hi everybody!

while testing a distributed scenario (1 server without tables as main entry
point; 1 server with tables and a resthost entry) I encountered some strange
behavior:

After several minutes of playing with approx. 200 simultaneous users, the
explain server suddenly stops working, not opening new connections, and
hanging at 100% cpu. The cpu usage still goes on even when no user is
connected to the server anymore. This outcome happens every time we started
the load test.

my first guess was the explain server did not close the restclients
correctly, but the logs show no increase of opened connex/second (approx.
100 connex/second). I tried to visualize the load of the participating
servers, the graph can be found at http://db.tt/beJLtab


Does anybody deployed the pokerservers in a distributed scenario and he
experienced something similar? Also, if anybody has suggestions for the
currently tested scenario, please let me know! I only recently started load
testing and would be glad to hear of working deployment scenarios.


For the aforementioned test, 4 large EC2 Instances were used, containing:
- server1: nginx, php, memcache, nodejs
- server2: mysql 5.5
- server3: pythonpoker, not containing any tables. nginx points the
/POKER_REST path to this server
- server4: pythonpoker, containing tables, and having a resthost entry for
itself.

All machines are Ubuntu 10.04, on m1.large Amazon EC2 instances (x86_64
platforms)
Pythonpoker was installed from source from (svn trunk)

By eliminating server3 and routing /POKER_REST to server4, the setup works
as expected, so I suppose it has nothing to do with the setup, os, or other
external factors.



Andor
_______________________________________________
Pokersource-users mailing list
[email protected]
https://mail.gna.org/listinfo/pokersource-users

Reply via email to