When trying to diagnose the nature of the crash, I was experimenting with adding a buffering scheme in HTTPGet and quickly discovered that buffering requests (and setting a limit on how often they are allowed to be sent) clearly helped reduce the risc of crashes. It was however when I modified the HTTPGet even further to make it use syncronized http requests (rather than the default async), that I noticed that this seemed to completely prevent the server from crashing. So the final clue was that something was disturbing the server between the http request being sent and the resulting (asyncronous) reply.
I now have the new development build (z-way-4.0.2-4-ge4433cb0-lws16_armhf.deb) installed and running the usual stress testing environment and must say that the results so far is very encouraging - it has been running smoothly for >27 hours by now! (I have of course reverted to the original HTTPGet module, and with the stressful test case I am running this would typically lead to a SIGSEGV crash within 2-6 hours)
I do want to keep the test running for a while before I dare stating the problem is 100% cured, but this is looking very promising indeed.
Huge thanks to PoltoS for always listening to feedback and never giving up!
