Question about Cluster/DSync/Load Balance

According to the admin doc of iplanet, primary server is
the "manager" for data sync, is there any impact on
load balance when the iAS run as primary or backup?
will the primary kxs get the request first and do dispatching?
Thanks.
Heng

First of all lets discuss load balancing....
The type of load balancing you are using will determine which process manages the load balancing. If you are using Response time (per server or per component response time) or round robin (regular or weighted) the web connector does the load balancing. If you are using User Defined (iAS based) load balancing then the kxs process becomes involved with load balancing of requests since the "Load Balancing System" is part of the kxs process.
Now for Dsync and how it impacts load balancing.
When a server is a sync primary or a sync backup role it is doing more work. For the sync primary the extra work is making sure the backup has the latest Dsync Data and processing requests from the other servers in the cluster about the Distributed data. All state/session information is updated/created/deleted on the sync primary, when this happens the sync primary immediately updates the sync backup(s) with this new information. As you can guess managing the Dsync information and making the updates to the sync backups causes extra processing on the sync primary, so this will impact the overall performance of the machine (whether it be in server load or response time of processing). All lookup of state/session information is done on the sync primary only so the more lookups/updates you have to more impact on the server.
The sync backup(s) also have the extra work of managing their copy of the Dsync Data which will impact server performance but to a lessor degree of the sync primary.
Ultimately the extra overhead involved does have an impact on loadbalancing due to the extra load on the sync primary and sync backups.
Hope that helps,
Chris Buzzetta

Similar Messages

CF8/JRun4 Cluster for Load Balancing

Does anyone have an example of how to set up a CF8/JRun4
cluster for load balancing?
I have three servers:
x004 - Linux - Apache2 (10.0.0.54,10.1.0.54)
x020 - Linux - JRun4/CF8 (10.0.0.70,10.1.0.70)
x021 - Linux - JRun4/CF8 (10.0.0.71,10.1.0.71)
Every server in our network has two network cards. One
network card is attached to 10.0.x.x which has a gateway to the
internet and runs at 100Mbps and is firewalled, and the other is
attached to 10.1.x.x which runs at 1Gbps and is internal with no
gateway. I'm trying to set it up so web traffic arrives on
10.0.0.54 into Apache and mod_jrun20 bootstraps a cluster named
STST using 10.1.0.54 which consists of STST_x020 coldfusion server
running on x020 and STST_x021 running on x021. I want the
communications between JRun4 on x020 and x021 to occur on the
10.1.x.x network and eventhough JRun and ColdFusion will only use
the 10.1.x.x network I still need the 10.0.x.x network card
attached for other purposes which require a gateway. I have
installed JRun4/CF8 about 10 times already and it seems I have no
control over what network JRun4 clusters on... sometimes it will
communicate on one, sometimes the other and without being able to
set which network is being used there always seems to be "network
error" on at least one of the two CF8 servers. I was able to get
everything working fine by disabling the network cards on the
10.0.x.x network and re-installing everything... but as soon as I
added the network cards back the whole thing was broken again.
How is this supposed to work? Most of the examples are either
no clustering or clustering on the same machine with Apache running
on the same box... I don't see any clustering across machines
examples.
How do I install a connector on a web server which doesn't
have JRun on it and get wsconfig to connect to a multi-machine
cluster when wsconfig only accepts a single IP address as a host
and the cluster is not listed?
How do I get JRun to bind to a specific network card?
Does this work if I choose a J2EE server other than JRun?
Any help anyone can provide is greatly appreciated. I'm
getting close to giving up which means staying on the non-clustered
environment and figuring out how we can deal with scalability by
switching to something else.

