Hello,
I'm beginning to give up on making Linux-HA's heartbeat work for my environment (CentOS x86_64) and am wondering what other option have I got to help me: 1. Use IPVS to maintain a cluster of virtual servers, either master/slave or load-balanced. 2. Use DRBD in master/slave fashion to keep a home-grown application highly-available.
The first thing I stumbled upon is RedHat Cluster Suite ( http://www.linuxtopia.org/online_books/centos_linux_guides/centos_cluster_co...), from which I also saw some packages on my CentOS servers.
I've never heard of it before and am just starting to dig its docs, but if someone here can confirm/deny that this is a possible route to take it might save me some time or doubts.
Thanks,
--Amos
On Fri, 2007-11-30 at 09:30 +0000, Amos Shapira wrote:
Hello,
I'm beginning to give up on making Linux-HA's heartbeat work for my environment (CentOS x86_64) and am wondering what other option have I got to help me:
- Use IPVS to maintain a cluster of virtual servers, either
master/slave or load-balanced. 2. Use DRBD in master/slave fashion to keep a home-grown application highly-available.
The first thing I stumbled upon is RedHat Cluster Suite (http://www.linuxtopia.org/online_books/centos_linux_guides/centos_cluster_co...), from which I also saw some packages on my CentOS servers.
I've never heard of it before and am just starting to dig its docs, but if someone here can confirm/deny that this is a possible route to take it might save me some time or doubts.
Thanks,
--Amos
Perhaps you should read the CentOS docs - see http://www.centos.org/docs/5/ - cluster suite overview and admin
Rob
On Nov 30, 2007 4:30 AM, Amos Shapira amos.shapira@gmail.com wrote:
Hello,
I'm beginning to give up on making Linux-HA's heartbeat work for my environment (CentOS x86_64) and am wondering what other option have I got to help me:
- Use IPVS to maintain a cluster of virtual servers, either master/slave or
load-balanced. 2. Use DRBD in master/slave fashion to keep a home-grown application highly-available.
The first thing I stumbled upon is RedHat Cluster Suite ( http://www.linuxtopia.org/online_books/centos_linux_guides/centos_cluster_co...), from which I also saw some packages on my CentOS servers.
I've never heard of it before and am just starting to dig its docs, but if someone here can confirm/deny that this is a possible route to take it might save me some time or doubts.
Dump the cluste suite and use the LinuxVirtualServer.org packages. You won't have a gui, but it will be better in the long run. We're using that for quite a few clusters and handling about 30MBit/s on each of the clusters, I think it's around 10k concurrent connections.
On 30/11/2007, Matt Shields mattboston@gmail.com wrote:
Dump the cluste suite and use the LinuxVirtualServer.org packages. You won't have a gui, but it will be better in the long run. We're using that for quite a few clusters and handling about 30MBit/s on each of the clusters, I think it's around 10k concurrent connections.
I also need to fail-over DRBD (i.e. so if the primary goes down the secondary will notice this, mount that DRBD partition and start the server which uses the files on it) - will LVS give me that by itself or will I need something else on top of it to do that? I got the impression that this what Linux-HA's heartbeat adds to the plain LVS but it doesn't work for me.
I'm really not concerned about GUI's - I'd rather edit config files manually if they are documented well enough.
--Amos
On Nov 30, 2007 3:21 PM, Amos Shapira amos.shapira@gmail.com wrote:
On 30/11/2007, Matt Shields mattboston@gmail.com wrote:
Dump the cluste suite and use the LinuxVirtualServer.org packages. You won't have a gui, but it will be better in the long run. We're using that for quite a few clusters and handling about 30MBit/s on each of the clusters, I think it's around 10k concurrent connections.
I also need to fail-over DRBD (i.e. so if the primary goes down the secondary will notice this, mount that DRBD partition and start the server which uses the files on it) - will LVS give me that by itself or will I need something else on top of it to do that? I got the impression that this what Linux-HA's heartbeat adds to the plain LVS but it doesn't work for me.
