[
https://issues.apache.org/jira/browse/DISPATCH-1008?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16540187#comment-16540187
]
Chuck Rolke commented on DISPATCH-1008:
---------------------------------------
In diagnosing Commit cc3ab655 the issue was not the number of attempts to find
the correct failover list. The failover list from Router B includes a *bogus
host 'third-host'* whose name never resolves. Since the host name is a simple
name and does not include a domain name field, a test system, like my laptop,
might spend a lot of time querying DNS (internet provider DNS, corporate VPN
DNS, etc) trying to resolve the name. In my case I can see that Router A
disconnects from Router B and reconnects to Router C 4.5 seconds later. That's
why the test takes so long on some systems.
◊ ◊◊ 3.233983 {color:#000000} Frame 309 127.0.0.1:45172 -> 127.0.0.1:20902
-> {color} *flow* [0,2] (1,250)
◊ ◊◊ 3.381848 {color:#000000} Frame 356 127.0.0.1:45172 -> 127.0.0.1:20902
-> {color} *close* [0] <-- connection to B lost
◊ ◊◊ 7.936659 Frame 479 127.0.0.1:50038 -> 127.0.0.1:20904 -> *init*
SASL (3): (1.0.0) <-- Connection to C started
◊ ◊◊ 7.939116 Frame 481 127.0.0.1:50038 <- 127.0.0.1:20904 <- *init*
SASL (3): (1.0.0), *method* Method:
> Router should preserve original connection information when attempting to
> make failover connections
> ---------------------------------------------------------------------------------------------------
>
> Key: DISPATCH-1008
> URL: https://issues.apache.org/jira/browse/DISPATCH-1008
> Project: Qpid Dispatch
> Issue Type: Bug
> Reporter: Ganesh Murthy
> Assignee: Ganesh Murthy
> Priority: Major
> Fix For: 1.2.0
>
> Attachments: broker-slave.xml, broker.xml, qdrouterd-failover.conf
>
>
> # Start artemis master and slave brokers and the router with the attached
> config files.
> # Notice that the router receives an open frame from the master broker with
> the following failover information
> #
> {noformat}
> 2018-05-22 22:11:11.830106 -0230 SERVER (trace) [1]:0 <- @open(16)
> [container-id="localhost", max-frame-size=4294967295, channel-max=65535,
> idle-time-out=30000,
> offered-capabilities=@PN_SYMBOL[:"sole-connection-for-container",
> :"DELAYED_DELIVERY", :"SHARED-SUBS", :"ANONYMOUS-RELAY"],
> properties={:product="apache-activemq-artemis",
> :"failover-server-list"=[{:hostname="0.0.0.8", :scheme="amqp", :port=61617,
> :"network-host"="0.0.0.0"}]"}]{noformat}
>
> # Now, kill the master broker and notice that the router correctly fails
> over to the slave broker. But the slave broker does not provide any failover
> information in its open frame and hence the router erases its original master
> broker connection information
> # When the master broker is now restarted and the slave broker is killed,
> the router attempts to repeatedly connect only to the slave broker but never
> attempts a connection to the master broker.
> # If the router did not erase its failover list but preserved the original
> master connection information, it would have connected the master broker.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
