Dynamically adjust probe->app (or probe->probe) relationships

[2opremio]

In the ECS AMI, when using Scope in Standalone Mode, Scope is launched providing the IPs of the current members of the Autoscaling group to which the Container Instance belongs so that probes forward reports to all apps.

As I understand it (and unless the integration with Weave Net helps here) Scope won’t gracefully handle scaling up (down) the AutoScaling group. For instance:

• If I add a new instance to the group, the existing probes won’t send reports to the app in the new instance. In the same way
• If we remove an instance for the Autoscaling Group, the existing probes will try to send reports to that instance although it doesn’t exist anymore (this is less problematic).

Quoting @peterbourgon :

At the moment, there is only 1 way to dynamically update the set of apps that a probe is publishing to. And that is by specifying a [Weave] DNS name as the publish target. In that case, the probe will regularly resolve the name to IPs, and update its internal state accordingly.

It is not difficult to add some code to the probe, to allow other methods of dynamically updating the set of publish targets. But that code does not exist now. To the best of my knowledge.

[2opremio]

Further reference: https://groups.google.com/a/weave.works/d/msg/weave-users/VJ6fYfbmNlA/4h-CyV9cDAAJ

[tomwilkie]

Strangely, I was discussing this with @rade recently, the idea being we could use weave's newly refactored mesh library to propagate cluster membership info.

On Saturday, 21 November 2015, Alfonso Acosta notifications@github.com wrote:

Further reference: https://groups.google.com/a/weave.works/d/msg/weave-users/VJ6fYfbmNlA/4h-CyV9cDAAJ

— Reply to this email directly or view it on GitHub https://github.com/weaveworks/scope/issues/692#issuecomment-158464359.

[2opremio]

After https://github.com/weaveworks/integrations/pull/79 , this is not a problem anymore when running Scope in ECS (using the Weaveworks AMI).