You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@tomcat.apache.org by Ch Praveena <ps...@gmail.com> on 2008/04/08 09:30:54 UTC
Split-brain problem.
Hi all,
While going on with Terracotta article,
-- http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works
,
For implementing application server clustering, I found this phrase,
"Split-brain problem."..
Please let me know about this in detail..
Regards,
Praveena Chalamcharla,
Securview....
Re: [users@httpd] Split-brain problem.
Posted by Ch Praveena <ps...@gmail.com>.
Thank you Oscar, for your information.
2008/4/8 Oscar Haeger <os...@sigma.se>:
> Hi.
> Heartbeat is usually a private network between the nodes, used to send
> data about which nodes are alive. You'll want a low latency network for
> this. Google for clusters and heartbeat and you'll find lots of information
> I'm sure.
>
> Clusters can be used as High-Availiblity where server B is indeed just a
> backup, but it can also be used as High-Performance where two or more
> servers share the requests between them. Obviously the performance will
> degrade if parts of that cluster goes down. So yes, B can both be a backup
> and an equally performing server, depending on your setup.
>
> I'm a bit out of my compentence area here as I haven't done webserver
> clustering, but I do believe the general principle is about right. I'm sure
> someone on this list will correct me where I'm wrong.
>
> Cheers,
>
> Oscar
>
> ________________________________
>
> Från: Ch Praveena [mailto:ps050202@gmail.com]
> Skickat: ti 2008-04-08 10:47
> Till: users@httpd.apache.org
> Ämne: Re: [users@httpd] Split-brain problem.
>
>
> Thank you Oscar,
>
> As I am new to clustering Tomcat, Let me know some information about
> Heartbeat too..
>
> And according to your information, Is that Server B a backup for Server
> A(So that when A fails, B attain its responsibilites) or an equally
> performing server with A.
>
>
>
>
> 2008/4/8 Oscar Haeger <os...@sigma.se>:
>
>
> From my experience with clustering that is when the connection
> between parts of the clusters are lost and they can't agree on which one is
> supposed to keep serving the data. The most common way around it afaik is to
> have a priority and heartbeat between the nodes and allow them to kill each
> other off (something like STONITH = Shoot The Other Node In The Head) or
> similar.
>
> So, if we have two servers A and B and a heartbeat (with STONITH)
> between them and node A misses a few heartbeats (perhaps a lockup, perhaps
> something else), then node B will reset node A and take over that nodes
> capabilities.
>
> This is without reading about Terracotta so things might be
> different there, but that's the split-brain problem from my experience.
>
> Hope this helps
>
> Cheers
>
> Oscar
>
> ________________________________
>
> Från: Ch Praveena [mailto:ps050202@gmail.com]
> Skickat: ti 2008-04-08 09:30
> Till: users@httpd.apache.org
> Ämne: [users@httpd] Split-brain problem.
>
>
>
>
> Hi all,
>
> While going on with Terracotta article,
> --
> http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works,
> For implementing application server clustering, I found this
> phrase, "Split-brain problem."..
>
> Please let me know about this in detail..
>
>
>
> Regards,
>
> Praveena Chalamcharla,
> Securview....
>
>
>
> ---------------------------------------------------------------------
> The official User-To-User support forum of the Apache HTTP Server
> Project.
> See <URL:http://httpd.apache.org/userslist.html> for more info.
> To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
> " from the digest: users-digest-unsubscribe@httpd.apache.org
> For additional commands, e-mail: users-help@httpd.apache.org
>
>
>
>
>
> --
> Regards,
>
> Praveena Chalamcharla,
> Securview....
>
> ---------------------------------------------------------------------
> The official User-To-User support forum of the Apache HTTP Server Project.
> See <URL:http://httpd.apache.org/userslist.html> for more info.
> To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
> " from the digest: users-digest-unsubscribe@httpd.apache.org
> For additional commands, e-mail: users-help@httpd.apache.org
>
--
Regards,
Praveena Chalamcharla,
Securview....
SV: [users@httpd] Split-brain problem.
Posted by Oscar Haeger <os...@sigma.se>.
Hi.
Heartbeat is usually a private network between the nodes, used to send data about which nodes are alive. You'll want a low latency network for this. Google for clusters and heartbeat and you'll find lots of information I'm sure.
