cancel
Showing results for 
Search instead for 
Did you mean: 

vcs node

linuly
Level 4

hi

i like to clarify.

when a node join to vcs then in its memory all the information from the other nodes are filled.thus this node sees if there he should bring up the resource group that normally start on it.when he leave the cluster the information is updated on remaining cluster nodes and the service group fail over.right?

tnx

1 ACCEPTED SOLUTION

Accepted Solutions

frankgfan
Moderator
Moderator
   VIP   

Hello,

In CVM configuration, any vx* commands which will lead to a vxvm configuration change (commjands like vxassist, vxmake, vxsd, vxplex etc) need to be executed on the CVM master (with new releases from 6.2, the command can be issued on any nodes as now there is a GAB port y which can ship commands to the cvm master for execution).

VCS commands like hares, hagrp, haclus etc do not change VxVM metadata so can be executed on any nodes in the cluster, so long as the node is in online running state in the cluster.

Frank

View solution in original post

10 REPLIES 10

RiaanBadenhorst
Moderator
Moderator
Partner    VIP    Accredited Certified
It works like this.

There is a special network between the cluster nodes which is called llt. This network allows another protocol called gab to keep all cluster nodes up to date with the status of all cluster resources. Any status changes and configuration updates are distributed via gab.

When a node starts up it joins network and check if anybody else is online. If there are enough nodes online it gets the configuration from them and decides based on the rules in the main.cf if it needs to online anything (based on the current state in cluster).

If it finds that nobody is online it has to wait for enough nodes to join it. When they do, it distributes the configuration to the others and then again based on the rules, they online the resources.

For nodes leaving, or resources failing, the state is again distributed to all nodes and based on the rules in the configuration they decide where to online resources

Thats it at a high level

frankgfan
Moderator
Moderator
   VIP   

Hello,

 

when a node starts and if there is another node in running state, the node will take a snapshot of cvs configuration from the running node via llt network.  you can see this info from engine_A.log.  if there is not any node in running state, the node will read on disk vcs configuation (main,cf) and buid up the cluster.

 

if a node is shutdown, hastop command will be executed as part of system shutdown,  since a system shutdown is considered as a system admin action, not a true failure., vcs service groups may not be switched over.

 

here are some hastop command options and their meanings:

    The -local option stops the VCS server on the system on which the command is issued. When the VCS server stops a system, the system’s service groups are also stopped.

   The -force option stops the VCS server, but allows the service groups to continue running.

   The -evacuate option migrates all groups to other systems, if possible.

   The -noautodisable option ensures that service groups that can run on the node for which hastop is invoked are not autodisabled while VCS is stopped. The -noautodisable option can also be used with -evacuate option, but not with -force option.

please also read this technote https://support.symantec.com/en_US/article.TECH11477.htm to know some vcs behavior in a system restart.

Frank

 

Hi,

In vcs cluster an designated coordinator is elected.When the cluster is running every hagrp,hares including resource switch  so on command will talk to DC.right?

tnx

 

frankgfan
Moderator
Moderator
   VIP   

Hello,

there are two types VCS configurations:  failover and CVM (cluster volume manager, also known as parallel configuration).  In CVM configuration there is one node elected as a CVM master.

In failover configuration (a service group is onlined on one node at any given time) there is no master node.

Can you please elaberate the followings?

1. an designated coordinator 

2. DC

is your cluster running on Windows platform or Unix?

Frank

hi,

it was my mistake.Unix i run.

In vcs case if i add or modify a resource from any node with hares then the main.cf of every node is updated and data in memory the same.right?

hares command from the node i given is copied in every node and in every node hares talk to had and this write to main.cf.right?

tnx

 

frankgfan
Moderator
Moderator
   VIP   

Hello,

In CVM configuration, any vx* commands which will lead to a vxvm configuration change (commjands like vxassist, vxmake, vxsd, vxplex etc) need to be executed on the CVM master (with new releases from 6.2, the command can be issued on any nodes as now there is a GAB port y which can ship commands to the cvm master for execution).

VCS commands like hares, hagrp, haclus etc do not change VxVM metadata so can be executed on any nodes in the cluster, so long as the node is in online running state in the cluster.

Frank

hi,

cvm need a master node in order to make sure that the resources are not corrupted and access to them  is synchronized from the nodes that share them.right?

the master node  is not necessary to keep the cluster in sync when a node join.

right?

tnx

CliffordB
Level 4
Employee

There have been good answers, but I want to make a few clarifications:

In VCS (outside of cluster file system), there is no "master" node.  All servers that participate in the cluster have identical state status for all resources.  All VCS commands are atomic.  Either all servers get updated successfully, or the command falis.  VCS is a state machine that works across mulitple separate operating system instances.   Each OS has a complete picture of the state of all servers participating in the cluster.  

As host enter and exit the cluster, all active nodes are updated.  Every time a resoruce or service group is changed: turned on or off or moves, created or deleted, all nodes are notified.  Again, either all nodes are updated, or the change fails.

 

Regarding behavior of ServiceGroups when a node joins or leaves the cluster:  the behavior is dictated by several paramenters.  As a default, when a node fails, its ServiceGroups are moved to other existing cluster members.  How the ServiceGroups are distributed is beyond this discussion.  The Admin guide give very detailed examples on how to control behavior of ServiceGroups when a node leaves on purpose (admin shutsdown a server for maintenance) versus a hardware/softawre failure.

Cluster Volume Manager does have a master/slave system for lock managment only.  If you are not using CFS, then you don't need to consider the idea of master/slave.


_______________________________
My customers spend the weekends at home with family, not in the datacenter.

frankgfan
Moderator
Moderator
   VIP   

CVM master node is responsible for making and recording hanges made to VxVM configration, populate and update the new VxVM configurations (changes) to the slave nodes.

Here is an example.

there is a CVM shared dg called datadg with two volumems vol1 and vol2.

 

If the system admin adds a new volume called vol3 (500G) on the cvm master node systemA to the datadg, the other nodes would not have this updated VxVM configuration (a 500G vol3) info in memory.  the CVM master node will communicate these upadted info to other nodes via GAB ports u and w so that all the nodes know there is now a new volume called vol3 with capacity of 500G.

 

Until now the new VxVM vol3 is not in VCS configuration so its not a cluster resource yet.  You actually do not have to make the new vol3 as a vcs resoruce.  You do not even have to have VCS (HA) to run CVM.  (as this is beyond the scope of this post, I will just stop here ...).

Cheers,

 

Frank

 

 

 

hi,

Tnx Frankgfan and Riaan.I posed this question on Saturday.I have not expected somebody to reply on weekend.The people were very kind and replied me straightaway and very in depth.

You Clifford just shows me how much you are uneducated  arrogant and so on.It were better if you abstained to reply.Guys like you i met very often.You can not understand to be polite.

"My customers spend the weekends at home with family, not in the datacenter."