The article at
http://www.adobe.com/go/1e8e9170
is specific to configuring two or more cluster nodes that reside on
separate networks, e.g. 10.0.1.0/24 and 10.0.2.0/24. (The article
doesn't state it, but you can only use unicast peers if your
cluster nodes host a single instance of JRun or multiple instances
of JRun in the same cluster domain. When performing unicast
discovery, JRun looks for all Jini groups and not just the cluster
group.)
Anyhow, that's not your problem. The simplest solution is you
haven't enabled the jrun.servlet.jrpp.JRunProxyService service. I'm
most familiar with the Windows version of JRun, but I'm assuming
the directory structure is similar across platforms. In
<jrun_root>/servers/<name>/SERVER-INF/jrun.xml, set the
deactivated attribute of the jrun.servlet.jrpp.JRunProxyService
service to false and restart JRun. You should now see JRun
listening on the appropriate port. (The default for the first
manually created instance is 51000.) You can limit the proxy
service to a single interface using the interface attribute.
If you have enabled the proxy service, verify your security
settings in <jrun_root>/lib/security.properties. It's usually
best to limit access to specific hosts. Comment out the
jrun.subnet.restriction parameter and set the jrun.trusted.hosts to
the IP address of your web server, e.g. 10.1.0.54.
Forcing all JRun processes/services to listen on a single
interface isn't difficult, but it does require modifying quite a
few configuration files by hand. If you need assistance with that,
I can elaborate.
Configuring the JRun module under Apache is pretty
straightforward. If you're not using virtual hosts, it's very
simple. If you are using virtual hosts, it's still simple, but your
JRun configuration can be virtual host-specific.
On your Apache server, you'll want to create a directory
structure for the JRun module. I'll assume
/opt/jrun/lib/wsconfig/1, but you can use anything you want. Once
the directory structure is created, extract the appropriate JRun
module from wsconfig.jar to the new directory. You're most likely
interested in the Apache 2.0 module,
wsconfig.jar/connectors/apache/intel-linux/prebuilt/mod_jrun20.so.
Let's assume you've extracted the module to
/opt/jrun/lib/wsconfig/1/mod_jrun20.so. Your Apache service account
should have read, write, and execute permissions on the
/opt/jrun/lib/wsconfig/1 directory.
The JRun module configuration is normally appended to your
current httpd.conf file by wsconfig. Here's a sample configuration:
LoadModule jrun_module
"/opt/jrun/lib/wsconfig/1/mod_jrun20.so"
<IfModule mod_jrun20.c>
JRunConfig Verbose false
JRunConfig Apialloc false
JRunConfig Ssl false
JRunConfig Ignoresuffixmap false
JRunConfig Serverstore
"/opt/jrun/lib/wsconfig/1/jrunserver.store"
JRunConfig Bootstrap 10.1.0.70:51000
#JRunConfig Errorurl <optionally redirect to this URL on
errors>
#JRunConfig ProxyRetryInterval 600
#JRunConfig ConnectTimeout 30
#JRunConfig RecvTimeout 30
#JRunConfig SendTimeout 30
AddHandler jrun-handler .jsp .jws .cfm .cfml .cfc .cfr
.cfswf
</IfModule>
You may also want to update your DirectoryIndex directive
with an appropriate index page, e.g. index.cfm.
After the first request to a page handled by the JRun module
is received, the module will query the boostrap server,
10.1.0.70:51000, for a list of cluster peers. If you've configured
your cluster correctly, a line similar to following will be written
to /opt/jrun/lib/wsconfig/1/jrunserver.store:
proxyservers=10.1.0.70:51000;10.1.0.71:51000
You can create/edit this file manually as well.
Unfortunately, the bootstrap option only accepts one server. If
your bootstrap server is down, the JRun module will use the values
in jrunserver.store directly, if the file exists.
Here's a complete list of JRun module options:
metrics *
debugger *
ssl *
verbose
traceflags
serverstore
bootstrap
errorurl
apialloc
ignoresuffixmap
proxyretryinterval
connecttimeout
recvtimeout
sendtimeout
sslcalist
Options flagged with an asterisk can only be configured at
the Apache server level. All other options can be configured at the
server level and/or the virtual host level. The usage of these
options is in the JRun documentation, and the JRun module source
code is included in wsconfig.jar. Keep in mind that versions of the
JRun module shipped prior to ColdFusion 8 were coded to assign the
connecttimeout and sendtimeout options to the socket connection
timeout. Whichever option appeared last in your configuration ended
up as the final value. This has been fixed in ColdFusion 8 and
presumably the next release of the JRun updater.
I think that's a good start. If you need more information or
can't find what you need in the JRun or ColdFusion documentation,
let me know.
If you're looking for resiliency, I highly recommend
expanding your configuration to include a second web server and a
hardware load-balancer (preferably one that supports redudancy via
multiple paths and devices, e.g. devices from Cisco, F5, or Foundry
Networks). Often, however, running Apache on the ColdFusion
server(s) provides adequate performance, and round-robin DNS
records coupled with the ability to update DNS quickly in the event
of a failure may be all you need for load-balancing and
failover.

