Mantis - Resin
Viewing Issue Advanced Details
ID: Category: Severity: Reproducibility: Date Submitted: Last Update:
1195 major always 06-13-06 01:12 06-15-06 17:08
Reporter: jornsvendsen Platform:  
Assigned To: ferg OS:  
Priority: normal OS Version:  
Status: closed Product Version: 3.0.18  
Product Build: Resolution: fixed  
Projection: none      
ETA: none Fixed in Version: 3.0.20  
Summary: 0001195: Loadbalancer servlet give long delays when a backend server is powered off
Description: When the loadbalancing servlet is defined for load distribution to 2 other servers and one of the servers is powered off, we get unreasonable long times for serving the content.
If the powered off server is removed from the cluster definition, content is again served with no problems.

We consider finding another way to do the load distribution because a failed server makes the service useless.
I think the failing server must be marked unavailable and no live requests should be sent to it before it is verified (background task) that the server is actually up and running.
Steps To Reproduce:
Additional Information:
Relationships
Attached Files:

Notes
(0001300)
ferg   
06-15-06 17:08   
A number of changes in 3.0.20 will help in this case.

client-connect-timeout - will limit the time trying to connect to a starting server

client-fail-recover-time (was client-dead-time) - configures how lot to treat a dead server as dead before retrying

client-slow-start-time - throttles connections for a newly started server (which also means only one request would be penalized by the connect timeout)