I'm really not concerned about GUI's - I'd rather edit config files manually if they are documented well enough.
--Amos
Yup. We use LVS for all types of failover senarios. We use it for redundant firewall/vpn servers which use heartbeat for failing over the virtual IPs and services. We also use LVS with ldirectord as redundant load balancers. Read the docs, they explain how to set up a service to be started/stopped on failover
On 30/11/2007, Matt Shields mattboston@gmail.com wrote:
Yup. We use LVS for all types of failover senarios. We use it for redundant firewall/vpn servers which use heartbeat for failing over
So you are using the same heartbeat that doesn't work for me? Or are you refering to another package which provides a similar functionality?
the virtual IPs and services. We also use LVS with ldirectord as redundant load balancers. Read the docs, they explain how to set up a service to be started/stopped on failover
I've been digging the web for over a week now but it just doesn't work the way it's supposed to, whatever I try.
Any other hints?
Thanks,
--Amos
On Nov 30, 2007 3:57 PM, John R Pierce pierce@hogranch.com wrote:
Matt Shields wrote:
Dump the cluste suite and use the LinuxVirtualServer.org packages.
isn't that heartbeat and stuff repackaged?
With a GUI that actually makes it more difficult to manage. Learn to use the command line tools and config files, it's so much easier.
then why in Gaea's name . <snip>
gaea?
whoa nellie, I got first dibs on preaching Jesus here
;->
Get in line.
:-)
- rh
On 30/11/2007, John R Pierce pierce@hogranch.com wrote:
Matt Shields wrote:
With a GUI that actually makes it more difficult to manage. Learn to use the command line tools and config files, it's so much easier.
then why in Gaea's name did they make the heartbeat config files XML ??
while XML -can- be human read, its a freekin' mess to read and edit and maintain sanity.
I wish THAT was my problem :).
While I'm not fond of manually manipulating XML (XML is usually meant to be touched by programs, not humans), I can cope with it if the f*** programs executed properly.
As it is now, even the BasicSanityCheck fails.
I'm trying to run the system-config-cluster thing and see what happens.
Cheers,
--Amos
On 30/11/2007, Matt Shields mattboston@gmail.com wrote:
On Nov 30, 2007 3:57 PM, John R Pierce pierce@hogranch.com wrote:
Matt Shields wrote:
Dump the cluste suite and use the LinuxVirtualServer.org packages.
isn't that heartbeat and stuff repackaged?
With a GUI that actually makes it more difficult to manage. Learn to use the command line tools and config files, it's so much easier.
So it uses the same heartbeat as the one which comes in the "heartbeat" CentOS 5 package? I was hoping they implement their own thing. The last thing I need now is glossy interface which hides the little details that might help me understand what's wrong.
Unless that interface can magically configure heartbeat in a way that it'll actually start running without core-dump'ing some of the programs it comes with.
Thanks,
--Amos
John R Pierce wrote:
Matt Shields wrote:
Dump the cluste suite and use the LinuxVirtualServer.org packages.
isn't that heartbeat and stuff repackaged?
Visiting the web site it appears to be a load-balancer, not that that wouldn't be useful in some scenarios, but it isn't really "clustering" software that is to have an application run in active/passive or active/active between multiple nodes cooperatively, with fencing and shared storage, locking and all that goes with that.
-Ross
______________________________________________________________________ This e-mail, and any attachments thereto, is intended only for use by the addressee(s) named herein and may contain legally privileged and/or confidential information. If you are not the intended recipient of this e-mail, you are hereby notified that any dissemination, distribution or copying of this e-mail, and any attachments thereto, is strictly prohibited. If you have received this e-mail in error, please immediately notify the sender and permanently delete the original and any copy or printout thereof.
On Nov 30, 2007 4:12 PM, Ross S. W. Walker rwalker@medallion.com wrote:
John R Pierce wrote:
Matt Shields wrote:
Dump the cluste suite and use the LinuxVirtualServer.org packages.
isn't that heartbeat and stuff repackaged?
