Ask is moving to Stack Overflow and Serverfault.com! Please use the "opendaylight" tag on either of these sites. This site is now in Read-Only mode

Revision history [back]

click to hide/show revision 1
initial version

Clarification on 3-Node Clustering Failure Condition

I've seen at https://ask.opendaylight.org/question/3927/odl-clustering-with-2-nodes/ and many other locations that if you are running 3 node clustering and two nodes go down, clustering operations are suspended and the controller is unusable. In my understanding, this means that if two nodes go down, the remaining single node will not function and is incapable of managing the network. I've been discussing this with a colleague and they would like to know the exact definition of "cluster operations are suspended"; they want to believe that if two nodes go down, the last node can still manage the network and just write everything to its local copy of the store, but with no ability to replicate that data. Could you please verify if this is in in fact possible, or if not, please explain in a very clear way exactly what the suspension means.

(I believe it's meaning that since when using HA clustering, everything is transaction based and the leader needs to forward data to the replicas and receive confirmation that at least one replica committed the transaction before it can be committed as actually having occurred, in accordance with the three-phase transaction model ODL is stated to follow. With two nodes down, further transactions cannot possibly be committed as having occurred because there are no replicas to communicate with. The discussion here is on whether or not ODL currently supports in Lithium the ability for that last cluster to manage the network alone without replication.)

Clarification on 3-Node Clustering Failure ConditionFailure

I've seen at https://ask.opendaylight.org/question/3927/odl-clustering-with-2-nodes/ and many other locations that if you are running 3 node clustering and two nodes go down, clustering operations are suspended and the controller is unusable. In my understanding, this means that if two nodes go down, the remaining single node will not function and is incapable of managing the network. I've been discussing this with a colleague and they would like to know the exact definition of "cluster operations are suspended"; they want to believe that if two nodes go down, the last node can still manage the network and just write everything to its local copy of the store, but with no ability to replicate that data. Could you please verify if this is in in fact possible, or if not, please explain in a very clear way exactly what the suspension means.

(I believe it's meaning that since when using HA clustering, everything is transaction based and the leader needs to forward data to the replicas and receive confirmation that at least one replica committed the transaction before it can be committed as actually having occurred, in accordance with the three-phase transaction model ODL is stated to follow. With two nodes down, further transactions cannot possibly be committed as having occurred because there are no replicas to communicate with. The discussion here is on whether or not ODL currently supports in Lithium the ability for that last cluster to manage the network alone without replication.)

Clarification on 3-Node Clustering Failure"Cluster Operation Suspension"

I've seen at https://ask.opendaylight.org/question/3927/odl-clustering-with-2-nodes/ and many other locations that if you are running 3 node clustering and two nodes go down, clustering operations are suspended and the controller is unusable. In my understanding, this means that if two nodes go down, the remaining single node will not function and is incapable of managing the network. I've been discussing this with a colleague and they would like to know the exact definition of "cluster operations are suspended"; they want to believe that if two nodes go down, the last node can still manage the network and just write everything to its local copy of the store, but with no ability to replicate that data. Could you please verify if this is in in fact possible, or if not, please explain in a very clear way exactly what the suspension means.

(I believe it's meaning that since when using HA clustering, everything is transaction based and the leader needs to forward data to the replicas and receive confirmation that at least one replica committed the transaction before it can be committed as actually having occurred, in accordance with the three-phase transaction model ODL is stated to follow. With two nodes down, further transactions cannot possibly be committed as having occurred because there are no replicas to communicate with. The discussion here is on whether or not ODL currently supports in Lithium the ability for that last cluster to manage the network alone without replication.)

Clarification on 3-Node "Cluster Operation Suspension"Can you use a single node in a 3-node cluster to manage the network? (I think not, but confirming)

I've seen at https://ask.opendaylight.org/question/3927/odl-clustering-with-2-nodes/ and many other locations that if you are running 3 node clustering and two nodes go down, clustering operations are suspended and the controller is unusable. In my understanding, this means that if two nodes go down, the remaining single node will not function and is incapable of managing the network. I've been discussing this with a colleague and they would like to know the exact definition of "cluster operations are suspended"; they want to believe that if two nodes go down, the last node can could still potentially manage the network and just write everything to its local copy of the store, but with no ability to replicate that data. Could you please verify if this is in in fact possible, or if not, please explain in a very clear way exactly what the suspension means.