OCS on a cluster with Load balancing and fail safe environment

Dear all,
i want to ask is there any documat or hints on how to do an OCS R2 installaiotn on 3 server with RAC option (clustered Fail Safe), how can i install OCS on a cluster with Load balancing and fail safe environment.
plz i need ur help
thanking u
[email protected]

Dear all,
i want to ask is there any documat or hints on how to do an OCS R2 installaiotn on 3 server with RAC option (clustered Fail Safe), how can i install OCS on a cluster with Load balancing and fail safe environment.
plz i need ur help
thanking u
[email protected]

How to setup Adobe Media Server Professional x 2 run as cluster for load balance?

How to setup Adobe Media Server Professional x 2 run as cluster for load balance?

Hi,
Welcome to adobe forums,
Please refer to these help files in order to setup AMS as a cluster : https://helpx.adobe.com/adobe-media-server/config-admin/load-balancing.html
https://helpx.adobe.com/adobe-media-server/tech-overview/scaling-server.html
Let me know if you need any help.
Regards,
Puspendra

OAM 11g integration with Kerberos on cluster with load-balanced virtualhost

Hello!
I need to make a Kerberos integration with OAM.
I find following notes about OAM 11g: WNA Configuration for HA Clusters [ID 1365888.1] (https://support.oracle.com/epmos/faces/ui/km/SearchDocDisplay.jspx?_afrLoop=223640518878014&type=DOCUMENT&id=1365888.1&displayIndex=1&_afrWindowMode=0&_adf.ctrl-state=14ehvbh4z2_61).
"In an OAM Clustered environment, the OAM Principal for WNA must be the same on all tiers i.e. the load-balanced virtualhost for the OAM cluster.
Therefore each OAM managed server will reference the same keytab file, generated for Principal HTTP/<virtualhost.domain>, and the keytab file will be in the same location on all OAM managed servers.
For example: ${DOMAIN_HOME}/domains/${DOMAIN_NAME}/config/fmwconfig/oam/<keytab filename>.
After copying the keytab file to the same directory on all OAM managed server machines, complete the configuration of the Kerberos authentication module in OAM Administration Console (/oamconsole).
The AdminServer will ensure that the oam-config.xml file on all OAM managed server tiers in the cluster is updated with this configuration."
The question is; When I generate oam.keytab with following command, What is the name of the server that I will must put in the command? Virtualhost (load-balanced), Node1 or Node2?
ktpass -princ HTTP/<servername>@DOMAIN -pass XXXXXXX mapuser DOMAIN\user -out oam.keytab.
Thanks in advance and best regards!
PS: Sorry if my english is not clear.

David,
Your Principal name should be the SSO LB URL.(ie :sso.mycomany.com)
ktpass -princ HTTP/sso.mycomany.com@DOMAIN -pass XXXXXXX mapuser DOMAIN\user -out oam.keytab.
Also make sure sso.mycomany.com has a reverse DNS configured correctly.
you can check using dig command
ping sso.mycomany.com
What ever the ip-address
dig -x <IP-ADDRESS>
Check in the reverse DNS section there should be 1 record.
;; ANSWER SECTION:
1.1.1.1.in-addr.arpa. 3600 IN PTR sso.mycomany.com.
Let me know if you have more questions.
Thanks
Saurabh

OBPM Enterprise Deployment on WLS - No Cluster, But Load Balanced

All,
Does anyone know of any gotchas when deploying BPM to WLS on 2 separate nodes, sharing the same directory, but not clustered. The system is load balanced based on F5. Basically we are talking a hot server/cold server deployment.
When we deploy projects, they default to the hot server even if the cold server is specified for deployment.
Anyone done this before?
TIA,
IGS

