On 31 Oct 2024, at 17:42, Ilya Maximets wrote:

> On 10/31/24 12:46, Eelco Chaudron wrote:
>>
>>
>> On 30 Oct 2024, at 14:50, Ilya Maximets wrote:
>>
>>> Expand the NxN test with the network connectivity check between all the
>>> nodes.  Unfortunately, we can't really run this test with Libreswan 4.x,
>>> since, due to internal issues in these versions, we are getting into
>>> states where everything is loaded and active, but no traffic can pass.
>>> This is an internal issue in Libreswan that we can't workaround from
>>> the outside.  So, the fix is required in Libreswan itself.  4.5 and
>>> earlier versions seem to not be affected by this problem, at least not
>>> severely affected, but it's easier to just cut off all the 4.x versions
>>> from the test.
>>>
>>> 3.32 version from Ubuntu 22.04 and Libreswna 5.1 work just fine with
>>> this test.
>>>
>>> Test is relatively long, but it is very valuable, IMO.  Besides
>>> stressing ovs-monitor-ipsec with various failure and asynchronous
>>> connection establishment conditions, which are important for OVS, it
>>> also was used to reproduce and fix several bugs in Libreswan 4.x.
>>> Unfortunately, not all the issues are understood and fixed yet.
>>
>> I do not like the fact that now the inclusion or not for the traffic test
>> is not visible. I rather split up the test and get a “skipped” indication
>> for the traffic part.
>>
>> WDYT?
>
> Yeah, I'm not a big fan of this either.  We have a few options:
>
> 1. Copy the test and only test the configuration in one and
>    ping in the other.
>
> 2. Put the test into a macro and call it twice with and without
>    ping, so the one with ping can be skipped.
>
> 3. Add SKIP_IF() check in the middle of the test.
>
> The first option will cause some code duplication, which doesn't
> sound great.  The second one is OK, but it will obscure line numbers
> when something inside the macro fails, which is not very friendly
> while debugging the test.
>
> The third option will solve the code duplication and we'll also
> not run the configuration check twice saving the time.  We will
> still run the configuration part with Libreswan 4, even if the
> test will be skipped afterwards, so the test will fail if the
> configuration fails.  This is not very straightforward, but at
> least the test will actually run on all systems and it will be
> clear that it wasn't run in full on systems with libreswan 4,
> so there will be no ambiguity.
>
> WDYT?

Option 3 appears to be the most suitable choice for me.


> Best regards, Ilya Maximets.

_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to