[389-users] replication failure - clock skew

[Rich Megginson]

On 02/10/2012 09:07 AM, Greg Kuchyt wrote:
> On 02/09/2012 03:03 PM, Rich Megginson wrote:
>> On 02/09/2012 12:52 PM, Greg Kuchyt wrote:
>>> On 02/09/2012 02:38 PM, Rich Megginson wrote:
>>>> On 02/09/2012 12:23 PM, Greg Kuchyt wrote:
>>>>> Yesterday afternoon, one of my consumers randomly crashed/rebooted.
>>>>> Upon rebooting, its replication agreement with its master failed with
>>>>> the following error:
>>>>>
>>>>> Unable to acquire replica: Excessive clock skew between the supplier
>>>>> and the consumer. Replication is aborting
>>>>>
>>>>> I did a little bit of Google searching and found some list traffic
>>>>> from a few years ago. From that I derived that this replica was hosed
>>>>> and I would need to re-initialize it. No problem. A re-initialization
>>>>> didn't do anything, same error. Starting from scratch from a
>>>>> completely new/fresh replica produces the same result. That's when I
>>>>> noticed the following errors in the logs on the master.
>>>>>
>>>>> csngen_new_csn - Warning: too much time skew (-115319 secs). Current
>>>>> seqnum=1
>>>>>
>>>>> I downloaded the readNsState.py script attached to the following
>>>>> ticket (https://bugzilla.redhat.com/show_bug.cgi?id=233642). Running
>>>>> this on the master produced the following output
>>>>>
>>>>> For replica cn=replica,cn=o\3Dpotsdam.edu,cn=mapping tree,cn=config
>>>>> len of nsstate is 40
>>>>> CSN generator state:
>>>>> Replica ID : 6560
>>>>> Sampled Time : 1328928777
>>>>> Time in hex : 0x4f35d809
>>>>> Time as str : Fri Feb 10 21:52:57 2012
>>>>> Local Offset : 0
>>>>> Remote Offset : 261
>>>>> Seq. num : 1
>>>>> System time : Thu Feb 9 14:00:01 2012
>>>>> Diff in sec. : -114776
>>>>>
>>>>> This leads me to believe that the clock skew problem is on the 
>>>>> master.
>>>>>
>>>>> I am not really sure how the clock skew happened. All of these 
>>>>> systems
>>>>> synchronize their clocks via a centralized time server and all the
>>>>> times on their clocks are correct. There are 3 or 4 other replicas
>>>>> that are still receiving incremental updates fine, but any attempt to
>>>>> add a new replica results in a failed replication agreement due to
>>>>> excessive clock skew.
>>>>>
>>>>> I am writing to get a better understanding of the situation and 
>>>>> see if
>>>>> there is anything to be done to resolve this. At the moment it seems
>>>>> as if I am caught in an unfortunate situation that will require
>>>>> re-initialization of my master from a back-up.
>>>>>
>>>>> Thanks for any help that can be provided.
>>>> What is your 389-ds-base version and platform?
>>>>> -- 
>>>>> 389 users mailing list
>>>>> 389-users at lists.fedoraproject.org
>>>>> https://admin.fedoraproject.org/mailman/listinfo/389-users
>>>>
>>>
>>> Rich,
>>>
>>> The master and a few replicas are running on Scientific Linux 6.1
>>> x86_64. Here, we're using the stock packages along with the modified
>>> 389-ds-base packages in your fedorapoeople.org repo. So that puts it
>>> at 1.2.9.9-1 for 389-ds-base I believe.
>>>
>>> Two replicas (including the one that rebooted/failed) are on Fedora 12
>>> x86_64 and their 389-ds-base is 1.2.5-1.
>> There was a known problem with clock skew calculation and handling in
>> 1.2.5 - please try upgrading everything to 1.2.9.9. I realize fedora 12
>> is no longer supported.
>>>
>>> Let me know what other info you need. Thanks.
>>> -- 
>>> 389 users mailing list
>>> 389-users at lists.fedoraproject.org
>>> https://admin.fedoraproject.org/mailman/listinfo/389-users
>>
>
> Rich,
> The F12 systems were in production and were slated for replacement by 
> SL 6.1 systems. I just took the F12 systems out of the mix rather than 
> upgrade them, so everything is now SL 6.1 and 389-ds-base 1.2.9.9.
>
> When attempting to add a new replica I still see the following in the 
> error logs on the master.
>
> "Unable to acquire replica: Excessive clock skew between the supplier 
> and the consumer. Replication is aborting."
>
> As well, I am seeing a lot of these messages in the logs on the master.
>
> "csngen_new_csn - Warning: too much time skew (-123525 secs). Current 
> seqnum=1"
Are your clocks on the servers all in sync?
> -- 
> 389 users mailing list
> 389-users at lists.fedoraproject.org
> https://admin.fedoraproject.org/mailman/listinfo/389-users