Hi,
Sorry, but I could not understand completely your architecture.
Are you talking about the Workspace (not clustered but load balanced). That's supported.
Or, are you trying to load balance the engine? (a single engine with 2 or more nodes)
If so.... I wouldn't recommend that you to do that.
Let me explain you why.
The engine uses the queue to balance the work among the different nodes. (that's why you have to configure a Distributed Queue and disable the server affinity in the connection factory).
Even more, the engine has some internal mechanism of synchronization among nodes so as to avoid some inter-node locking. If your engine nodes are not in a cluster, that mechanism will be disabled and the overall engine performance will be significantly degraded.
I'm not sure if I have answered your question. If not, please add more details of your configuration.
Hope this helps,
Ariel

Question on how does load balancing work on Firewall Services Module (FWSM)

Hi everyone,
I have a question about the algorithm of load balancing on Firewall Services Module (FWSM).
I understand that the FWSM supports up to three equal cost routes on the same interface for load balancing.
Please see a lower simple figure.
outside inside
--- L3 SW --+
|
MHSRP +--- FWSM ----
|
--- L3 SW --+
I am going to configure the following default routes on FWSM point to each MHSRP VIP (192.168.13.29 and 192.168.13.30) for load balancing.
route outside_1 0.0.0.0 0.0.0.0 192.168.13.29 1
route outside_1 0.0.0.0 0.0.0.0 192.168.13.30 1　　　　　　
However I don't know how load balancing work on FWSM.
On FWSM, load balancing work based on
Per-Destination ?
Per-Source ?
Per-Packet ?
or
Other criteria ?
Your information would be greatly appreciated.
Best Regards,

Configuring "tunnel default gateway' on the concentrator allowed traffic to flow as desired through the FWSM.
FWSM is not capable of performing policy based routing, the additional static routes for the VPN load balancing caused half of the packets to be lost. As a result, it appears that the VPN concentrators will not be able to load balance.

OCS 10g Cluster Installation - Load Balancing

Hi all,
Anybody have sucessfully install and configure OCS 10g Cluster ?, with load balancing ?
I'm trying to install OCS 10g cluster with a two node server setup, and using Oracle web Cache as the load balancer, but not sucessfull. Any hints ?
Regards
Lanang

Just found out that Oracle Web Cache support HTTP and HTTPS only, no LDAP traffic yet. That why the cluster node installation failed. Trying using iptables NAT for the LDAP traffic, and the HTTP will use web cache.
Regards
Din

A question about cluster

Hi all,
I have a question about the usage of cluster. I create a cluaster
which includes three indicators on the front panel, but I can't use
unbundle function on the diagram. If I change the indicator to control,
it works. I didn't find on the manual that the unbundle function only
accept cluster of control as input.
Thanks for your help!
Regards,
Tao

This is because of the way dataflow programming works. It probably didn't
explicitly mention.
Labview has data sources (controls, inputs) and data sinks (indicators,
outputs). Unbundling accepts cluster as input and outputs the individual
pieces of data. Similarly bundling accepts individual pieces of data as
input and ouputs a cluster. So logically your cluster must be a control
(input) for unbundling, and an indicator (output) for bundling.
If I'm misunderstanding the question or oversimplifying the problem email me
or post again.
-joey
"tsong" wrote in message news:b11154$qmq$[email protected]..
> Hi all,
>
> I have a question about the usage of cluster. I create a cluaster
> which includes three indicators on the front panel, but I can'
t use
> unbundle function on the diagram. If I change the indicator to control,
> it works. I didn't find on the manual that the unbundle function only
> accept cluster of control as input.
>
> Thanks for your help!
>
> Regards,
>
> Tao
>
>
>

Cluster not load-balancing, ideas?

I've been struggling to identify why my JMS producers are not load-balancing against a remote cluster.
          I've ruled out the producer as being the problem (I see the same non-load-balancing behavior regardless of what I use to create messages - Hermes, ALSB, simple Java producer...) I also don't think the JMS Connection Factory config is the problem, judging by the help I've received from folks over on the jms forum.
          I believe something is wrong with our cluster setup because in addition to the problem I just mentioned, we also are not seeing JNDI entries propagate to all managed servers - for example, if I create one jms queue on m1, that queue does not appear in the jndi tree on m2.
          I've been trying to find any documentation on what settings I should look at to verify the cluster configuration. If I go through the WLS console and look at the Cluster settings, I see both managed servers there, is there some other place that the configuration could be messed up?
          Added 6/11, 9:30 am:
          We're focusing on multicast now as the most likely problem. Can anyone tell me whether clusters on the same multicast address but different ports will interfere with each other? It looks like the infrastructure team has set up 5 clusters like that (same multicast address in each cluster, but different ports).
          We've got a ticket open with BEA but it's been two weeks now and nothing except requests for more information.
          Any ideas/help are much appreciated!
          Meghan
          Edited by pietila at 06/11/2008 7:38 AM

