Keep reference to DomainParticipantFactory

[MiguelCompany]

We've seen several issues related with destruction order lately.

This PR aims to mitigate them by leveraging Fast DDS DomainParticipantFactory::get_shared_instance().

It should then avoid the automatic destruction of the factory (and the participants it created) until all the contexts in the process have been destroyed.

This can be backported to Jazzy and Iron, but not Humble, since get_shared_instance() is only available from Fast DDS 2.8.0 onwards.

[fujitatomoya]

@MiguelCompany thanks for the explanation, lgtm with green CI.

[ahcorde]

• Linux
• Linux-aarch64
• Linux-rhel
• Windows

[MiguelCompany]

@fujitatomoya @ahcorde I rebased this one to get the changes from #763 (I think the test failures were due to that)

[fujitatomoya]

• Linux
• Linux-aarch64
• Linux-rhel
• Windows

[ahcorde]

• Linux
• Linux-aarch64
• Linux-rhel
• Windows

[clalancette]

I'm going to run one more CI here, just to make sure everything is good, before merging.

• Linux
• Linux-aarch64
• Linux-rhel
• Windows
• Windows Debug

[clalancette]

Ah, CI isn't working because this needs to be rebased. @MiguelCompany can you do that?

[MiguelCompany]

Ah, CI isn't working because this needs to be rebased. @MiguelCompany can you do that?

Done!

[clalancette]

Here we go:

• Linux
• Linux-aarch64
• Linux-rhel
• Windows
• Windows Debug

[clalancette]

The failures here are all in known flakes, so going ahead and merging this.

[MiguelCompany]

@clalancette @fujitatomoya Should we backport this?

[fujitatomoya]

@MiguelCompany i say we should especially for jazzy and humble, but this is breaking ABI change?

[MiguelCompany]

I say we should especially for jazzy and humble, but this is breaking ABI change?

@fujitatomoya Strictly speaking, this is an ABI break. But since it is on the internal package rmw_fastrtps_shared_cpp, which is only used by the other two packages, it would not be a problem if we release new versions of the three packages in this repo at the same time.

[MiguelCompany]

BTW, it cannot be backported to Humble, because get_shared_instance() was added in Fast DDS 2.8.0, and Humble uses 2.6.x

[fujitatomoya]

@MiguelCompany i am okay with backport for jazzy, lets hear from other opinion.

[gavanderhoorn]

@MiguelCompany i am okay with backport for jazzy, lets hear from other opinion.

If this hasn't yet been backported, I would really appreciate a backport to Jazzy.

[fujitatomoya]

@clalancette what do you think about backporting jazzy? IMO, this should be no problem and better bug fix for jazzy LTS.

[clalancette]

@clalancette what do you think about backporting jazzy? IMO, this should be no problem and better bug fix for jazzy LTS.

I'm not entirely sure, but I think this may be involved in what is causing our crashes on RHEL with flake8. I'm still debugging that, but commenting out the call to delete_participant here makes the crash stop happening.

Thus I think we should hold off on backporting this until we figure out what that problem is.