Function host does not spin up with nonexisting SB subscription

[azyl-scnsoft]

I have SB namespace without topics and subscriptions.
I created simple function with trigger bindings, configured with default values for topic\subscription names.
After restarting the app, Function host doesn't start anymore: 'Function host is not running.'
In host logs I see next log message:

40400: Endpoint not found., Resource:sb://alexzmeetupdashboard.servicebus.windows.net/mytopic/subscriptions/mysubscription. TrackingId:17db49ca-6428-4e23-937f-e39e10b44f62_G9, SystemTracker:alexzmeetupdashboard.servicebus.windows.net:mytopic/Subscriptions/mysubscription, Timestamp:9/2/2016 4:29:21 PM
Stopping Host

From Webjob source code, I see that topics and subscriptions are created only when access rights are set to Manage.

However, not sure if such configuration failure should make the host broken.

[lindydonna]

Can you explain exactly what is going on? Is that connection string invalid? It sounds like you're saying that the misconfiguration caused the host to shut down, which does indeed seem like a poor user experience.

[azyl-scnsoft]

@lindydonna

Connection string is valid. I mostly use RootManageSharedAccessKey connection string for my functions.
The problem lays in using AccessRights.Listen

Steps to reproduce:

1. Add new SB connection string (even with Manage permissions)
2. Create SB subscription trigger function with following configuration:

• 1. Set existing\nonexisting topic name
• 2. Set nonexisting subscription name
• 3. Set AccessRights to Listen

Expected result:
Binding error occured

Actual result:
Entire host became unresponsive.

[mathewc]

@azyl-scnsoft I gather that you know that the last repro steps you provided would result in an expected error (i.e. can't listen on non-existent entities if we only have Listen rights and can't create them). The question is whether this should cause host startup failure.

We need to investigate more, but I believe that currently we don't have good support for errors involving the startup of listeners, which require valid connection strings / connectivity to underlying entities. In such cases, I do believe that a host level unhandled exception may occurr, meaning the host cannot start until the error is fixed. We need to verify that is the case, and improve if possible.

The host error you showed above should have been shown to you in the portal in a red bubble window. Can you confirm that is the case? While our current handling for these error paths is not ideal, we need to at least ensure the error is surfaced.

[azyl-scnsoft]

@mathewc
Yes, I got popup message saying that webhost is in 'Internal Server Error' state.

[mathewc]

Did you also get a descriptive error message pointing to the problem? You should have.

[mathewc]

Remaining work here is to repro the issue and verify that the red bubble error includes a detailed error message indicating the issue. If it does, we can close this out as by design at this point.

[christopheranderson]

For RC timeframe, can we do a more descriptive error message?

We likely also want to look at, later on, having the host not crash and provide more graceful error messages. Out of scope for now, though.

[alex-zyl]

@mathewc
Sorry for long delay. Yes, I confirm that I'm getting the error popup with the error message that I mentioned previously.

After a while, I also got this error popup

Error:
We are unable to reach your function app (Service Unavailable). Please try again later.
Session Id: fbb6e4a4c33a42cfa70b88c53899e65a
Timestamp: 2016-10-18T15:00:52.689Z

[christopheranderson]

Moving this out as it is more complex than originally designed.

[mathewc]

This has now been fixed by @mamaso in the latest release.