Automating BGP Troubleshooting with Kentik AI Advisor

Welcome to a brand new demo of Kentik's AI Advisor. In this session, we'll walk through a real world troubleshooting scenario where a BGP session goes down due to a misconfiguration. We'll see how AI Advisor helps us quickly identify the issue, assess the impact, and guides us to resolution. First, using natural language queries, and then using additional AI Advisor features that make this process even faster and more consistent. First, let's take a look at how AI Advisor works. It comes in two access modes, a dedicated page and an overlay window. Users can access the AI Advisor page from the main menu in Kentik. Here, you can ask any question about your network. We'll use a simple example to explain how AI Advisor works. AI Advisor is an AI agent that can reason about the user's request, plan a solution using the tools it is equipped with, and call those tools to collect relevant data from Kentik. Each step performed by AI Advisor can be verified by the user who can view details of the tool calls and their results. Once AI Advisor has collected all relevant data, it will provide a final answer to the request. Users can copy the response or provide thumbs up down feedback, which helps Kentik improve the accuracy of AI Advisor answers and the overall user experience. With the history button, you can access previous conversations you had with AI Advisor, as well as conversations colleagues in your company had, making them easily shareable across the team. Here, you can also rename or delete past conversations. AI Advisor is also available on all Kentik portal pages via the ask button in the top right corner of the screen. This button opens an overlay window over the Portal's page. Here, you can select one of the suggested questions related to the Open Kentik portal or simply type your own question. The blue notification above the text field indicates that additional context from the portal page has been provided to the AI Advisor. In this case, AI Advisor is aware that I am on the device details page viewing this device. Now, if I ask the question, check health for this device, AI Advisor knows which device I am looking at on the screen, thanks to the additional context provided. All other features and interactions are available in the overlay window just as in the dedicated AI Advisor page. Now let's see our demo scenario in more detail. We have a small network with two edge devices connected to multiple external links. Our network connects to multiple providers. The left router peers with Charter and Server Central. The right router peers with Cogent and Equinix Internet Exchange. In this demo, we will simulate the problem on the BGP session to Charter. This problem will be detected by Kentik NMS using SNMP polling and traps. We are now on the Kentik alerting screen, which currently lists all the active alerts. Kentik detects the BGP session drop and raises an alert. Let's open the alert details page to investigate it further. We are now on the alerts details page, which provides all relevant details about the alert, such as main dimensions, start time, trigger reason, and historical alert occurrences. Let's use AI Advisor to troubleshoot this issue step by step using natural language questions. Notice the blue notification above the text field, which indicates that AI Advisor is aware of the context of our Kentik portal page and that we are looking at a specific alert. We started the investigation with the first question, over which interface is this BGP session established? AI Advisor performs necessary tool calls to find this information. After collecting enough information, it answered that the interface involved is Ethernet one slash two, which is our PNI towards Charter Communications. As a next step, we ask about possible interface flapping or going down, which can be a common cause of a BGP session going down. AI Advisor queries the relevant metrics to find the status of the affected interface. However, the interface is up, so that is not the root cause of our problem. As a next step, we ask AI Advisor to review syslog messages from the affected device and identify possible reasons. It queries the collected syslog messages through Kentik NMS. Those messages reveal the problem. The root cause is the BGP maximum prefix limit being exceeded, which can cause the session to be turned down. The syslog messages provide clear evidence about this. AI Advisor suggests the following steps: contacting Charter to understand the reason for the increased prefix number and establishing the correct number, which should be configured on the device. As you can see, AI Advisor helped us troubleshoot the network issue through natural language interaction and autonomously collected and surfaced the relevant data needed to understand the root cause of this problem. In this conversation, I guided the AI Advisor through the appropriate steps to troubleshoot this issue. However, since AI Advisor is an autonomous agent, let's look at how to make this even faster and more efficient using custom network context and natural language runbooks. We will navigate to Kentik AI settings and first check custom network context, which allows us to inject organization specific knowledge directly into AI Advisor's reasoning. This additional user configurable text is where users can provide information like device types and naming conventions, IP addressing schema, network segments, operational procedures, and even more. It is recommended to provide this additional context in the markdown format, but it is not necessary. In this example, we have included information about the most critical applications in our network and the support contact details for our service providers. Now let's check the runbooks feature. Unlike traditional runbooks, Kentik AI Advisor runbooks are human readable natural language guides that teach AI Advisor how to troubleshoot specific alerts step by step, just like a senior engineer would guide a junior colleague. They are also provided in markdown format and can include the specific steps that AI Advisor should follow while troubleshooting a particular alert type. In this specific case, we have detailed instructions on how to troubleshoot BGP down alerts in four easy steps. One, identify the affected interface. Two, assess traffic impact, three, find the root cause, and four, provide recommendations. Finally, the runbook needs to be assigned to an alerting policy so that each alert raised by that policy is associated with the provided runbook. Now let's go back to our BGP down alert and see how these features will affect the AI adviser's behavior when troubleshooting it. This time, we'll use the investigate with AI adviser button, which is just a shortcut to start an AI Advisor session in the overlay window with the default prompt. This time, AI Advisor will autonomously troubleshoot alerts. It is worth noting that AI Advisor can autonomously and successfully troubleshoot many alerts without any Runbook guidance. However, for frequent but complex network challenges, using runbooks will still provide better, more consistent behavior and results for your specific network. In the first step, AI Advisor successfully finds the affected interface. In the second step, AI Advisor follows our runbook and assesses the traffic impact. We can observe that the traffic on the Charter interface has completely dropped. However, looking at traffic on other external interfaces, we can see that traffic has been rerouted to another external link, in this case using Cogent Transit. AI Advisor provides a nice summary of the traffic impact findings and how they've affected our critical applications. In the third step, it executes root cause analysis, checking interface status metrics, BGP session transitions, device syslogs, and SNMP traps. It finds the root cause related to the maximum prefix limit and then emphasizes it. In the fourth step, it summarizes all findings and recommends actions to fix the issue. This time, the recommendation to contact our provider, Charter, includes the contact details configured in the custom network context. This concludes our demo. We've seen how AI Advisor helps identify and resolve network issues quickly and with greater reliability. First, through natural conversation and then through enhanced automation using custom network context and runbooks. This combination empowers every engineer to troubleshoot like an expert, while AI Advisor earned one thumbs up. Thanks for watching.