cancel
Showing results for 
Search instead for 
Did you mean: 

resource

linuly
Level 4

hi,

every time when a resource status change this information will be sent over the private link to all nodes.

hares -online res5 -sys test5 ->this command will online res5 on node test5 then this information will be updated on the cluster node.right?

tnx

1 ACCEPTED SOLUTION

Accepted Solutions

snockles
Level 3

Short answer, yes. Easy to see by running hastatus (not -sum) on one node, and changing the state of a resource on another. The output from hastatus updates almost instantly.

Longer answer. VCS can be regarded as a replicated state machine. in other words all nodes show the same state. There is regular communication, regarding the configuration and state of all the components, between HAD on all the nodes. It is part of the role of GAB to make sure that all the currently cluster members get this information, since GAB is the membership manager in the configuration. GAB uses LLT to actually pass the messages.

If you run any kind packet sniffer on the private networks you'll generally see two types of packets, small broadcasts, which are the heartbeats, and larger addressed ones, which are the state/configuration information being passed around. And while the heartbeats are not acknowledged directly, GAB on the node where a state change has happened, expects a response from the other cluster nodes so it can confirm that everyone has the same information.

In theory, although I've never seen this happen, if a node doesn't acknowledge a state change, it will be ejected from the cluster, because GAB on the other nodes will assume there's a major problem.

View solution in original post

1 REPLY 1

snockles
Level 3

Short answer, yes. Easy to see by running hastatus (not -sum) on one node, and changing the state of a resource on another. The output from hastatus updates almost instantly.

Longer answer. VCS can be regarded as a replicated state machine. in other words all nodes show the same state. There is regular communication, regarding the configuration and state of all the components, between HAD on all the nodes. It is part of the role of GAB to make sure that all the currently cluster members get this information, since GAB is the membership manager in the configuration. GAB uses LLT to actually pass the messages.

If you run any kind packet sniffer on the private networks you'll generally see two types of packets, small broadcasts, which are the heartbeats, and larger addressed ones, which are the state/configuration information being passed around. And while the heartbeats are not acknowledged directly, GAB on the node where a state change has happened, expects a response from the other cluster nodes so it can confirm that everyone has the same information.

In theory, although I've never seen this happen, if a node doesn't acknowledge a state change, it will be ejected from the cluster, because GAB on the other nodes will assume there's a major problem.