On Sunday, December 18, 2016 at 6:26:45 PM UTC-8, mwall wrote: > > sorry in advance for the lack of analysis on this, but here are some raw > data from the heap. > > this is the heap after initial startup, at the first archive interval. > note the total size. > > what follows are the *differences* in the heap at each subsequent archive > interval. these should be heap differences - the service does a > setrelheap() after it prints the heap. after the initial hiccup of almost > 13M, it looks like there is about 124K of memory added to the heap on each > archive interval. most of that is in dicts of weewx.accum.ScalarStats. > > > I guess it would be interesting to drop the system back to totally vanilla weewx with just the standard skin in simulator mode and see what happens.
I have a good Vagrantfile and playbook.yml for ansible that I can provide for automating building simulator VMs with or without forecast if you want. I'm running the latest Vagrant and VirtualBox which I'd recommend. I'm running a default debian VM set to 512MB RAM but of course that's tweakable easily in the Vagrantfile. Feel free to ask via email if you want a copy.... Where I get interested is the behavior where a reboot after the initial seeding of the forecast and runs to create the images+files resets where the system levels out at to a much lower VSZ value. The RSS doesn't change 'that' much, but the virtual size is pretty significantly different. FWIW, I redid the weewx-only test: - first boot stabilized at 167,000 VSZ and 37,000 RSZ - after a reboot, values leveled out at the same values (wow) - so my interpretation is there's something in forecast that is eating ram initially, but less so after it's been initialized and stable for a few runs -- You received this message because you are subscribed to the Google Groups "weewx-user" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
