[ServiceBus] Adding default idle timeout for amqpConnection object #7944
Conversation
Merge from Azure/master
Merge from master
jsquire
added
the
Client
| @@ -128,7 +128,8 @@ public static AmqpConnectionSettings CreateAmqpConnectionSettings(uint maxFrameS | |||
| { | |||
| MaxFrameSize = maxFrameSize, | |||
| ContainerId = containerId, | |||
| HostName = hostName | |||
| HostName = hostName, | |||
| IdleTimeOut = (uint)Constants.DefaultOperationTimeout.TotalMilliseconds | |||
There was a problem hiding this comment.
What was the behavior without the explicit setting? What's the driver behind this change? I'm curious on two fronts:
-
It likely changes the behavior of the library in a subtle way that current consumers may not expect and could possibly need to be a breaking change.
-
I'm curious if this is something that should be applied to Event Hubs as well, or if this is a case that is specific to Service Bus.
There was a problem hiding this comment.
So essentially, during a network communication issue, when packets are not sent being sent from client to service (or vice versa) this parameter dictates how long should the connection be alive. On windows, the connection gets dropped automatically after sometime. However, in Linux, the connection stays in the ESTABLISHED state (using netstat) even after several minutes.
From the customer view, it looks like MessageSender.Send() is called, but the operation just times out and throws ServiceBusTimeoutException instead of communication exception, since the connection never gets into faulted state.
With this change, we are forcing that 1 minute of idle time on the connection would force it to drop the connection and reestablish when required.
It likely changes the behavior of the library in a subtle way that current consumers may not expect and could possibly need to be a breaking change.
I wouldn't exactly consider it a breaking change and would think of it as a bugfix. It already works as expected on windows machine, and this would start working on linux as well.
There was a problem hiding this comment.
I'm curious if this is something that should be applied to Event Hubs as well, or if this is a case that is specific to Service Bus.
Yes. It should be applied to Event Hubs as well.
There was a problem hiding this comment.
Thanks, Neeraj. I appreciate the context.
AMQP Foundation
- Adding an idle timeout to the AMQP connection to ensure that it does
not hang in an inconsistent state on some platforms.
(see: Azure#7944)
- Restructuring of retry loops for service operations for more uniform
exception handling and logging.
- Adjusted close for client types to mark the object as closed before
taking any action to ensure that the instance is not usable for normal
operations while in a closing state; this will revert should an
exception occur to allow for attempting to close again.
- Removing unused annotations type; the members used have been folded into
the AMQP Properties class, where they are a more logical fit.
- General light tweaking and tuning AMQP-related constructs for efficiency
and to normalize patterns.
AMQP Foundation
- Adding an idle timeout to the AMQP connection to ensure that it does
not hang in an inconsistent state on some platforms.
(see: #7944)
- Restructuring of retry loops for service operations for more uniform
exception handling and logging.
- Adjusted close for client types to mark the object as closed before
taking any action to ensure that the instance is not usable for normal
operations while in a closing state; this will revert should an
exception occur to allow for attempting to close again.
- Removing unused annotations type; the members used have been folded into
the AMQP Properties class, where they are a more logical fit.
- General light tweaking and tuning AMQP-related constructs for efficiency
and to normalize patterns.
Adding default idle timeout of 60 seconds for amqpConnection object.