The new feature of JK1.2.19: -reference- This directive allows to copy configurations ... doesn't seem to work. I tried several configurations, different names for my workers, with or without "worker." - prefix. The workers won't inherit the properties of my template-worker.
Created attachment 18982 [details] My isapi_redirect.properties The config is done here, not in the registry.
Created attachment 18983 [details] My workers.properties As you can see, i've been experimenting a lot with our 6-node-cluster. nothing worked. Now nearly all is commented out, so that only 2 TCs are working. Still no inheritance. The settings from worker "template" won't appear in the jk manager.
Uriworkermap.properties consists only of 4 lines: /myweb.mydomain.de*/herstportal/*.jsp=LBPTomHH /myweb.mydomain.de*/herstportal/*.do=LBPTomHH /myweb.mydomain.de*/herstportal/*.servlet=LBPTomHH /myweb.mydomain.de*/jkmanager=jkstatus no magic. Btw, all works fine if I define the settings in the workers directly. So I got a workaround. But we also got much bigger clusters, so a function like "reference" would realy come in handy.
Created attachment 18984 [details] Log file in debug Mode Here the log in debug mode, with just one request to jkmanager.
Don't know why it doesn't work. Anyhow you have some errors in the workers.propeties worker.template.connect_timeout=90 worker.template.reply_timeout=900 Remember that those values are in miliseconds (unlike other timeouts that are in seconds). I think I'll add a simple rule: (if timeout < 1000 then timeout = timeout * 1000) Having timeouts smaller then 1 second is insane anyhow, and this would help deling with obvious typos like this.
I had in the past used timeouts of 500 milliseconds so that if a server was un-responsive jk would fail over to another server very quickly without the customer waiting. Half a second doesn't seem that short when our average page time delivered back to the customer across the internet is only 330 milliseconds.
Of cause Mladen Turk is right, i misread the configuration there. Set it to 5000 onthe connect timeout and 900000 on the reply timeout (yes, we have some verry long running servlets). Jeff Johnson might have a point there, but we got a multitier environment with database and mainframe-backends. so no way in answering in that time. Nevertheless, nice comments, but the reported bug is still there, even with the new more sane values.
I agree that the reply_timeout below 1000 is aggressive but the connect_timeout and prepost_timeout (ping/pong) below a second seems very reasonable. Especially when running a cluster that doesn't need sessions. If tomcat can't say that its alive in less than a second, something is probably wrong with it. IIRC we used to have this problem when our garbage collection wasn't tuned well. Regarding the reference problem, I almost reported a bug yesterday saying that reference didn't work but I had forgot the "worker." prefix (which I see that you tried). As soon as I fixed that, my workers did indeed inherit the lbfactor setting so I know it works at least for that in my case.
You are right. The reference feature is missing for non-Apache webservers. We'll clarify the docs and add it to the other webservers. Thank you for your observation and sorry for any inconvenience.
Created attachment 18986 [details] Adding the reference feature to IIS and Netscape You might want to try this patch, since the next release will not be available soon.
The feature will work for IIS and Netscape to, sarting in 1.2.20. Since that version will not be released soon, you might want to try the attached patch for 1.2.19. The docs have been updated with a warning.
Move a couple of fixed JK issues from resolved to closed.