Loading DXData
OBSERVABILITY
Query-level tracing, flamegraphs, dataset freshness and quality metrics, and OpenTelemetry export — no bolt-on agents required.
// the problem
Most data stacks stitch together four different tools to answer basic questions — “is the pipeline running?”, “is the table fresh?”, “why is the dashboard slow?”, “who does this cost belong to?” By the time someone correlates them, consumers have already noticed.
DXData observability is first-class, not a bolt-on. Every query, pipeline step, and dataset has the same identity across traces, freshness, quality, and cost — so you see the problem in one place, route it to the right owner, and fix it before an angry Slack message shows up.
Every query gets a trace with node-by-node timing and I/O breakdown. No separate agent, no sample gaps.
Watermark tracking, source lag, and continuous assertions on every write — with downstream impact maps.
Export traces, metrics, and logs to Datadog, Honeycomb, Grafana, Jaeger — or anything that speaks OTLP.
// query.trace
DXData emits OpenTelemetry spans from inside the executor, so every query — ad-hoc, scheduled, or API-driven — has a flamegraph the moment it finishes. No sampling gaps for slow queries, no separate infrastructure to run.
// visualize.hotpath
Flamegraphs tell you the truth — whether slowdowns live in the planner, the executor, or raw I/O. Zoom into any frame to pivot the trace around the hot path, then copy a suggested fix from the query hints panel.
// datasets.freshness
Every dataset has a watermark, an SLA, and a source-lag signal. DXData tracks them continuously and knows which dashboards, jobs, and APIs depend on each table — so a stale source raises an alert with its blast-radius attached.
// quality.continuous
DXData runs assertions on every write — not on a cron — across completeness, uniqueness, validity, and consistency. Dashboards surface trends over time; breaches fail the pipeline stage and route to the owning team, instead of silently polluting downstream tables.
SLA breach · ops.inventory stale 47m (SLA 15m). Downstream: 3 dashboards, 2 pipelines.
INC-2841 · crm.accounts_daily — source lag 25h, downstream billing export paused.
// alerts.routed
Routing rules match alerts to owners by team, namespace, or SLA; deduplication collapses alert storms to one incident; quiet hours and maintenance windows keep your on-call focused. PagerDuty, Opsgenie, and Slack are first-party integrations — not glue code.
// otlp.export
Traces, metrics, and logs all speak OTLP. Fan-out to Datadog, Honeycomb, Grafana Tempo, Loki, Jaeger, Splunk, or New Relic — or all of the above at once — without touching application code. Headers, tags, and sampling are configured per exporter.
// cost.attribution
Every query, pipeline run, and dashboard render carries a team tag and a cost breakdown — slot-hours, CPU, and bytes scanned. Chargeback reports drop straight into your finance workflow; guardrails keep runaway jobs from torching the budget.
// connectors
100+ native connectors for databases, warehouses, SaaS tools, object stores, and streams. Query in place or materialize into Iceberg — your call.
// use cases
Pin the offending query from the dashboard card, open its flamegraph, and see where the time actually goes.
Freshness SLAs and watermark tracking fire alerts before a consumer notices stale numbers downstream.
Attribute every slot-hour, byte scanned, and dollar to a team — no more one-team-eats-the-bill drama.
// questions we get a lot
// related capabilities
Orchestration
Declarative DAGs with incremental materialization, backfills, and SLA alerts built-in.
ExploreQuery Engine
Federated SQL across your lake, warehouses, and operational sources — no ingest required.
ExploreGovernance
Row, column, and table-level RBAC with immutable audit logs for every query and change.
ExploreOBSERVABILITY, BUILT IN
Tracing, freshness, quality, alerts, and OTLP export — one surface, one bill, zero bolt-on agents.