Observability: Issues, monitors and OTLP

The Observability module (Dashboard → project → Observability) brings together what tells you about application health: logs, errors, availability and query performance.

Issues

HostBase reads Edge Function and Worker logs every 2 minutes and groups errors by fingerprint: error type, normalized message (ids, numbers and addresses removed) and the top three stack frames in your code (without line numbers, so a new deploy does not split the group). An issue has a counter, first and last occurrence, and a status: open, resolved or ignored. A resolved issue reopens when the error comes back.

• Notifications — new and regressed issues go to dashboard notifications and organization channels (Slack, Discord, email, webhook).
• Copy context for agent — markdown with the stack trace, log lines before the error and the last deploy (secrets and tokens are redacted), ready to paste into Claude Code, Cursor or Codex.
• Project webhook — JSON payload with an agent_prompt field; events issue.created and issue.regressed.

Health Check Advisors

The Advisors tab shows live rules: API 5xx rate (warning from 1%, critical from 5%), connection saturation (70% / 90%), replica lag (30 s / 5 min), data disk fill (80% / 90%) and failed or stale backups. Each result has a severity and an "Investigate" link; warnings and critical findings create notifications.

HTTP monitors

Synthetic checks of URLs (custom domains, endpoints) every 1, 5, 15, 30 or 60 minutes with an expected status, history, an SLO target (7 or 30 days) and an error budget. Alerts fire after two consecutive failures. Monitors marked public appear on /status/project/{ref}. Probes run from the dashboard server.

OpenTelemetry (OTLP/HTTP)

In Settings → Log Drains choose "OpenTelemetry (OTLP/HTTP)": enter the collector base URL (Grafana Cloud, Honeycomb, SigNoz…), auth headers and the signals: logs, metrics (entry and request counters) and traces (API requests as spans). "Test connection" sends a test record. Entries reach the collector through a HostBase relay as OTLP JSON (/v1/logs, /v1/metrics, /v1/traces).

Index recommendations

In Database → Overview HostBase analyzes pg_stat_statements and points at queries that filter on columns without a matching index. With hypopg installed the best suggestions are verified with a cost simulation without creating an index. Indexes are built with CREATE INDEX CONCURRENTLY.

Management API and MCP

GET   /api/v1/projects/{ref}/issues?status=open
GET   /api/v1/projects/{ref}/issues/{id}          # includes agent_context (markdown)
PATCH /api/v1/projects/{ref}/issues/{id}          {"status":"resolved"}
GET|POST /api/v1/projects/{ref}/monitors          DELETE .../monitors/{id}
GET   /api/v1/projects/{ref}/runtime-advisors

MCP tools: list_issues, get_issue, set_issue_status, list_monitors, create_monitor, delete_monitor, get_runtime_advisors.