Visiting the web site it appears to be a load-balancer, not that that wouldn't be useful in some scenarios, but it isn't really "clustering" software that is to have an application run in active/passive or active/active between multiple nodes cooperatively, with fencing and shared storage, locking and all that goes with that.
-Ross
LVS is a group of tools that do a lot of different things, the two that you are interested in are:
- heartbeat - provides failover if you have two nodes (active/active or active/passive) - ipvsadm/ldirectord - provides load balancing (ie. http(s) load balancer in front of multiple web servers)
As stated in a previous post we have a number of these setup in our network and we handle a lot of traffic. Some we're using for http(s) traffic, others smtp/pop/imap, others mysql (read only queries off replicas). There's no end to what what you could use heartbeat or ipvsadm/ldirectord or both for. Both packages can be installed from dag's repo.
On 30/11/2007, Matt Shields mattboston@gmail.com wrote:
LVS is a group of tools that do a lot of different things, the two that you are interested in are:
- heartbeat - provides failover if you have two nodes (active/active
or active/passive)
- ipvsadm/ldirectord - provides load balancing (ie. http(s) load
balancer in front of multiple web servers)
As stated in a previous post we have a number of these setup in our network and we handle a lot of traffic. Some we're using for http(s) traffic, others smtp/pop/imap, others mysql (read only queries off replicas). There's no end to what what you could use heartbeat or ipvsadm/ldirectord or both for. Both packages can be installed from dag's repo.
Thanks.
What platform are you using? Mine is CentOS 5 on x86_64. It runs as a Xen DomU but from what I read on the linux-ha users mailing list this shouldn't be the issue. The production system will run on the bare metal (not under Xen).
My experience with LVS at a previous workplace (a very large ISP) was also excellent - they had a couple of LVS servers in front of hundreds of mini-clusters (each such cluster service its own web or other network application, sometimes sharing disks using DRBD).
The difference, I suspect, is that I'm trying this now with version 2.1.2 on CentOS 5 and x86_64, as opposed to possibly older version of everything (RedHat version, LVS, hardware (i386)).
Thanks for your input,
--Amos
On Nov 30, 2007 6:28 PM, Amos Shapira amos.shapira@gmail.com wrote:
On 30/11/2007, Matt Shields mattboston@gmail.com wrote:
LVS is a group of tools that do a lot of different things, the two that you are interested in are:
- heartbeat - provides failover if you have two nodes (active/active
or active/passive)
- ipvsadm/ldirectord - provides load balancing (ie. http(s) load
balancer in front of multiple web servers)
As stated in a previous post we have a number of these setup in our network and we handle a lot of traffic. Some we're using for http(s) traffic, others smtp/pop/imap, others mysql (read only queries off replicas). There's no end to what what you could use heartbeat or ipvsadm/ldirectord or both for. Both packages can be installed from dag's repo.
Thanks.
What platform are you using? Mine is CentOS 5 on x86_64. It runs as a Xen DomU but from what I read on the linux-ha users mailing list this shouldn't be the issue. The production system will run on the bare metal (not under Xen).
My experience with LVS at a previous workplace (a very large ISP) was also excellent - they had a couple of LVS servers in front of hundreds of mini-clusters (each such cluster service its own web or other network application, sometimes sharing disks using DRBD).
The difference, I suspect, is that I'm trying this now with version 2.1.2 on CentOS 5 and x86_64, as opposed to possibly older version of everything (RedHat version, LVS, hardware (i386)).
Thanks for your input,
Take Xen out of the picture until you learn how heartbeat and ipvsadm/ldirectord actually work. You could be having network issues because you are hosting it on a virtual server instead of on a real server. So it's kinda hard to troubleshoot if you don't even know if your configs are broken. Get two crappy boxes that you can load everything up on, configure them with heartbeat, get that working where it will failover an IP. then add some other service like ipvsadm/ldirectord, and take things one step at a time. Don't try to setup everything all at once, it makes it harder to try to debug problems.
