Hosted beta

sigiro ingests OpenTelemetry (traces, logs, metrics, profiles) and lets you query it back with SQL. During beta, onboarding is manual. Replace <your-sigiro-host> below with the host your operator gives you.

1. Get your API key

Your operator sends you a key shaped like sk_<yourtenant>_<random>. Treat it like a password — it is shown once, cannot be recovered (only rotated), and authenticates both ingest and querying.

2. Point your OpenTelemetry exporter at sigiro

Standard OTLP over HTTP or gRPC; auth is a standard Authorization: Bearer header.

OTLP/HTTP (port 4318, recommended):

export OTEL_EXPORTER_OTLP_ENDPOINT="https://<your-sigiro-host>:4318"
export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer sk_<yourtenant>_<random>"
export OTEL_EXPORTER_OTLP_PROTOCOL="http/protobuf"

OTLP/gRPC (port 4317):

export OTEL_EXPORTER_OTLP_ENDPOINT="https://<your-sigiro-host>:4317"
export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer sk_<yourtenant>_<random>"
export OTEL_EXPORTER_OTLP_PROTOCOL="grpc"

Notes: signals post to the standard OTLP paths (/v1/traces, /v1/logs, /v1/metrics, /v1/profiles) — your SDK handles that. gzip is supported on HTTP. Both ports use TLS (https://). A single request (after decompression) must be ≤ 8 MB — default SDK batch sizes stay well under this.

3. Verify data is flowing

Send some telemetry, wait a few seconds (flush cadence ~1s), then query:

curl -s "https://<your-sigiro-host>/v1/query" \
  -H "Authorization: Bearer sk_<yourtenant>_<random>" \
  --data 'SELECT count(*) AS n FROM sigiro_spans'

A non-zero n means traces are landing. (Query after your first batch — a brand-new tenant that has never ingested has no storage yet.)

4. Query your data (SQL API)

• Endpoint: POST /v1/query on the main HTTPS port.
• Auth: Authorization: Bearer sk_<yourtenant>_<random>.
• Body: the raw SQL string (not JSON).
• Response: JSON array of row objects; header x-sigiro-truncated: true|false indicates whether results were capped.

You only ever see your own data (isolated per-tenant catalog).

Tables: sigiro_spans, sigiro_logs, sigiro_log_templates, sigiro_metrics_gauge, sigiro_metrics_sum, sigiro_metrics_histogram, sigiro_metrics_exp_histogram, sigiro_profiles, and the sigiro_anomalies table — precomputed regime shifts written continuously by the anomaly pass (also served typed at GET /v1/anomalies). Attribute columns (*_attributes, events_json, …) are JSON text — read fields with json_extract(col, '$.key') or col ->> 'key'.

Allowed SQL: SELECT only, against the sigiro_* tables. No writes/DDL, and no file/URL/S3 readers (read_csv, read_parquet, glob, … are blocked). Joins, window functions, aggregates and json_extract all work.

CTEs are rejected. A WITH clause fails validation whatever it contains — rewrite it as a derived-table subquery, SELECT ... FROM (SELECT ...) t. On hosted, information_schema is blocked too, since one shared catalog holds every tenant.

Examples:

-- Slowest operations, last hour
SELECT service_name, span_name,
       approx_quantile(duration, 0.95) / 1000.0 AS p95_ms, count(*) AS n
FROM sigiro_spans
WHERE timestamp > now() - INTERVAL '1 hour'
GROUP BY 1, 2 ORDER BY p95_ms DESC LIMIT 10;

-- Error-log rate per route, last 15 min
SELECT service_name, log_attributes ->> 'http.route' AS route, count(*) AS errors
FROM sigiro_logs
WHERE timestamp > now() - INTERVAL '15 minutes' AND severity_number >= 17
GROUP BY 1, 2 ORDER BY errors DESC;

5. Limits

There are no secondary indexes, so bound queries by timestamp for speed and to avoid timeouts/truncation.

6. Data retention

During beta, treat sigiro as a live query surface, not a system of record. Ask your operator for the current retention window and export anything you need to keep long-term.