Clusters can be used as High-Availiblity where server B is indeed just a backup, but it can also be used as High-Performance where two or more servers share the requests between them. Obviously the performance will degrade if parts of that cluster goes down. So yes, B can both be a backup and an equally performing server, depending on your setup.
I'm a bit out of my compentence area here as I haven't done webserver clustering, but I do believe the general principle is about right. I'm sure someone on this list will correct me where I'm wrong.
Cheers,
Oscar
________________________________
Från: Ch Praveena [mailto:ps050202@gmail.com]
Skickat: ti 2008-04-08 10:47
Till: users@httpd.apache.org
Ämne: Re: [users@httpd] Split-brain problem.
Thank you Oscar,
As I am new to clustering Tomcat, Let me know some information about Heartbeat too..
And according to your information, Is that Server B a backup for Server A(So that when A fails, B attain its responsibilites) or an equally performing server with A.
2008/4/8 Oscar Haeger <os...@sigma.se>:
From my experience with clustering that is when the connection between parts of the clusters are lost and they can't agree on which one is supposed to keep serving the data. The most common way around it afaik is to have a priority and heartbeat between the nodes and allow them to kill each other off (something like STONITH = Shoot The Other Node In The Head) or similar.
So, if we have two servers A and B and a heartbeat (with STONITH) between them and node A misses a few heartbeats (perhaps a lockup, perhaps something else), then node B will reset node A and take over that nodes capabilities.
This is without reading about Terracotta so things might be different there, but that's the split-brain problem from my experience.
Hope this helps
Cheers
Oscar
________________________________
Från: Ch Praveena [mailto:ps050202@gmail.com]
Skickat: ti 2008-04-08 09:30
Till: users@httpd.apache.org
Ämne: [users@httpd] Split-brain problem.
Hi all,
While going on with Terracotta article,
-- http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works,
For implementing application server clustering, I found this phrase, "Split-brain problem."..
Please let me know about this in detail..
Regards,
Praveena Chalamcharla,
Securview....
---------------------------------------------------------------------
The official User-To-User support forum of the Apache HTTP Server Project.
See <URL:http://httpd.apache.org/userslist.html> for more info.
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
" from the digest: users-digest-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org
--
Regards,
Praveena Chalamcharla,
Securview....
Re: [users@httpd] Split-brain problem.
Posted by Ch Praveena <ps...@gmail.com>.
Thank you Oscar,
As I am new to clustering Tomcat, Let me know some information about
Heartbeat too..
And according to your information, Is that Server B a backup for Server A(So
that when A fails, B attain its responsibilites) or an equally performing
server with A.
2008/4/8 Oscar Haeger <os...@sigma.se>:
> From my experience with clustering that is when the connection between
> parts of the clusters are lost and they can't agree on which one is supposed
> to keep serving the data. The most common way around it afaik is to have a
> priority and heartbeat between the nodes and allow them to kill each other
> off (something like STONITH = Shoot The Other Node In The Head) or similar.
>
> So, if we have two servers A and B and a heartbeat (with STONITH) between
> them and node A misses a few heartbeats (perhaps a lockup, perhaps something
> else), then node B will reset node A and take over that nodes capabilities.
>
> This is without reading about Terracotta so things might be different
> there, but that's the split-brain problem from my experience.
>
> Hope this helps
>
> Cheers
>
> Oscar
>
> ________________________________
>
> Från: Ch Praveena [mailto:ps050202@gmail.com]
> Skickat: ti 2008-04-08 09:30
> Till: users@httpd.apache.org
> Ämne: [users@httpd] Split-brain problem.
>
>
>
> Hi all,
>
> While going on with Terracotta article,
> --
> http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works,
> For implementing application server clustering, I found this phrase,
> "Split-brain problem."..
>
> Please let me know about this in detail..
>
>
>
> Regards,
>
> Praveena Chalamcharla,
> Securview....
>
> ---------------------------------------------------------------------
> The official User-To-User support forum of the Apache HTTP Server Project.
> See <URL:http://httpd.apache.org/userslist.html> for more info.
> To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
> " from the digest: users-digest-unsubscribe@httpd.apache.org
> For additional commands, e-mail: users-help@httpd.apache.org
>
--
Regards,
Praveena Chalamcharla,
Securview....
