Reliability and failures

The worst failure is a silent one. So in ALERTIS every breakage has someone who will learn about it, and a moment when that happens. And nobody gets woken up at night.

How the night goes

1. 02:00 UTC. Nightly collection. The service pulls every client's yesterday data from GA4 one by one (never in parallel). If someone's pull fails, nobody is disturbed at night — the attempt is just logged.
2. 02:30 and 03:30. Automatic retries. Only for clients whose previous attempt failed. Most transient issues (network, API hiccups) heal themselves here.
3. 05:00. The final attempt. If this one fails too, it's no longer a coincidence — and it's morning, time to notify.

Who gets notified after the final attempt

• The cause is on your side (GA4 access expired or revoked): you get a Telegram message explaining it, with a one-minute reconnect instruction. The service owner gets a copy.
• The cause is on our side (Google API failure, network, quotas): only the service owner is notified and deals with it. You're not pinged, because there's nothing you need to do.

What the dashboard shows

• A red "Reconnect GA4" banner on the Overview when analytics access is broken.
• An amber "Telegram reports are off" banner when data flows but delivery is disconnected or paused.
• An amber "Refresh data" banner if data hasn't updated for over 36 hours.

Safety nets

• A daily ops digest to the owner: how many collections succeeded, how many failed, with error texts.
• An independent watchdog on separate infrastructure checks daily that the nightly collection ran at all and that everyone who should have received a report actually did. If the main scheduler dies completely, the watchdog notices and reports it.
• Exact-time delivery is dedup-protected: every successful send is recorded, so even a manual blast can't cause a second message the same day.