Integrations
Your integrations are failing silently
When an integration breaks, there is rarely an error page. The service stays up, the dashboard stays green, and the data simply stops moving. A booking doesn't arrive, a vehicle status is an hour old, an employee record never updates. By the time someone notices, it is usually a customer.
Health checks that tell the truth
A typical health check asks whether the service responds. For an integration, that is the wrong question. A sync worker can answer every request perfectly while its last successful run was six hours ago.
We judge integration health by heartbeats: each live data source reports when it last delivered something. Sources that are intentionally retired or paused are excluded, so the check doesn't cry wolf, and the ones that matter can't hide behind a healthy process.
Measure freshness, per source
“Is the data current?” means something different for every source. Fleet positions might be stale after a few minutes; a nightly export is fine for a day. Give each source its own freshness threshold, track the time of the last successful sync, and alert when the gap crosses that line, not when the process dies.
Alert on what people would notice
Alerts tied to CPU and memory tell engineers about machines. Alerts tied to the business tell everyone about the problem that actually matters:
- Bookings stopped arriving from a channel.
- Vehicles stopped reporting telematics data.
- A sync is running far longer than it usually does.
Each of these points straight at a cause, and each would otherwise surface as a confused customer or a wrong decision.
Make it visible beyond engineering
The operations team is often the first to feel stale data, and the last to be told. A simple status view showing each integration, when it last succeeded, and whether it is within its threshold, gives them the same picture engineers have. It also cuts the number of “is something wrong with the system?” messages dramatically.
Every alert needs an owner and a next step
An alert nobody acts on trains people to ignore alerts. For every one we add, we agree who receives it and write down the first thing to check. If an alert fires often and nobody needs to act, it gets fixed at the source or removed.
Bottom line
Integrations fail quietly by nature. The fix is not more monitoring, but the right monitoring: freshness per source, heartbeats from the things that matter, and alerts that describe a business problem with a clear owner. Then a broken sync becomes a five-minute fix instead of a bad week.