Re: cvs commit: jakarta-tomcat-catalina/catalina/src/share/org/apache/catalina/clusterJGCluster.java JGManager.java ReplicatedSession.java ReplicationStream.javaSerializablePrincipal.java SessionMessage.java

[Remy Maucherat]

Filip Hanik wrote:

Remy wrote:



I'd like to get the save/restore on shutdown features of the std
manager, and can't really see any major problem it would cause.



could be a little tricky. how would you do this without getting the nodes
out of sync.


After reloading and passivating, the node will pull the list of sessions 
from the monitor. I agree this could end up out of sync, but this is a 
useful feature, otherwise the sessions go with the cluster.

then, when a server joins a cluster, there is a possibility that some
sessions might not get replicated correctly, because a state


transfer has to


happen. Thanks to Bela Ban pointing this out

and later in the future, instead of having every node keep a copy of the
session, just use primary and secondary servers for the cluster data.


So only two servers are active members of the cluster ?



nope, it would be more weblogic style, so all servers are members of the
cluster, but data only gets replicated between two of them.


Ok, why not.


Is a cluster with let's say 6 members really too expensive network wise
(assuming a dedicated gig ether link between the cluster members) ?



depends on how much data is being transferred. so the answer could be yes
and no, the current implementation is more expensive than the
primary/secondary solution I was talking about.


Obviously. It's a lot simpler of course ;-)


Anyway, I think by far the most important feature to add is a TCP or
HTTP redirector, possibly written with NBIO. Until then, the clustering
doesn't have much practical interest (usless you're willing to buy
additional expensive hardware; could squid get the job done, BTW ?).



you mean a software load balancer, there are a bunch of them out there
already. I use my homegrown in Java.


Yes, I mean that. If it is useful and works, maybe you'd want to add it 
in JG or contribute it to Tomcat.

I'm ok with proposing you as a committer, provided you accept to respect



I would like to continue the work on this. I think that Bela and myself
would be a big resource to Tomcat clustering.


Ok, I will nominate you then.

Remy


--
To unsubscribe, e-mail:   
For additional commands, e-mail: 

Re: cvs commit: jakarta-tomcat-catalina/catalina/src/share/org/apache/catalina/clusterJGCluster.java JGManager.java ReplicatedSession.java ReplicationStream.javaSerializablePrincipal.java SessionMessage.java

[Remy Maucherat]

Filip Hanik wrote:

Mmm, ok, and would would these be ?
I see a few changes in the sf.net project which haven't been merged, but
it doesn't look like critical changes.


first of all, I wouldn't have the InMemoryReplicationManager extend the
standard manager, since some workarounds had to be done with the principal.


I'd like to get the save/restore on shutdown features of the std 
manager, and can't really see any major problem it would cause.

then, when a server joins a cluster, there is a possibility that some
sessions might not get replicated correctly, because a state transfer has to
happen. Thanks to Bela Ban pointing this out

and later in the future, instead of having every node keep a copy of the
session, just use primary and secondary servers for the cluster data.


So only two servers are active members of the cluster ?

Is a cluster with let's say 6 members really too expensive network wise 
(assuming a dedicated gig ether link between the cluster members) ?

Anyway, I think by far the most important feature to add is a TCP or 
HTTP redirector, possibly written with NBIO. Until then, the clustering 
doesn't have much practical interest (usless you're willing to buy 
additional expensive hardware; could squid get the job done, BTW ?).

I'm ok with proposing you as a committer, provided you accept to respect
the coding style and general design in Tomcat (I've had to do
significant work to adapt the code submitted).



I have no problem with this. You could have even asked me to do this, I had
no idea you were gonna commit the code into the Tomcat CVS :)


Mmmh, ok. I thought when code was contributed, it was to get it in the 
Tomcat CVS ;-)

lemme know what you plan to do


Remy


--
To unsubscribe, e-mail:   
For additional commands, e-mail: 

Re: cvs commit: jakarta-tomcat-catalina/catalina/src/share/org/apache/catalina/clusterJGCluster.java JGManager.java ReplicatedSession.java ReplicationStream.javaSerializablePrincipal.java SessionMessage.java

[Remy Maucherat]

Filip Hanik wrote:

Hi Remy,
we have been actually been working on this at
sourceforge.net/projects/tomcat-jg

could I continue the work in TC?
I have several improvements to make, over the next few months


Mmm, ok, and would would these be ?
I see a few changes in the sf.net project which haven't been merged, but 
it doesn't look like critical changes.

I'm ok with proposing you as a committer, provided you accept to respect 
the coding style and general design in Tomcat (I've had to do 
significant work to adapt the code submitted).

Remy


--
To unsubscribe, e-mail:   
For additional commands, e-mail: 

Re: cvs commit: jakarta-tomcat-catalina/catalina/src/share/org/apache/catalina/clusterJGCluster.java JGManager.java ReplicatedSession.java ReplicationStream.javaSerializablePrincipal.java SessionMessage.java

[Remy Maucherat]

[EMAIL PROTECTED] wrote:

remm2002/12/05 04:25:09

  Added:   catalina/src/share/org/apache/catalina/cluster
JGCluster.java JGManager.java
ReplicatedSession.java ReplicationStream.java
SerializablePrincipal.java SessionMessage.java
  Log:
  - Clustered JG support (untested for now; it builds, and doesn't appear to crash TC).


For extended credits:
- written by Filip Hanik
- resubmitted privately to me by Bela Ban

Remy


--
To unsubscribe, e-mail:   
For additional commands, e-mail: