Slack¶

rlwatch posts alerts to Slack via incoming webhooks. The webhook URL is the only required setting.

Setup¶

Create an incoming webhook in your Slack workspace: Apps → Incoming Webhooks → Add to Slack.
Copy the webhook URL (looks like https://hooks.slack.com/services/T.../B.../...).
Configure rlwatch one of two ways:

Environment variable (recommended for CI/CD):

export RLWATCH_SLACK_WEBHOOK_URL="https://hooks.slack.com/services/..."

rlwatch.yaml:

alerts:
  slack:
    webhook_url: "https://hooks.slack.com/services/..."

You don't need to set enabled: true explicitly — rlwatch flips the flag automatically when the webhook URL is set.

Payload format¶

Each alert becomes a Slack message with block kit blocks:

Header with severity emoji (🚨 for critical, ⚠️ for warning) and detector name
Section with run id and step
Section with the alert message
Section with the recommended action
Section with up to 10 metric values from the alert

Mentioning users¶

alerts:
  slack:
    webhook_url: "..."
    channel: "#ml-alerts"
    mention_users: ["U12345", "U67890"]

mention_users takes Slack member IDs (the ones that look like U...), not display names. You can find a member ID by clicking a user's profile and copying it from "More → Copy member ID".

Failure handling¶

If the webhook returns a non-200, the failure is logged at error level. If the connection fails, the failure is logged. Either way, training is not interrupted — Slack delivery runs in a daemon thread and exceptions are caught.