I'm using CentOS4 and RHEL4 using dag'd rpms on a few of the CentOS and RHEL boxes and built from source on some of the other ones. I haven't had a chance to try out a CentOS 5 system yet. But as to your stability questions, we've been using LVS for about 3 or 4 years now and never, ever had stability problems.
On Nov 30, 2007 6:40 PM, Matt Shields mattboston@gmail.com wrote:
On Nov 30, 2007 6:28 PM, Amos Shapira amos.shapira@gmail.com wrote:
On 30/11/2007, Matt Shields mattboston@gmail.com wrote:
LVS is a group of tools that do a lot of different things, the two that you are interested in are:
- heartbeat - provides failover if you have two nodes (active/active
or active/passive)
- ipvsadm/ldirectord - provides load balancing (ie. http(s) load
balancer in front of multiple web servers)
As stated in a previous post we have a number of these setup in our network and we handle a lot of traffic. Some we're using for http(s) traffic, others smtp/pop/imap, others mysql (read only queries off replicas). There's no end to what what you could use heartbeat or ipvsadm/ldirectord or both for. Both packages can be installed from dag's repo.
Thanks.
What platform are you using? Mine is CentOS 5 on x86_64. It runs as a Xen DomU but from what I read on the linux-ha users mailing list this shouldn't be the issue. The production system will run on the bare metal (not under Xen).
My experience with LVS at a previous workplace (a very large ISP) was also excellent - they had a couple of LVS servers in front of hundreds of mini-clusters (each such cluster service its own web or other network application, sometimes sharing disks using DRBD).
The difference, I suspect, is that I'm trying this now with version 2.1.2 on CentOS 5 and x86_64, as opposed to possibly older version of everything (RedHat version, LVS, hardware (i386)).
Thanks for your input,
Take Xen out of the picture until you learn how heartbeat and ipvsadm/ldirectord actually work. You could be having network issues because you are hosting it on a virtual server instead of on a real server. So it's kinda hard to troubleshoot if you don't even know if your configs are broken. Get two crappy boxes that you can load everything up on, configure them with heartbeat, get that working where it will failover an IP. then add some other service like ipvsadm/ldirectord, and take things one step at a time. Don't try to setup everything all at once, it makes it harder to try to debug problems.
I'm using CentOS4 and RHEL4 using dag'd rpms on a few of the CentOS and RHEL boxes and built from source on some of the other ones. I haven't had a chance to try out a CentOS 5 system yet. But as to your stability questions, we've been using LVS for about 3 or 4 years now and never, ever had stability problems.
Also, we're on a mix of i386 and x86_64 systems. But for each cluster the pair of nodes is identicle.
On 30/11/2007, Matt Shields mattboston@gmail.com wrote:
Take Xen out of the picture until you learn how heartbeat and ipvsadm/ldirectord actually work. You could be having network issues because you are hosting it on a virtual server instead of on a real server. So it's kinda hard to troubleshoot if you don't even know if your configs are broken. Get two crappy boxes that you can load everything up on, configure them with heartbeat, get that working
Thanks for your suggestion. The reason I use Xen (beyond the huge convenience) is that I don't have spare hardware to play with.
where it will failover an IP. then add some other service like ipvsadm/ldirectord, and take things one step at a time. Don't try to setup everything all at once, it makes it harder to try to debug problems.
That's exactly what (I think) I did - just stuck to instructions from someone who seems to have been in exactly the same position and got it working.
As for network issues - I see the packets coming and going all right. But I also see programs just crash and burn - I've just executed BasicSanityCheck on the primary node which appeared to be working relatively fine a couple of minutes ago (at least it got more processes running after three minutes than the other node) and that failed too with core dumps.
I'm using CentOS4 and RHEL4 using dag'd rpms on a few of the CentOS and RHEL boxes and built from source on some of the other ones. I haven't had a chance to try out a CentOS 5 system yet. But as to your stability questions, we've been using LVS for about 3 or 4 years now and never, ever had stability problems.
