On 07/01/2014 12:19 PM, Martin Kosek wrote:
> On 06/26/2014 10:44 AM, Jan Cholasta wrote:
>> On 26.6.2014 10:39, Petr Viktorin wrote:
>>> On 06/26/2014 10:33 AM, Jan Cholasta wrote:
>>>> On 26.6.2014 09:40, Petr Viktorin wrote:
>>>>> On 06/26/2014 09:33 AM, Jan Cholasta wrote:
>>>>>> On 26.6.2014 09:21, Petr Viktorin wrote:
>>>>>>> On 06/26/2014 08:30 AM, Jan Cholasta wrote:
>>>>>>>> On 25.6.2014 18:25, Petr Viktorin wrote:
>>>>>>>>> On 06/25/2014 05:29 PM, Jan Cholasta wrote:
>>>>>>>>>> Hi,
>>>>>>>>>> On 25.6.2014 17:17, Tomas Babej wrote:
>>>>>>>>>>> Hi,
>>>>>>>>>>> Our datetime conversion does not support full LDAP Generalized
>>>>>>>>>>> time syntax. In the unsupported cases, we should fall back
>>>>>>>>>>> to string representation of the attribute.
>>>>>>>>>>> In particular, '0' is used to denote no value of LDAP generalized
>>>>>>>>>>> time attribute.
>>>>>>>>>>> https://fedorahosted.org/freeipa/ticket/4350
>>>>>>>>>> NACK, this beats the purpose of decoding of the values, because it
>>>>>>>>>> requires you to check the type of the value before using it.
>>>>>>>>>> Instead, you should either fix the code that uses the
>>>>>>>>>> nsds5ReplicaLastUpdate{Start,End} attributes to access their raw
>>>>>>>>>> value
>>>>>>>>>> directly, or exclude the attributes from decoding to datetime by
>>>>>>>>>> overriding their type in IPASimpleLDAPObject._SYNTAX_OVERRIDE.
>>>>>>>>>> Honza
>>> [...]
>>>>> The problem is that "unset" is a valid value here,
>>>> It is not, according to Generalized Time syntax (RFC 4517 section
>>>> 3.3.13) "0" is an invalid value and we should treat it the same way as
>>>> any other invalid value (hence my original suggestion).
>>> OK, in that case ignore what I said here.
>>> So am I correct that 389-ds storing a value that doesn't comply with the
>>> attribute's syntax?  Should we file a DS bug?
>> AFAIK syntax checks are done only on LDAP adds and mods, so unless we are
>> setting "0" somewhere and DS does not complain, it is not a bug.
> What is the final resolution here? Do we plan to ignore the attribute
> conversion for these attributes as, fixing DS or are we fixing the framework?
> Surprisingly, I did not see the filed failure any more in my
> ipa-replica-install tests nor in the latest CI run, I wonder if anything
> changed in our code or if the issue is intermittent. Anyway, I am willing to
> close this ticket if this is no longer a problem.
> Martin
> _______________________________________________
> Freeipa-devel mailing list
> Freeipa-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/freeipa-devel

Issue is not 100% reproducible in my testing. The underlying problem
here is that sometimes the values of nsds5replicalastupdatestart and
nsds5replicalastupdateend attributes are not being set during
replication (intermittent issue) and thus 389 falls back and returns '0'
if any of these attributes were explicitily requested, which is not a
valid value for LDAP Generalized time.

The reason you're not hitting the conversion errors is probably that
during replication nsds5replicalastupdatestart and
nsds5replicalastupdateend were assigned proper values. You can check
that in the replication log (search for "Replication Update in progress:
%s: status: %s: start: %d: end: %d" in the logs).

I am checking with Ludvig, but if this behaviour is specific to the
nsds5replicalastupdatestart and nsds5replicalastupdateend I'd go with
the SYNTAX_OVERRIDE solution.

However, it is worth noting that these conversion errors are harmless

Tomas Babej
Associate Software Engineer | Red Hat | Identity Management
RHCE | Brno Site | IRC: tbabej | freeipa.org 

Freeipa-devel mailing list

Reply via email to