SV: [users@httpd] Split-brain problem.
Posted by Oscar Haeger <os...@sigma.se>.
>From my experience with clustering that is when the connection between parts of the clusters are lost and they can't agree on which one is supposed to keep serving the data. The most common way around it afaik is to have a priority and heartbeat between the nodes and allow them to kill each other off (something like STONITH = Shoot The Other Node In The Head) or similar.
So, if we have two servers A and B and a heartbeat (with STONITH) between them and node A misses a few heartbeats (perhaps a lockup, perhaps something else), then node B will reset node A and take over that nodes capabilities.
This is without reading about Terracotta so things might be different there, but that's the split-brain problem from my experience.
Hope this helps
Cheers
Oscar
________________________________
Från: Ch Praveena [mailto:ps050202@gmail.com]
Skickat: ti 2008-04-08 09:30
Till: users@httpd.apache.org
Ämne: [users@httpd] Split-brain problem.
Hi all,
While going on with Terracotta article,
-- http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works,
For implementing application server clustering, I found this phrase, "Split-brain problem."..
Please let me know about this in detail..
Regards,
Praveena Chalamcharla,
Securview....
Re: Split-brain problem.
Posted by Kevan Miller <ke...@gmail.com>.
Strange. The "To" address on this posting is
"users@httpd.apache.org" (at least in my mail reader...
Praveena,
"Split-brain" refers to a problem in clustering where cluster members
are unable to communicate with each other. If a cluster implementation
doesn't take precautions, you may end up with each of the cluster
members believing that they are the primary member -- thus a split
brain, one side doesn't know what the other is doing...
--kevan
On Apr 8, 2008, at 3:30 AM, Ch Praveena wrote:
>
> Hi all,
>
> While going on with Terracotta article,
> -- http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works
> ,
> For implementing application server clustering, I found this phrase,
> "Split-brain problem."..
>
> Please let me know about this in detail..
Re: Split-brain problem.
Posted by wes <we...@the-wes.com>.
Anything can cause a netsplit. Anything that makes the 2 sets of servers
lose connection to each other will cause a netsplit. Dead router, cut wire,
power outage, configuration changes, any number of things.
What I don't understand is, if there's a netsplit, the servers on the other
side should be "unreachable" to the load balancer, and thus should not have
any more requests directed to them until they become reachable again.
-wes
On 4/8/08, Ch Praveena <ps...@gmail.com> wrote:
>
> Thankyou Marcus,
> But let me know the reasons for the occurence of the Netsplit, which lead
> Split brain problem.
>
>
>
>
> 2008/4/8 Marcus Franke <mf...@evendi.de>:
>
>
> > On Tue, Apr 08, 2008 at 01:00:54PM +0530, Ch Praveena wrote:
> > > Hi all,
> > >
> > > While going on with Terracotta article,
> > > --
> >
> http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works
> > > ,
> > > For implementing application server clustering, I found this phrase,
> > > "Split-brain problem."..
> > >
> > > Please let me know about this in detail..
> > >
> >
> > Easy,
> >
> > if you have a cluster of 4 nodes, and you have a netsplit, you'll end
> with
> > two clusters with two nodes. Split brain problem is that both parts are
> > equal in size, so how can the software decide, which side is the correct
> > one and which nodes shall drop dead and restart?
> >
> > Because of this you may want an uneven number of nodes in your cluster.
> > This
> > way you cannot have a split'ed cluster with equal numbers of nodes on
> each
> > side.
> >
> >
> >
> > regards,
> > Marcus
> >
> > ---------------------------------------------------------------------
> > To start a new topic, e-mail: users@tomcat.apache.org
> > To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
> > For additional commands, e-mail: users-help@tomcat.apache.org
> >
> >
>
>
>
> --
> Regards,
>
> Praveena Chalamcharla,
> Securview....
>
Re: Split-brain problem.
Posted by Ch Praveena <ps...@gmail.com>.
Thankyou Marcus,
But let me know the reasons for the occurence of the Netsplit, which lead
Split brain problem.
