing about 16k keys
and this seemed to cause problems.

As a temporary basis I've de-peered them and there's been no timeouts
since then with any of the remaining (all-SKS) servers.

>> Brief addendum: I was looking through the man page and it looks like the
>> "reconciliation_timeout" option may do the trick. The description says
>> it sets the "Timeout for reconciliation runs in minutes."
>>
>> If I were to add "reconciliation_timeout: 2" to my sksconf file, would
>> that cause the recon process to timeout after 2 minutes rather than
>> (seemingly the default) 60 minutes, or am I misinterpreting that option?
> 
> I think you are right. But 2 minutes sounds much too short. If I
> understand correctly the reconciliation process, it is *very* CPU
> intensive when the peers are too much out of sync and could require
> more time.

Interesting. Is there a recommended timeout period? 10 minutes? 20?
Something else?

Cheers!
-Pete