DOCS

Quickstart.

The full documentation ships with early access. This page carries what evaluating engineers actually need first: the connection model and the sharp edges.

01 — CONNECT A WORKSPACE

main.tf · telemetria moduleread-only scopes

module "telemetria" {
  source        = "vrahad/telemetria/databricks"
  workspace_ids = [var.prod_workspace]
  scopes        = ["system.access", "system.billing",
                   "event_logs.read", "delta.history.read"]
  region_pin    = "us-east-1"
}

# terraform apply → first baselines within one scheduling cycle (~30 min)
# terraform destroy → clean revoke, nothing left behind

02 — THE QUESTIONS ENGINEERS ASK FIRST

What exactly can telemetria read?: System tables (access, billing, lineage where enabled), Spark event logs, Delta transaction history, and query history. It cannot run SELECT on your tables — the service principal has no data-plane grants to revoke because it never had them.
What leaves our VPC?: Job/run metadata, stage metrics, table statistics, schema versions, and cost figures. Row-level data does not leave; column names do (they are required for lineage). If column names are sensitive, masking rules apply before ingest.
How do expectations get enforced if you are read-only?: Enforcement runs inside your pipelines via a lightweight SQL/DLT macro library that you version in your own repo. The plane observes results; your code does the halting. Nothing in your critical path calls out to us.
What happens when telemetria is down?: Nothing. Pipelines run, expectations still enforce locally, telemetry buffers in your event logs. The plane catches up on reconnect. We are not in your execution path by design.
Structured Streaming?: Not natively yet. Streaming jobs appear as batch runs with reduced fidelity. If streaming is your core workload, we are not ready for you — ask us for the roadmap before buying.

Scope questions? We’ll walk your security team through every grant.

Run a free health check Read the architecture