I have looked through the code and only purging sessions that the
connection is idle goes back to day one for context.c 

Throughout the code (modules too), kill_me gets set in all sorts of
conditions.  I am not sure what the state of the connection is in all
of these cases, but the one I am running into is when the client has a
connection open, but does not do anything in the time configured, it
sets the kill_me to KILLME_IDLE.  Since the connection is not idle, the
session never is ended.  I get hundreds of these in production. 

Looking at the other places that set kill_me, I think they all expect
the purge to kill them, but I cannot be sure, and I don't even know how
to test every condition. 

On my server, I added purging sessions that were set to KILLME_IDLE and
it seems to clean up my issue.  Not sure how to test if we purge all
kill_me scenarios. 

I think adding the check for KILLME_IDLE is worth adding to the code,
but what are the thoughts about the risk of purging all?

Reply via email to