[jboss-jira] [JBoss JIRA] (WFLY-11750) Allow cluster to use DNS addresses instead of IP addresses

[Paul Ferraro (Jira)]

    [ https://issues.jboss.org/browse/WFLY-11750?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13700244#comment-13700244 ] 

Paul Ferraro commented on WFLY-11750:
-------------------------------------

[~tomekadamski] Thanks for elaborating - I think I understand the issue now.
{quote}But if I understand correctly HA features and correct transaction recovery are independent. {quote}
Correct.

{quote}As a result, the client has a record in its permanent object store and tries to finish the transaction. As the record contains the IP address of the failed node the client will wait till the server is up again to finish the transaction.{quote}
The address would have been obtained from the service URL of the ejb client connection.

{quote}That node will have the new cluster identity. My understanding is that it has no effect on transaction recovery which will work fine.{quote}
Correct.

{quote}But the key thing here IMO is, the way remoting gets node information.{quote}
Agreed.  The server supplies this information via client mappings, the destination address for which is stored and supplied as a string.   Thus the server *should* already be capable of seeding the client with a hostname instead of an IP address.  However,  I suspect that the address used by the transaction client is taken from the remote connection itself - which likely use the resolved destination address (as the established connection will have already resolved the host name), rather than preferring the host name over the address, if it exists.

{quote}Another thing: does clustering takes care of replicating Narayana object store? If this is the case then the problem would be different.{quote}
Clustering does not currently perform any state replication on behalf of the the Narayana object store.

> Allow cluster to use DNS addresses instead of IP addresses
> ----------------------------------------------------------
>
>                 Key: WFLY-11750
>                 URL: https://issues.jboss.org/browse/WFLY-11750
>             Project: WildFly
>          Issue Type: Feature Request
>          Components: Clustering
>    Affects Versions: 16.0.0.Beta1
>            Reporter: Tomasz Adamski
>            Assignee: Paul Ferraro
>            Priority: Major
>
> We would need a configuration that would allow for the cluster to use DNS addresses instead of IP addresses. The reason is that OpenShift guarantees the node identity under DNS address and not under IP address.
> Sample scenario that may currently fail when application are deployed in OpenShift:
> A (application)
> B (clustered application)
> 1. A calls transactional invocation on B
> 2. as a result of discovery process A obtains a cluster topology from B and uses one of obtained IP addresses for the connection
> 3. as the invocation is transactional the object-store records are written in A's persistent object store; those records are based on the data obtained from the cluster => subordinate node is identified by the IP address from point two
> 4. B node fails
> 5. OpenShift restarts node B on another IP address
> 6. A attempts recovery and persistently fails
> OTOH OpenShift guarantees node identity under DNS address. As a result, at point 5 node is guaranteed to restart on established DNS address so if the cluster used this address instead of physical addresses the scenario above will finish with A being able to recover the transaction.



--
This message was sent by Atlassian Jira
(v7.12.1#712002)