I have a web2py server running on heroku. So far my server had little traffic, and Heroku being a cloud PaaS service means they put processes to sleep if they're not used for a while.
Because of that, I never really noticed that the memory signature of my application was growing at every request until my server got put to sleep due to inactivity (at night most often), and thus the memory was reset upon restart. This is no longer the case, and now my server often becomes unavailable due to memory quotas, which is a big issue on production. I've been using several tools to track the memory usage of web2py, one of which is the logs of Heroku itself, eg: 2015-02-01T09:01:46.965380+00:00 heroku[web.1]: source=web.1 dyno=heroku. 28228261.c4fede81-d205-4dad-b07e-2ad6dcc49a0f sample#*memory_total=186.27MB sample#memory_rss=180.75MB sample#memory_cache=5.52MB* sample#memory_swap=0.00MB sample#memory_pgpgin=71767pages sample#memory_pgpgout=24081pages As far as I can tell, the "memory_cache" signature is rather small and is stable around 5MB (I don't use cache a lot on my website, precisely for fear of memory leaks). Based on the documentation <https://devcenter.heroku.com/articles/log-runtime-metrics>, "memory_rss" on Heroku refers to: > > - *Resident Memory* (memory_rss): The portion of the dyno’s memory > (megabytes) held in RAM. > > This variable keeps growing at every request, seemingly regardless of which page is accessed. You will find attached the memory chart of my app, taken from Heroku's Metrics back-office : every vertical line is a hard reset I had to do manually to prevent the app from exceeding quotas. The red zone is a server downtime because of swap usage making requests to be served very very slowly. I've tried using the "memory_profiler" library to decorate some functions in my code and try to track memory usage increase, but it is a very tedious task and so far I haven't managed to find the root of my problem. I suspect my issue to be in models since those are loaded on every request, but I can't use memory_profiler on them since models aren't natively encapsulated in any function. Is there a function somewhere in Gluon that loads every model, and which I could decorate to scope which model might have memory leaks ? -- Resources: - http://web2py.com - http://web2py.com/book (Documentation) - http://github.com/web2py/web2py (Source code) - https://code.google.com/p/web2py/issues/list (Report Issues) --- You received this message because you are subscribed to the Google Groups "web2py-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to web2py+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
