DROP ALERT NOTES
Measure how late your alerts arrive
Every alert carries a Unix timestamp. Comparing it with your receipt time tells you the true latency of your pipeline, and where it leaks.
Every drop alert stamps its birth in the alerted field, in Unix seconds, as the drop alerts guide lists the event's fields, checked 2026-10-08. Comparing that stamp with the moment your code sees the alert gives you the end-to-end latency of your whole path: the stream's delivery plus your processing. If you act on alerts, that number is worth measuring, because a fast threshold rule on a slow pipeline is a slow rule.
The two clocks in every alert
The alerted field tells you when the feed detected the drop. Your receipt time, taken the moment the line is parsed, tells you when you heard about it. The difference is your total delay, and it deserves a log line or a metric per alert, the same way you would track any other performance figure.
A third timestamp sits downstream: when your message actually posts to its destination. The gap between receipt and post is your own processing, and it is the part you control entirely.
Where the delay hides
In a well-built relay the delay should be small and flat, but three places add seconds quietly. The reading loop: a client that buffers instead of streaming reads holds alerts in a local queue while the connection looks open. The formatter: any synchronous call in the message path, a lookup, a slow template, stacks onto every alert behind it. The webhook worker: a queue that drains too slowly turns into a backlog during a busy slate, which is why the relay guide keeps the POST off the reading loop.
Measure per stage and the slow stage names itself.
What to watch over a week
Latency is a distribution, not an average. Track the typical delay and the worst minute of each day, because the tail is what costs you: a relay that is fast at 3 a.m. and three seconds behind during a Saturday slate is a slow relay when it matters. Match days and live slates are the load test you get for free.
If the delay grows with alert volume, your bottleneck is in your own code. If it is flat regardless, you are reading the stream's own cadence, and that is the number to know.
Why it matters for your rules
Your thresholds and cooldowns assume the alert you see is fresh. A cooldown keyed on a three-minute-old alert is a cooldown spent on history, and a rule tuned on delayed alerts is tuned on the wrong data. The measurement closes the loop: you can only trust your rules' timing if you know when the alerts really arrived.
The timestamp is already in the alert. Compare it, log it, watch its tail. A pipeline that knows its own delay is the one that can afford a fast rule.