Introduction
A customer is asking for the following - anomaly detection capability to monitor integration runtimes and
identify performance degradation trends across a group of integrations.
Here is a simple solution, based on an Agent, leveraging the OIC factory api.
OIC Factory API
I use the following api -
I test it out on the following simple integration from my project AA-Anomaly-Detection -
This create order integration is simply a stub, the only logic being to execute a WAIT of 20 seconds on orderNr 310138.
This is how I will generate an anomaly or 2.
I execute this integration a couple of times, including once with the orderNr 310138. I then execute the api in postman -
As you can see, the response contains fields such as - creationDate, instanceId, integrationId, duration etc.
The duration field gives us the execution time.
OIC Agent Tool
I create an integration that invokes the OIC Factory api, that we just covered.
This is then exposed as a tool.
This integration has the following request payload -
The response is set to the result I receive in postman.
OIC Agent
The agent uses the tool we've just discussed. Its configuration is as follows -
I now execute the
create order integration a couple of times -
I edit the prompt to also include the overall number of flows and then re-run the agent -
Looks good! Now let's add another integration to the mix.
This also includes the WAIT of 30 secs on orderNr 310138.
I amend the agent guidelines as follows -
I update the template -
I run the create order integration 3 times, the update order run 5 times. Each has an anomaly.
Summa Summarum
This is a simple demo, which I could improve on. For example, I could include a scheduled integration that runs every nn hours. This could invoke the agent via the OIC AI Agent Native action. I could also include a Notification tool - based on a simple integration that sends the output via email to all interested parties.