2008/4/8 Marcus Franke <mf...@evendi.de>:
> On Tue, Apr 08, 2008 at 01:00:54PM +0530, Ch Praveena wrote:
> > Hi all,
> >
> > While going on with Terracotta article,
> > --
> http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works
> > ,
> > For implementing application server clustering, I found this phrase,
> > "Split-brain problem."..
> >
> > Please let me know about this in detail..
> >
>
> Easy,
>
> if you have a cluster of 4 nodes, and you have a netsplit, you'll end with
> two clusters with two nodes. Split brain problem is that both parts are
> equal in size, so how can the software decide, which side is the correct
> one and which nodes shall drop dead and restart?
>
> Because of this you may want an uneven number of nodes in your cluster.
> This
> way you cannot have a split'ed cluster with equal numbers of nodes on each
> side.
>
>
>
> regards,
> Marcus
>
> ---------------------------------------------------------------------
> To start a new topic, e-mail: users@tomcat.apache.org
> To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
> For additional commands, e-mail: users-help@tomcat.apache.org
>
>
--
Regards,
Praveena Chalamcharla,
Securview....
Re: Split-brain problem.
Posted by Marcus Franke <mf...@evendi.de>.
On Tue, Apr 08, 2008 at 01:00:54PM +0530, Ch Praveena wrote:
> Hi all,
>
> While going on with Terracotta article,
> -- http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works
> ,
> For implementing application server clustering, I found this phrase,
> "Split-brain problem."..
>
> Please let me know about this in detail..
>
Easy,
if you have a cluster of 4 nodes, and you have a netsplit, you'll end with
two clusters with two nodes. Split brain problem is that both parts are
equal in size, so how can the software decide, which side is the correct
one and which nodes shall drop dead and restart?
Because of this you may want an uneven number of nodes in your cluster. This
way you cannot have a split'ed cluster with equal numbers of nodes on each
side.
regards,
Marcus
---------------------------------------------------------------------
To start a new topic, e-mail: users@tomcat.apache.org
To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
For additional commands, e-mail: users-help@tomcat.apache.org
Re: Split-brain problem.
Posted by Jason Warner <ja...@gmail.com>.
Hi Praveena,
Terracotta is actually a completely separate project. If you have questions
regarding their documentation, I suggest asking on the Terracotta user lists
or on their IRC channel, both of which can be found using the documentation
on their website, www.terracotta.org.
Thanks!
On Tue, Apr 8, 2008 at 3:30 AM, Ch Praveena <ps...@gmail.com> wrote:
>
> Hi all,
>
> While going on with Terracotta article,
> --
> http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works,
> For implementing application server clustering, I found this phrase,
> "Split-brain problem."..
>
> Please let me know about this in detail..
>
>
>
> Regards,
>
> Praveena Chalamcharla,
> Securview....
--
~Jason Warner
Re: Split-brain problem.
Posted by SriSri <sr...@gmail.com>.
Split-Brain in that context might have meant that the individual cluster app
server is good enough to serve all the requests.
On Tue, Apr 8, 2008 at 3:30 AM, Ch Praveena <ps...@gmail.com> wrote:
> Hi all,
>
> While going on with Terracotta article,
> --
> http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works
> ,
> For implementing application server clustering, I found this phrase,
> "Split-brain problem."..
>
> Please let me know about this in detail..
>
>
>
> Regards,
>
> Praveena Chalamcharla,
> Securview....
>
Re: [users@httpd] Split-brain problem.
Posted by Nils Jeppe <ni...@pandemonium.de>.
On 08.04.2008, at 09:30, Ch Praveena wrote:
> While going on with Terracotta article,
> -- http://www.terracotta.org/confluence/display/explore/How+Terracotta+Works
> ,
> For implementing application server clustering, I found this phrase,
> "Split-brain problem."..
>
> Please let me know about this in detail..
Split brain refers to the problem where you have two equal servers,
and a problem occurs that causes the state to diverge between the two.
The cluster won't be able to decide on its own which version of the
state is "correct". One solution would be to employ at least three
systems and have them go by a majority decision. It's really more an
application / database problem that transcends Apache.
Best wishes
Nils
---------------------------------------------------------------------
The official User-To-User support forum of the Apache HTTP Server Project.
See <URL:http://httpd.apache.org/userslist.html> for more info.
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
" from the digest: users-digest-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org