Debugging performance issue.

[thruflo]

Hi,

I put turnstile in front of a production site for a few hours today. Unfortunately, I found that the response time steadily increased. I believe that the graph below shows the time that the request was taking to get through the middleware. The two blue lines are deploys: LHS when I deployed turnstile, RHS when I pulled it.

I was running gunicorn with multiple gevent workers and three WSGI apps in a Paste url map all behind turnstile as a filter. I haven't configured a remote control daemon, because I figured initially it was fine to re-publish XML -> set limits -> restart processes to pick up new config.

In order to workaround my redis host requiring a username, I'm using this as a redis client entry point: https://github.com/thruflo/pyramid_redis/blob/master/src/pyramid_redis/turnstile.py I'm configuring the connection pool with a redis.connection.BlockingConnectionPool and max_connections=4.

Can you give me any pointers as to why the response time was increasing -- i.e.: do I need to look at the control and compaction config, or is it more likely to be my custom redis client factory?

Thanks,

James.