profiler.py:ProfilerWSGIMiddleware:__call__ is not threadsafe

[holte]

This method seems to have a race condition that eventually leads to max recursion depth exceeded errors:

The code has the following pattern:

old_memcache_add = memcache.add
memcache.add = (lambda ... : ... old_memcache_add)
...
memcache.add = old_memcache_add

If this is executed by two threads in a particular order it leaves memcache.add in the wrong state, e.g.:

Thread 1: old_memcache_add = memcache.add [1.old_memcache_add = MA]
Thread 1: memcache.add = (lambda ... : ... old_memcache_add) [ memcache.add = L(MA)]
Thread 2: old_memcache_add = memcache.add [2.old_memcache_add = L(MA)]
Thread 2: memcache.add = (lambda ... : ... old_memcache_add) [ memcache.add = L(L(MA))]
Thread 1: memcache.add = old_memcache_add [ memcache.add = MA]
Thread 2: memcache.add = old_memcache_add [ memcache.add = L(MA)]

[benjaminjkraft]

Thanks for the report! This does seem like a potential issue. At the moment I don't think we at KA use the mini-profiler much in multithreaded contexts so this won't be high-priority for us to fix, but if it affects others I'd be happy to accept a pull request.