Meghan Pietila wrote:
          > I've been struggling to identify why my JMS producers are not load-balancing against a remote cluster.
          >
          > I've ruled out the producer as being the problem (I see the same non-load-balancing behavior regardless of what I use to create messages - Hermes, ALSB, simple Java producer...) I also don't think the JMS Connection Factory config is the problem, judging by the help I've received from folks over on the jms forum.
          >
          > I believe something is wrong with our cluster setup because in addition to the problem I just mentioned, we also are not seeing JNDI entries propagate to all managed servers - for example, if I create one jms queue on m1, that queue does not appear in the jndi tree on m2.
          >
          > I've been trying to find any documentation on what settings I should look at to verify the cluster configuration. If I go through the WLS console and look at the Cluster settings, I see both managed servers there, is there some other place that the configuration could be messed up?
          >
          > Added 6/11, 9:30 am:
          > We're focusing on multicast now as the most likely problem. Can anyone tell me whether clusters on the same multicast address but different ports will interfere with each other? It looks like the infrastructure team has set up 5 clusters like that (same multicast address in each cluster, but different ports).
          >
          > We've got a ticket open with BEA but it's been two weeks now and nothing except requests for more information.
          >
          > Any ideas/help are much appreciated!
          >
          > Meghan
          >
          > --
          > Edited by pietila at 06/11/2008 7:38 AM
          You could be right. I think we have had problems where the same IP but
          different ports were used for multicast. This is on 8.1 though.
          I think as a rule, it's best to have a different ip and port for each
          cluster.
          Also - can you be sure that no one else is using the multicast addresses
          on the network for anything else - we had someone bring up a test
          cluster using our addresses which caused a few issues and took a while
          to find! We also have security cameras which also use multicast, which
          if they are using the same address/port can cause issues!
          We're using 239.192.1.4:8001 for one cluster and 239.192.1.3:7001 for
          the other - I think it's best to keep those as different as you can.
          In 8.1, there is also the multicast monitor utility - there's a support
          pattern on e-support on how to diagnose it. I've found this useful in
          the past when I've suspected a cluster issue.
          https://support.bea.com/application_content/product_portlets/support_patterns/wls/MulticastErrorsPattern.html
          Check also that you're using a valid range for the address - we weren't
          for a while and had odd problems from time to time.
          There are also cluster debug flags available which you'll see listed in
          the support document.
          Are you seeing dropped multicast packets?
          Hope that helps.
          Pete

Cluster without load balancer?

Can two FMS interactive servers work this way:
1. they both serve the same VOD flv file existing on both
machines
2. whe one server gets X users connected, next user (X+1) is
routed to the next server
3. there is no hardware for the load balancer
Is this possible and if it is, how?

Yes.... it's possible, but there's nothing built-in to FMS to
handle it. You need to write your own application to do it.
I like to handle this by building a little app that polls the
admin service of each involved server once a second or so, and
retrieves stats about the application instances i need to keep
track of (we'll call it the "load balancer app" . In this
application, I include functions to loop through the stats data for
each server, and determine which is the most logical to send the
next client to.
On the client side, I first connect to the load balancer app,
providing an identifier for the application I want to connect to as
an argument in the connect() call. The load balancer makes the best
server determination, and returns the host name of the target
server. The client then disconnects from the load balancer, and
connects to the target host.

