Reliability is important to us. We even inject failure into our systems every Friday to prove it. But when it comes to sending alerts, reliability goes beyonds writing flawless code. We rely on several third-party carriers to deliver alerts to our customers. If an SMS isn’t delivered, you aren’t notified of an outage. We can’t stick our heads in the sand, telling you that it was because our providers were down. That’s not good enough. We need a way to ensure that you will receive your alerts no matter what.

Initially, we provided reliability with our providers with redundancy. We have several providers who aid us in delivering our SMS alerts to our customers. Redundancy is great and it’s something that we will continue to do. But we felt we can do even better.

Over the last 4 months, we’ve introduced End-to-End SMS Provider Testing to proactively identify any delays or outages our providers may be experiencing, even if their publicly facing status pages are out-of-date. We often found that our providers would show that their systems were fully available, but SMS messages would not be received, even going as far as sending us a delivery confirmation. Using End-to-End Provider testing we can quickly switch the order of our primary, secondary and tertiary providers to ensure that our customers alerts are not affected by unexpected outages with our third-party providers. Before introducing this testing, we had no choice other than to believe our providers and react accordingly.

How We Conduct End-to-End SMS Provider Testing

We are continuously testing our SMS providers. For our short codes we send a test SMS every two-minutes, while varying the frequency for our less-frequently used long codes. To conduct testing, we send alerts to ourselves all day using each of our providers. We can measure how long it takes for the message to be received, if at all, to determine if a provider is available (and how well they are performing). We then determine if we should demote or promote the usage of our providers.

How We Automate End-to-End SMS Provider Testing

• Currently, we have 3 Android phones with different mobile carrier networks: AT&T, Verizon and T-Mobile. (We will be adding Sprint soon)

• Using our internally built android app for this testing system we send test SMS alerts from our system to each of our Android phones in a round-robin rotation.

• Using Datadog, we calculate the time taken for each SMS to reach the designated testing phone and how long our testing app takes to reply back to us.

• Based on measurements we determine if a provider is down or degraded, then take the appropriate action.

We consider a provider to be degraded if their SMS delivery latency is more than 3 minutes or if an SMS is not received for the last two attempts across each of the mobile carrier networks. If a provider is degraded, our team is alerted and replaces the provider to proactively avoid our customers missing any of their SMS alerts.

The Future of End-to-End SMS Provider Testing

Currently, the process to switch the priority level of our SMS providers is a manual process. We receive an alert that a provider is degraded and we react accordingly. As we move forward we will be automating by building a probabilistic model, moving away from receiving failure provider alerts.

This process has allowed our team to get appropriate alerts from our testing systems so we don’t have to set-up a variety of alerts on different resources to be aware of issues. This reduces the overall number of pages our on-call engineer receives, allowing us to spend time solving issues and improving processes without getting distracted by noise.

We will continue to focus on reliability here at PagerDuty. Not only do these tests and automation practices keep our product available for you, we also are compiling many learnings from the state of the systems we connect with. We have definitive knowledge into the performance of our mobile providers without having to rely on external reports, allowing us to be proactive and deliver you an awesome PagerDuty experience.