(I believe it's meaning that since when using HA clustering, everything is transaction based and the leader needs to forward data to the replicas and receive confirmation that at least one replica committed the transaction before it can be committed as actually having occurred, in accordance with the three-phase transaction model ODL is stated to follow. With two nodes down, further transactions cannot possibly be committed as having occurred because there are no replicas to communicate with. The discussion here is on whether or not ODL currently supports in Lithium the ability for that last cluster node to manage the network alone without replication.)replication. It would be a similar case to the 2-Node cluster, which I don't believe is supported in Lithium, but with 3-Node clustering.)

Can you use a single node in a 3-node cluster to manage the network? (I think not, but confirming)

I've seen at https://ask.opendaylight.org/question/3927/odl-clustering-with-2-nodes/ and many other locations that if you are running 3 node clustering and two nodes go down, clustering operations are suspended and the controller is unusable. In my understanding, this means that if two nodes go down, the remaining single node will not function and is incapable of managing the network. I've been discussing this with a colleague and they would like to know the exact definition of "cluster operations are suspended"; they want to believe that if two nodes go down, the last node could still potentially manage the network and just write everything to its local copy of the store, but with no ability to replicate that data. Could you please verify if this is in in fact possible, or if not, please explain in a very clear way exactly what the suspension means. means?

(I believe it's meaning that since when using HA clustering, everything is transaction based and the leader needs to forward data to the replicas and receive confirmation that at least one replica committed the transaction before it can be committed as actually having occurred, in accordance with the three-phase transaction model ODL is stated to follow. With two nodes down, further transactions cannot possibly be committed as having occurred because there are no replicas to communicate with. The discussion here is on whether or not ODL currently supports in Lithium the ability for that last cluster node to manage the network alone without replication. It would be a similar case to the 2-Node cluster, which I don't believe is supported in Lithium, Lithium as the page for that explicitly notes it's a design proposal but doesn't actually mention that it's been implemented, but with 3-Node clustering.)clustering's final functioning node.)

Can you currently use a single node in a 3-node cluster to manage the network? (I think not, am certain the answer is no, but confirming)was requested to get confirmation)

I've seen at https://ask.opendaylight.org/question/3927/odl-clustering-with-2-nodes/ and many other locations that if you are running 3 node clustering and two nodes go down, clustering operations are suspended and the controller is unusable. In my understanding, this means that if two nodes go down, the remaining single node will not function and is incapable of managing the network. I've been discussing this with a colleague and they would like to know the exact definition of "cluster operations are suspended"; they believe that if two nodes go down, the last node could still potentially manage the network and just write everything to its local copy of the store, but with no ability to replicate that data. Could you please verify if this is in in fact possible, or if not, please explain in a very clear clear, undeniably obvious way exactly what the suspension means?

(I believe it's meaning that since The explanation I gave was that when using HA clustering, everything is transaction based and the leader needs to forward data to the replicas and receive confirmation that at least one replica committed the transaction before it can be committed as actually having occurred, occurred. This process is in accordance with the three-phase transaction model ODL is stated to follow. With two nodes down, further transactions cannot possibly be committed as having occurred because there are no replicas to communicate with. If we cannot commit transactions, we cannot process state changes or new information. If we can't process incoming information, we can no longer manage the network. The discussion here is on whether or not ODL currently supports in Lithium the ability for that last cluster node to manage the network alone without replication. It would be a similar case to the 2-Node cluster, which I don't believe is supported in Lithium as the page for that explicitly notes it's a design proposal but doesn't actually mention that it's been implemented, but with 3-Node clustering's final functioning node.)node.

If I am correct, an answer simply affirming that my reasoning is correct would be sufficient closure on this issue.