Alright, let's get this out there. I've been running production systems since some of you were in diapers, and I've seen my fair share of "magical" platforms that promise to simplify your life while quietly adding a whole new layer of complexity and failure modes. Granola, with its promise of unified cloud cost visibility, seemed like a sensible idea. But the execution, particularly around its notification system, feels like it was built by a team that's never been paged at 3 AM.
Here's the recurring nightmare: you set up your budget alerts, anomaly detection, whatever. It works flawlessly for a week, maybe two. You start to trust it. Then, without any visible error in the Granola UI, no warnings, no failed status checks—the notifications just stop. For days. Silence. Meanwhile, your dev team has decided to spin up a few dozen `c6gd.16xlarge` instances for a "quick test," and you're blissfully unaware until the weekly finance report hits.
My configuration isn't rocket science. It's all webhook-based, pointing to a simple Slack endpoint and a PagerDuty integration. The same endpoints handle alerts from five other systems without issue.
```yaml
# Example of the alert config (Granola's UI exports this)
alert:
name: "Prod AWS Spike"
threshold: 15% increase
window: 1 hour
channels:
- slack: "#alerts-finops"
- pagerduty: "cost_anomaly_service"
enabled: true
```
The dashboard says "Active." The last test notification succeeded. But actual cost events? Crickets.
What I've observed so far:
* The outage isn't global. Colleagues in other regions sometimes get their alerts while mine are dead.
* There's no correlation with maintenance windows announced by Granola.
* Toggling the alert "off" and "on" again sometimes revives it for a short period.
* The Granola status page remains a serene, unblemished green throughout.
This isn't a minor bug. This is a fundamental breach of trust for a service whose primary value proposition is *awareness*. It forces you to implement redundant alerting from the source clouds, which begs the question: why am I paying for Granola's "unified" layer if I need to keep the underlying systems wired up anyway?
So, I'm putting it to the community: am I the only one experiencing this? Has anyone managed to find a pattern, a workaround, or a configuration nuance that actually makes this thing reliable? Or are we all just nodding along, accepting that our cost management tool periodically takes a multi-day nap?
keep it simple