So maybe I should try to get packages from dag, even though there are ones included in CentOS?
Which exact version of hearbeat are you using right now? From reading the history of Linux-HA it appears there was a huge change between 1.x and 2.x
Thanks,
--Amos
On Nov 30, 2007 6:53 PM, Amos Shapira amos.shapira@gmail.com wrote:
On 30/11/2007, Matt Shields mattboston@gmail.com wrote:
Take Xen out of the picture until you learn how heartbeat and ipvsadm/ldirectord actually work. You could be having network issues because you are hosting it on a virtual server instead of on a real server. So it's kinda hard to troubleshoot if you don't even know if your configs are broken. Get two crappy boxes that you can load everything up on, configure them with heartbeat, get that working
Thanks for your suggestion. The reason I use Xen (beyond the huge convenience) is that I don't have spare hardware to play with.
where it will failover an IP. then add some other service like ipvsadm/ldirectord, and take things one step at a time. Don't try to setup everything all at once, it makes it harder to try to debug problems.
That's exactly what (I think) I did - just stuck to instructions from someone who seems to have been in exactly the same position and got it working.
As for network issues - I see the packets coming and going all right. But I also see programs just crash and burn - I've just executed BasicSanityCheck on the primary node which appeared to be working relatively fine a couple of minutes ago (at least it got more processes running after three minutes than the other node) and that failed too with core dumps.
I'm using CentOS4 and RHEL4 using dag'd rpms on a few of the CentOS and RHEL boxes and built from source on some of the other ones. I haven't had a chance to try out a CentOS 5 system yet. But as to your stability questions, we've been using LVS for about 3 or 4 years now and never, ever had stability problems.
So maybe I should try to get packages from dag, even though there are ones included in CentOS?
Which exact version of hearbeat are you using right now? From reading the history of Linux-HA it appears there was a huge change between 1.x and 2.x
2.x
Amos Shapira wrote:
Hello,
I'm beginning to give up on making Linux-HA's heartbeat work for my environment (CentOS x86_64) and am wondering what other option have I got to help me:
- Use IPVS to maintain a cluster of virtual servers, either
master/slave or load-balanced. 2. Use DRBD in master/slave fashion to keep a home-grown application highly-available.
The first thing I stumbled upon is RedHat Cluster Suite ( http://www.linuxtopia.org/online_books/centos_linux_guides/centos_cluster_co...), from which I also saw some packages on my CentOS servers.
I've never heard of it before and am just starting to dig its docs, but if someone here can confirm/deny that this is a possible route to take it might save me some time or doubts.
AFAIK, RHCS has at its core a spin of heartbeat.
Amos Shapira wrote:
Hello,
I'm beginning to give up on making Linux-HA's heartbeat work for my environment (CentOS x86_64) and am wondering what other option have I got to help me:
- Use IPVS to maintain a cluster of virtual servers, either
master/slave or load-balanced. 2. Use DRBD in master/slave fashion to keep a home-grown application highly-available.
The first thing I stumbled upon is RedHat Cluster Suite ( http://www.linuxtopia.org/online_books/centos_linux_guides/cen tos_cluster_configuration_and_management/ http://www.linuxtopia.org/online_books/centos_linux_guides/ce ntos_cluster_configuration_and_management/ ), from which I also saw some packages on my CentOS servers.
I've never heard of it before and am just starting to dig its docs, but if someone here can confirm/deny that this is a possible route to take it might save me some time or doubts.
What isn't working with heartbeat?
It may be just that some help getting heartbeat working properly is all you need?
If after working with it fully configured properly it still doesn't work the way you need it then I'd look at another clustering solution.
-Ross
______________________________________________________________________ This e-mail, and any attachments thereto, is intended only for use by the addressee(s) named herein and may contain legally privileged and/or confidential information. If you are not the intended recipient of this e-mail, you are hereby notified that any dissemination, distribution or copying of this e-mail, and any attachments thereto, is strictly prohibited. If you have received this e-mail in error, please immediately notify the sender and permanently delete the original and any copy or printout thereof.
