Tracked exceptions not showing up in portal

I’ve added the insights sdk to my code in order to attempt to track errors in my code for function apps. I can’t use the default method of tracking failures for function apps (passing the error to context.done(error)), because that interrupts the HTTP response the function is supposed to give.

However, the exceptions do not show up in my insights instance. I assume the configuration is correct, since my custom events are showing up (although not associated with the actual request…is that supposed to happen?)

The test function I’m using:

module.exports = function (context, myTimer) {
    const appInsights = require("applicationinsights");
    appInsights.setAutoDependencyCorrelation(true);
    const client = appInsights.getClient();
    client.trackException(new Error("handled exceptions can be logged with this method"));
    client.trackEvent('custom event');

    context.log('JavaScript timer trigger function ran!');   
    context.done();
};

I don’t know whether this is a bug in the SDK or in the portal. Errors passed to context.done() are properly marked as exceptions in the portal.

[itilitymaarten]

The names do indeed seem backwards, but thanks to your explanation I understand them now.

I’ve got the correlation working the way I want to now, except for one small detail, but I don’t believe that’s fixable. I would like the requests in application insights to be associated with the operationId as well, however since they’re generated by the Azure Functions runtime I have no way of accessing their tags. I assume this is impossible for me to fix right now, since I only know the operationId once my function run has already started.

However, thank you so much for your help. You’ve helped me understand a great deal and achieve what I wanted, within the bounds of possibility. I look forward to seeing what kind of improvements you guys bring in the next couple of months!

[OsvaldoRosado]

Thanks for the follow up! I linked #148 not because of similar situations, but because a resolution is the same - a way to track requests without a req/res pair.

As for why they cannot be correlated, we simply don’t currently support automatic request collection for anything but the standard node server. Our correlation logic also correlates on requests, so without an automatically generated request context, or a manually generated one, you have no correlation.

Your suggested solution "so couldn’t you treat any function call following as a single request, if no other method is available " is interesting, but presents some issues. For example, it is a perfectly valid use case to use this SDK outside of a server context (eg, a desktop app using node like VSCode or Electron apps in general). In this situation, assuming a correlation context merely from the lack of requests could be unexpected behavior.

This all said, because you have a single client for each request, you can work around this for your particular use case with a telemetry processor. Code like the following may work for you (make sure it’s done before recording any telemetry):

var correlationId = Math.random().toString(36).substr(2, 10);
appInsights.client.addTelemetryProcessor((envelope) => {
     envelope.tags = envelope.tags || {};
     envelope.tags['ai.operation.id']  = correlationId;
     return true;
});

We can look into creating an emulated response object if this workaround isn’t working out right - though the real solution is for us to stop requiring one to make request telemetry.