Proper Tear down of Objects (Rust)

[cerdelen]

Hello and thanks in advance to anyone helping me out.

I am using iceoryx2 in my project and generally work quite well with it. I am using the examples showcased in the example folder and build on top of this foundation a more abstract api to send and receive in messages in a pub sub pattern, as well as in a req res pattern. I use Events alongside these Publisher/Subscriber or Server/Client objects to be able to wake up threads to avoid busy polling the objects.

I also build a moderatly medium sized test suite for my use case (around 80-120 tests). When running these tests i realised that if i open and close many subscribers/publishers or servers/clients as well as their companying Events in quick succession my tests run quite flakey. Quite consistently random singular tests fail because the creation of the these Objects fail with errors such as (ServiceInCorruptedState, IsMarkedForDestruction).

The Tests each have unique ServiceNames they are using. So there is no clash in namings. Some of the tests are opening and closing these Objects within the test itself multiple times a Service/Objects using this services (Subscribers/Publishers/Listeners....). Is there maybe something i have to take into account when tearing down objects? In the examples Objects are just dropped when they move out of scope and it is insinuated that i dont have to deregister or destroy/close anything manually. Is this the case or can someone point me to proper tear down procedures?

[elBoberido]

@cerdelen which version of iceoryx2 are you using?

[cerdelen]

0.7.0

[elBoberido]

Are you creating services with the same name in these tests?

Edit: obviously my reading comprehension is bad. You already wrote that they are unique 😅

[elBoberido]

I guess that our dead node detection is interfering with the startup. We might need to tweak it a bit further.

Meanwhile, there are two workarounds

• use the local::Service service variant. This keeps the services contained in the process
• if you require inter-process tests, another workaround would be to pass a custom config object to the node builder and set the root path for each test to a different location

If you can share the code that causes the issues or create a minimal testcase to trigger the behavior, we would have a chance to properly fix the issue

[orecham]

Just a stab in the dark, but you could try using isolating each test into their own environment. We do the same in our tests to prevent any risk of interference, see:

1. https://github.com/eclipse-iceoryx/iceoryx2/blob/main/iceoryx2/conformance-tests/src/service_publish_subscribe.rs#L70
2. https://github.com/eclipse-iceoryx/iceoryx2/blob/main/iceoryx2/src/testing.rs#L34

This uses the domain functionality of iceoryx2 as shown in this example: https://github.com/eclipse-iceoryx/iceoryx2/tree/main/examples/rust/domains

[cerdelen]

With the local version i dont run into problems apparently and having small sleeps inbetween tear downs and new instantiations seem to completely eliminate the Problem.

The domain possibility wouldnt work for me/us since we want our api to be as open and non restrictive as possible.

Thanks for the hint with local.