On 30/11/2007, Ross S. W. Walker rwalker@medallion.com wrote:
What isn't working with heartbeat?
Basically it just doesn't start properly on the secondary node.
I describe all my trials and tribulations in the following thread on Linux-HA Users: http://lists.linux-ha.org/pipermail/linux-ha/2007-November/029068.html.
Looking through the archive it seems that I'm not alone and heartbeat is not stable yet.
It may be just that some help getting heartbeat working properly is all you need?
I'd love to get it. I've been chasing this for over a week now (I started this thread a few days after hitting the first problem) with no success.
I'm contemplating trying to switch to i386 to see if it'll help, but that's voodoo workaround, I don't trust a solution where the response from the coders about the problem is just "that's very weird", and I suspect that i386 might not take full advantage of the Dual Xeon 3150+8Gb memory (possibly more soon after) that I plan to have on those nodes.
If after working with it fully configured properly it still doesn't work the way you need it then I'd look at another clustering solution.
I tried the instructions in http://wiki.centos.org/HowTos/Ha-Drbd which seems to be explaining exactly the kind of configuration I want. I got DRBD working alright even before finding those instructions but the heartbeat part is the problem. The first execution of crm_mon on the first node doesn't behave according to the instructions (it takes about 3 minutes to start giving input, and then it only reports the node itself). On the secondary node it doesn't work at all.
Since all I want from it is a tiny part of its complex possibilities (just two nodes, and an active/passive configuration for now), I'd go even for setting up my own scripts though I'd MUCH rather stick to something which comes with CentOS, it would give me a feeling that it was better tested and supported.
ANY tips about anything related to that would be greatly appreciated. I'm just running out of time to set this up (customers are waiting).
Thanks,
--Amos
--Amos
On Friday 30 November 2007, Amos Shapira wrote:
Hello,
I'm beginning to give up on making Linux-HA's heartbeat work for my environment (CentOS x86_64) and am wondering what other option have I got to help me:
- Use IPVS to maintain a cluster of virtual servers, either master/slave
or load-balanced. 2. Use DRBD in master/slave fashion to keep a home-grown application highly-available.
The first thing I stumbled upon is RedHat Cluster Suite ( http://www.linuxtopia.org/online_books/centos_linux_guides/centos_cluster_c onfiguration_and_management/), from which I also saw some packages on my CentOS servers.
I've never heard of it before and am just starting to dig its docs, but if someone here can confirm/deny that this is a possible route to take it might save me some time or doubts.
We are in the middle of migrating to a new colo and I first heard about Cluster Suite with the release of 5.
Our old colo used 2 different 2-node clusters using hearbeat version 1. We had a 2-node cluster in Active/passive for the LVS director and 4 nodes as real servers. Our other 2-node cluster was file servers.
I saw the Redhat Cluster Suite (RCS) and spent 2 weeks trying to implement it- without success. I ran into bugs and couldn't get it to work right.
(Parenthecally let me say this: VERSION 2 ROCKS! With version 1, you are limited to 2 nodes. With 2, as many as you want.)
So I went back to heartbeat and learned version 2. Now, we have a 6-node cluster where ANY NODE can be a REAL SERVER OR a LVS DIRECTOR. It was really cool when I learned how to do it. I spent 2 more weeks learning it BUT I have a solution that works and has been stable since inception. Note that we left the file servers in their own 2 node cluster.
So, in summary, from my experience:
1. forget RCS 2. use Heartbeat in version 2 mode to control both LVS and REAL Server functionality. 3. This will allow you to sleep at night.
Enjoy!
Dave
On 02/12/2007, Dave Augustus davea@ingraftedsoftware.com wrote:
We are in the middle of migrating to a new colo and I first heard about Cluster Suite with the release of 5.
Our old colo used 2 different 2-node clusters using hearbeat version 1. We had a 2-node cluster in Active/passive for the LVS director and 4 nodes as real servers. Our other 2-node cluster was file servers.