Question about free down loading Mavericks. On checking Mavericks before downloading that there are no Bangla language in the list. I occasionally use Bangla in my present Mountain Lion. My concerns if I down load Mavericks then what will happen to Bangla

Question about free downloading Mavericks. On checking Mavericks before downloading that there are no Bangla language in the list. I occasionally use Bangla in my present Mountain Lion. My concerns is, if I download Mavericks then what will happen to Bangla language in mountain lion? Samar Laha

No reason to be embarassed. You are not alone.
You will need an external drive. If you have a Time Machine drive without enough disk space you can create the clone there however it's advised that you keep your Time Machine and clone on separate drives. If the drive dies you loose both types of backup.
Download and install the clone software. Everyone has their favorite. These are the two most common applications used. I prefer SuperDuper because of it's simplicity in setting up. It's also free for the first clone where with CCC you have to give a Credit Card then cancel if you do not want to keep.
SuperDuper! http://www.shirt-pocket.com/
CCC http://www.bombich.com/download.html
Select your drive in the To popup. Click on Copy Now. That's it.
Yes, it really is that simple....

Weblogic cluster software load balancer

Hi,
We are currently using Weblogic domain as a Proxy Plug-In for High Availability test as it’s explained in this blog http://andrejusb.blogspot.com/2009/04/weblogic-load-balancing-for-oracle-adf.html.
Its working fine for POC project but what software load balancer would you recommended for production environment on Linux? (Assume that we don’t have a hardware load balancer).
- Oracle active-passive OHS web-tier clustering.
-Using Linux open source Linux Software(e.g. HAProxy and KeepAlived as explained here http://biemond.blogspot.com/2010/04/high-availability-load-balancer- for.html
-Using any other software load Balancer
I would appreciate if anybody can provide some recommendations/links etc.
Thanks
Alex

Hi Alex,
Yes you should never use HttpClusterServlet, not even for fun ;-)
We use mod_wl (Web Server Plug-In) for Apache for several customers and that works fine.
Check this:
http://docs.oracle.com/cd/E23943_01/web.1111/e14395/toc.htm
Regards Peter

Distributed HA cluster with load-balancing and failover: advice?

My workplace has a Xeon Xserve, which acts as our primary external server, with an attached ActiveStorage XRAID. We have just purchased a second Xserve/XRAID set to act as a mirror, which we will colocate. Both have Leopard Server installed, along with an array of additional software.
What we want to do is have both servers load-balance between the two, with failover in case of a server or XRAID fault. I plan on using RSYNC to mirror static files between the two, and I'm looking into PostgreSQL replication and load-balancing solutions for our database. I gather that Apache supports web-server failover and load-balancing, as well. But, that still leaves the actual host and network setup to arrange.
Does Leopard server support such a thing? The only information I found on IP failover instructs the user to place the two servers on the same subnet, directly connected via ethernet cable; obviously, this would not work in my case.
Ideally, what we'd end up with is a situation in which the two systems kept each other in sync, both in static files and database data, and load-balanced between themselves; in cases of failure, the remaining system would transparently assume all duties until the other was restored, at which time they would resynchronize
Any suggestions on how I could arrange such a thing?

Interesting. Does this DNS-based approach support session tracking, though? I would need to have a user directed to just one of the two servers for the duration of their session, to avoid having to synchronize temporary files and such.
You can't have it both ways. You need to build tolerance into the app.
At the simplest level where you run all traffic to one site and use the second site as a failover/standby site you'll be OK most of the time - all users will go to the same server and their sessions will be intact.
However, under any failover situation (your primary site is down for some reason), there is going to be some level of session traffic that it going to switch over to the other site. If your site depends on sessions then you're going to need to tolerate this kind of situation - your app will need to be able to fail gracefully if a user comes in with an invalid session cookie.
Note, though, that this may be less of an issue than you at first think - all DNS clients will cache DNS data for whatever TTL you set. This means that if a user looks up your site name and you return an IP address with a 30 minute TTL, then that user is going to use the same IP address for the next 30 minutes and isn't going to ask the server again. This should negate most chance of a user suddenly switching from one server location to the other in mid-session.
The trick comes in setting the DNS TTL low enough to effect a failover, yet not so long that you impact performance - e.g. you don't want the user to perform a DNS lookup on every page load. You may find that 10 minutes is appropriate. Just bear in mind that this affects how long a user could see your site 'down' before the failover DNS kicks in. Clearly you don't want to set the DNS TTL to a day since that may prevent the user switching to the secondary site for 24 hours by which time, hopefully, the primary site is back up, anyway.
The 'right' TTL value may take some analysis on your traffic to see how long a typical user 'session' is. If the average user spends 20 minutes on your site, then it would make sense to set your TTL to somewhere around 20 minutes to give the best chance of their entire session staying on the same server.

