I will use connectionTTL if no NettyConnectionTimeout is specified. I don't think we should ever allow a connection to be trying to open in more than TTL anyways. it should been considered TTL if ping / pongs were already established.
@Robbie: how that affects the qpid AMQP client? Perhaps we should also add a test using failover with a network failure over AMQP and see how the timeout is configured in there? On Thu, Sep 6, 2018 at 9:55 AM Robbie Gemmell <robbie.gemm...@gmail.com> wrote: > > Yes, I think it should use connect timeout by default, though I'd go > with a considerably bigger number than 2 seconds personally. > > Robbie > > On Wed, 5 Sep 2018 at 23:18, Clebert Suconic <clebert.suco...@gmail.com> > wrote: > > > > TL;DR: Should we make use of Netty_CONNECTION_TIMEOUT by default, as > > the connection would block forever if the IP on the live server is > > gone. > > > > > > Long version: > > > > I'm kind of proud of this little test I wrote, which I have committed here: > > > > https://github.com/apache/activemq-artemis/commit/05ce7c6ecd1c70fc571764af9027767f04538ccd > > > > The test will use sudo to create and remove networks. Along the test > > the IP will disappear. > > > > The test will validate if you have authorization to sudo, and if you > > don't it will tell you what to do in order to run it. And it will be > > ignored if you don't have the required configuration. > > > > However, there's an issue I'm not sure I should define it as a bug or > > just an user configuration. hence the discuss here: > > > > > > If the network disappears, failover will not kick in unless you > > configure NETTY_CONNECTION_TIMEOUT. As the TCP would be blocking > > waiting forever. > > > > > > I'm inclined to ellect a connection_timeout by default (say 2 seconds) > > if none is selected. But I wanted to discuss and get some ideas > > around. > > > > > > > > Ideas? -- Clebert Suconic