On 12/4/25 11:23 AM, Dumitru Ceara wrote:
> On 12/4/25 10:35 AM, Eelco Chaudron via dev wrote:
>>
>>
>> On 3 Dec 2025, at 12:01, Ilya Maximets wrote:
>>
>>> The message about conflicting rows is trying to order the two rows
>>> in a consistent manner, so the log messages do not change in tests.
>>> But it fails to do so, because the order of columns in the column
>>> set depends on the order of columns inside the hash map, which
>>> depends on the hash function and the internal implementation details
>>> of the hash map.  This results in random test failures, when two
>>> rows end up in the opposite order.
>>>
>>> Uncovered while testing a different hash map implementation, but the
>>> failure is technically possible even without any changes in the code,
>>> e.g., by running on a different CPU architecture or with different
>>> compiler flags.
>>>
>>> Fix that by introducing a new function that constructs the column
>>> set with columns in a predictable order and without UUID columns that
>>> have random values in most cases and so not actually comparable.
>>
>> Thanks, Ilya.
> 
> Hi Ilya, Eelco,
> 
>>
>> The patch looks good to me. I reviewed the code, and OVSDB_TYPE_UUID is not 
>> used for indexing.
>>
>> Acked-by: Eelco Chaudron [email protected]
>>
> 
> Same here, the change looks good to me too:
> 
> Acked-by: Dumitru Ceara <[email protected]>

Thanks, Eelco and Dumitru!
Applied and backported down to 3.3.

Best regards, Ilya Maximets.
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to