Question about cluster node majority voting

We've been having problems with a DB instance crashing regularly. This weekend when it crashed, it seems to have taken the node it was on with it, or this was a separate incident...
Right now I have 3 nodes in the cluster. 2 nodes are running 3 instances (2 on 1). The 3rd node is in a state where the OS is mostly unusable and the Cluster service will not start.
Event Log:
"The failover cluster database could not be unloaded. If restarting the cluster service does not fix the problem, please restart the machine."
Cluster Log from that machine:
00003768.000067a0::2014/01/06-03:28:05.393 INFO -----------------------------+ LOG BEGIN +-----------------------------
00003768.000067a0::2014/01/06-03:28:05.393 INFO [CS] Starting clussvc as a service
00003768.000067a0::2014/01/06-03:28:05.394 INFO [CS] cluster service logging level is 2
00003768.00004c30::2014/01/06-03:28:05.521 DBG   [NETFTAPI] received NsiInitialNotification
00003768.00004c30::2014/01/06-03:28:05.523 DBG   [NETFTAPI] received NsiInitialNotification
00003768.000031f4::2014/01/06-03:28:05.588 DBG   [NETFTAPI] received NsiAddInstance for 169.254.3.47
00003768.00004eb4::2014/01/06-03:28:05.590 ERR   [DM] Error while restoring (refreshing) the hive: STATUS_INVALID_PARAMETER(c000000d
00003768.00004eb4::2014/01/06-03:28:05.592 ERR   [DM] mscs::DmAgent::Start: STATUS_INVALID_PARAMETER(c000000d' because of 'Load(NOTHROW(), securityAttributes, discardError )'
00003768.00004eb4::2014/01/06-03:28:05.592 ERR   [DM] Node 3: failed to unload cluster hive, error 87.
00003768.00004eb4::2014/01/06-03:28:05.592 ERR   Hive unload failed (status = 87)
00003768.00004eb4::2014/01/06-03:28:05.592 ERR   FatalError is Calling Exit Process.
This is a 3 node cluster set to node majority, I don't have an available drive letter for a witness disk. Since the cluster service won't start, I'm not certain how the cluster is still running, but am thankful that it is.
A reboot might fix everything, but I'm very worried that if I reboot the server, and the cluster service still fails to start... it may prevent the entire cluster from starting and we won't be able to run the instances on the other 2 nodes.
Does the 3rd server still act as an odd-number server, even if the cluster service won't start? If I reboot and the cluster service still fails to start, will the cluster itself be able to be in an UP state and run the DB instances on the other nodes?
I already need to open a MS Support incident on the DB instance crashing, so I'd rather not have to open a 2nd one just to answer this hopefully simple question.
Thanks in advance!
Mark

I'll answer it here, since it matters fundamentally to SQL High Availability.
There are a couple of entities you are conflating here, leading to much confusion. There is a difference between the Cluster and the cluster service.
The cluster service will run on a node once the Failover Cluster Feature is installed on that node. The cluster service will run, even if a cluster is not created. It may generate errors and not participate in a Cluster if it cannot talk to the
other nodes, but it will not shut down.
The Cluster itself requires a quorum, that is a majority of votes, in order to operate. With three nodes, you should choose Node Majority quorum model, which sounds like what you have. Any two votes will count, so the third node being offline
does not matter. You can safely restart the cluster service on the failed nod, and even restart the node. Note that with the third node down, you have no redundancy. (Windows 2012 and 2012 R2 have dynamic quorum, which adjusts the quorum
count based on the last "settled" quorum vote, but that doesn't apply here).
I am concerned with your statement that you are out of drive letters. With three instances, you should have plenty of drive letters left. I suggest investigating Mount Points. You only need one drive letter per instance when using Mount
Points.
Geoff N. Hiten Principal Consultant Microsoft SQL Server MVP

Question about Cluster/DSync/Load Balance

Similar Messages

Maybe you are looking for