Alerts

Pakyas uses a smart alert system designed to minimize noise while ensuring you never miss critical failures.

Alert Types

Down Alert

Triggered when a check misses its expected schedule.

Timing: period + grace after last ping

Example: If a check has a 1-hour period and 10-minute grace, you’ll be alerted 70 minutes after the last successful ping.

Failed Alert

Triggered when a check receives a fail ping or non-zero exit code.

Immediate: Sent as soon as fail ping is received.

Running Long Alert

Triggered when a job exceeds its maximum runtime.

Timing: When max_runtime is configured and exceeded.

Alert Channels

Configure how you receive alerts:

Channel	Setup
Email	Add email addresses to notification groups
Slack	Connect Slack workspace, select channels
Webhook	Add custom webhook URLs

Notification Groups

Notification groups let you organize alert recipients by team or severity:

1. Go to Settings > Notification Groups
2. Create groups like “On-Call Team”, “Database Admins”
3. Add email addresses, Slack channels, webhooks
4. Assign groups to checks or projects

Inheritance

• Organization defaults - Applied to all new projects
• Project overrides - Override org defaults per project
• Check overrides - Override project settings per check

Alert Conditions

Fine-tune when alerts fire:

Consecutive Failures

Only alert after N consecutive failures:

alert_after_failures: 3

Useful for flaky jobs that occasionally fail but self-recover.

Late Threshold

Alert when job is late but hasn’t missed entirely:

late_after_ratio: 0.5  # Alert at 50% of period

Example: If period is 1 hour and ratio is 0.5, alert after 30 minutes.

Missed Before Alert

Require N missed pings before alerting:

missed_before_alert: 2

Useful for jobs that may legitimately skip occasionally.

Recovery Alerts

When a check recovers (receives success ping after being down), a recovery alert is sent automatically.

Silencing Alerts

Maintenance Windows

Schedule maintenance windows to silence alerts:

• One-time windows - Specific date/time range
• Recurring windows - Weekly/monthly patterns

Pausing Checks

Pause a check to stop all monitoring:

pakyas check pause my-check

Paused checks don’t generate alerts and show as “paused” status.

Best Practices

1. Start with longer grace periods - Reduce as you understand job timing
2. Use notification groups - Don’t alert everyone for everything
3. Set consecutive failure thresholds - Reduce noise from flaky jobs
4. Schedule maintenance windows - Don’t get paged during deploys