Peter Rossbach wrote:
Here my wishes list:
a) Clearer/better Loadbalancing Algo
- better understanding of domain/partition based strategy ( I have
detect some problemes here)
- busy strategy -
http://issues.apache.org/bugzilla/show_bug.cgi?id=36138
- real load strategy ( ask tomcats before balance the load)
- good documentation of all use cases and the protocol, before
implementation ;-)
- make extensions possible
b) auto reconfig was very nice but this feature is very difficult to
implement. Currently graceful restart works good for me!
c) disable/enable mapping for some worker inside the same loadbalancer
Then it is possible to reload/redeploy a single application at
some cluster nodes
d) AJP client java implementation and integrate this inside a tomcat
proxy module.
e) Refactor/Redesign the current AJP server implementation
Right, all that is wishlist for 1.3 branch not a new protocol.
One of the major things I see as protocol limit is graceful connection
close notification. Something like that can be easily implemented by
extending the existing protocol with the sort of CPING/CPONG messages.
That way the graceful restart would rely work without spinning the
thread count on Tomcat, and using connectionTimeout for killing the
zombie sockets.
Like said, couple of additions to the existing AJP1.3 protocol would
allow us to resolve all (most) of the existing problems observed.
Regards,
Mladen.
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