I saw the Redhat Cluster Suite (RCS) and spent 2 weeks trying to implement it- without success. I ran into bugs and couldn't get it to work right.
Thanks. That's helpful to know.
(Parenthecally let me say this: VERSION 2 ROCKS! With version 1, you are limited to 2 nodes. With 2, as many as you want.)
Yes I know that heartbeat 2.x should rock - when it runs. But having multiple core dumps on my filesystem doesn't exactly increase my confidence in it.
So I went back to heartbeat and learned version 2. Now, we have a 6-node cluster where ANY NODE can be a REAL SERVER OR a LVS DIRECTOR. It was really
That's my plan - to put both director and "real servers" on the same two nodes. As far as I'm aware it's possible also with version 1.
cool when I learned how to do it. I spent 2 more weeks learning it BUT I have a solution that works and has been stable since inception. Note that we left the file servers in their own 2 node cluster.
Which platform is it? Is it CentOS 5 x86_64 on an Intel Xeon?
I suspect that maybe my problems are connected with this particular architecture.
And possibly a general CentOS question - Is it practical to just install i386 packages of heartbeat on an x86_64 system?
So, in summary, from my experience:
- forget RCS
- use Heartbeat in version 2 mode to control both LVS and REAL Server
functionality. 3. This will allow you to sleep at night.
Enjoy!
Thanks.
--Amos
Amos Shapira wrote:
On 02/12/2007, Dave Augustus davea@ingraftedsoftware.com wrote:
We are in the middle of migrating to a new colo and I first
heard about
Cluster Suite with the release of 5.
Our old colo used 2 different 2-node clusters using
hearbeat version 1. We had
a 2-node cluster in Active/passive for the LVS director and
4 nodes as real
servers. Our other 2-node cluster was file servers.
I saw the Redhat Cluster Suite (RCS) and spent 2 weeks
trying to implement it-
without success. I ran into bugs and couldn't get it to work right.
Thanks. That's helpful to know.
(Parenthecally let me say this: VERSION 2 ROCKS! With
version 1, you are
limited to 2 nodes. With 2, as many as you want.)
Yes I know that heartbeat 2.x should rock - when it runs. But having multiple core dumps on my filesystem doesn't exactly increase my confidence in it.
So I went back to heartbeat and learned version 2. Now, we
have a 6-node
cluster where ANY NODE can be a REAL SERVER OR a LVS
DIRECTOR. It was really
That's my plan - to put both director and "real servers" on the same two nodes. As far as I'm aware it's possible also with version 1.
cool when I learned how to do it. I spent 2 more weeks
learning it BUT I have
a solution that works and has been stable since inception.
Note that we left
the file servers in their own 2 node cluster.
Which platform is it? Is it CentOS 5 x86_64 on an Intel Xeon?
I suspect that maybe my problems are connected with this particular architecture.
And possibly a general CentOS question - Is it practical to just install i386 packages of heartbeat on an x86_64 system?
Your running this under a Xen domU right?
You could try the i386 versions to see, but I wouldn't be surprised if you end up with the same results.
Instead of a domU why not try running it out of an HVM which performs greater level of abstraction. IF heartbeat is trying to use the low level features of the network interface then that may be the reason it is segfaulting in the para-virtualized machine.
So, in summary, from my experience:
- forget RCS
- use Heartbeat in version 2 mode to control both LVS and
REAL Server
functionality. 3. This will allow you to sleep at night.
Enjoy!
______________________________________________________________________ This e-mail, and any attachments thereto, is intended only for use by the addressee(s) named herein and may contain legally privileged and/or confidential information. If you are not the intended recipient of this e-mail, you are hereby notified that any dissemination, distribution or copying of this e-mail, and any attachments thereto, is strictly prohibited. If you have received this e-mail in error, please immediately notify the sender and permanently delete the original and any copy or printout thereof.
-
Amos Shapira -
Dave Augustus -
John R Pierce -
Matt Shields -
Robert - elists -
Robert Slade -
Ross S. W. Walker