Comment 27 for bug 1874831

I was able to recreate the error now.
I managed to trigger that by subscribing after a server restart with no message posted:

$ sudo systemctl restart nginx
$ sudo tail -f /var/log/nginx/error.log
# wait here then in another console
$ curl -i --header 'Accept: text/event-stream' --header "Content-Type: application/json" --get --data-urlencode 'payload={"topics":["one","two"]}' http://localhost:8000/subscribe
# The error log will spit out:
ker process: /build/nginx-KTLRnK/nginx-1.18.0/debian/modules/nchan/src/store/memory/ipc-handlers.c:417: memstore_ipc_send_get_message: Assertion `data.shm_chid->len>1' failed.
2021/11/15 11:09:07 [alert] 5340#5340: worker process 5344 exited on signal 6 (core dumped)

And that is indeed fixed by the PPA.

But BTW if you point nchan_publisher_upstream_request to an invalid target, then even with the fix it does fail the same way. And actually with the new version I had a hard time to get it right. I have found cases which formerly worked (basic nchan config post-pub + get+sub) which now after the fix is installed trigger a crash like this. This is kind of the very definition of what the SRU policy tries to prevent :-/

The argument for "allowing" the full version bump of this embedded module was that without this it isn't usable at all (therefore we can't regress anyone). But as we've seen it seems there are (basic?) configurations which were fine before, therefore I'm not entirely sure we can really push this as an SRU :-/

@teward @racb - would you mind to have a look and comment what you think about it?