Alerts and monitoring
Know when something goes down, fills up or expires before your users do.
Pier watches what happens after the deploy and warns you through the channels you choose. The idea is that you learn about the problem before your user does.
Notification channels
In Notifications, register one or more channels:
| Channel | What to enter |
|---|---|
| Discord | Webhook URL of the channel |
| Slack | URL of the incoming webhook |
| Telegram | Bot token and chat id |
| Email (SMTP) | Server, port, user, password and recipients. TLS can be required |
| Webhook | Any URL that receives a JSON POST |
In each channel, choose the events it receives and use Test to check it.
What raises an alert
| Situation | When it fires |
|---|---|
| Deploy failed or finished | At each deploy (you choose which) |
| Container restarting in a loop | 3 restarts or more in 10 minutes |
| Container killed for lack of memory | When Docker records OOMKilled |
| Failing health check | When the container becomes unhealthy after the deploy |
| High disk, memory or CPU | Above 90%, 90% and 95% for 10 minutes (editable) |
| Certificate with a problem | Invalid, self-signed or expiring in less than 14 days |
| Backup or task that did not run | 10 minutes after the expected time without a successful run |
| Restore test failed | When the weekly test of a database fails |
| Site down | When an uptime check fails |
| Server unreachable | When Pier loses contact with a server |
"Did not run" is different from "failed". Pier also warns you when a scheduled backup simply did not happen.
Without flooding the inbox
- No repetition: while an alert stays active, the message is only repeated as a reminder every 6 hours.
- Recovery: when the problem goes away, you get a single message saying it is back to normal.
- Silence: during maintenance, silence everything or one specific resource, for up to 30 days. When the silence ends, Pier evaluates the situation again.
- The state of the alerts survives a Pier restart: you do not receive everything again.
Uptime check
In Infrastructure › Monitoring, create a check for an application or stack:
- URL: one of the published domains.
- Expected status: default 200.
- Interval: from 1 minute to 1 day.
Each result is kept in a 7-day history. The check starts from the Pier server itself. To know whether the whole server went down, also use an external monitor.
Live monitoring
In Infrastructure › Monitoring, each server shows:
- CPU;
- memory used and available;
- disk of the data directory and swap;
- uptime, load average and containers.
The readings update every 15 seconds while the page is open. The limits of the capacity alerts are edited on the same screen.
Metrics history
The server and resource charts let you choose a time window. Pier collects samples in the background; the page does not need to stay open. One-minute samples are kept for 24 hours; longer periods use 15-minute aggregates, kept for 30 days. Aggregated series show average and maximum when available.
A gap means a missing reading, an unreachable server or an unsupported metric; it does not mean zero consumption. Resources with several containers are aggregated. In applications with multiple targets, CPU and memory explicitly add up the configured targets, without depending on the collection order. Each field only has a total if all targets have that reading; a partial collection leaves a gap. Host metrics remain separate, and container CPU can go above 100% when it uses several cores. Old points, from before this fix, may represent only one target; overwritten values cannot be recovered. It does not replace external monitoring of Pier itself. For logs in external storage, see server operations.