Databricks Observability Tools Priced for Data Quality, Job Monitoring and Lineage

Databricks observability tools fall into three groups, and only one of them publishes a price you can multiply. Databricks' own data quality monitoring has no per-table price at all; it bills serverless DBUs at $0.35 per DBU on AWS Premium in US East and $0.45 on Azure Premium. Datadog meters Databricks job monitoring at $0.05 per host-hour. Monte Carlo and Sifflet sell $50,000 and $48,000 annual packages on AWS Marketplace, DataFlint charges $2,000 a month flat for Spark cost work, and Databricks' own Overwatch project is archived and deprecated. Every figure on this page was read from the vendor's own page or listing on 12 October 2026.

Page one for this search is a Sifflet guide, an Atlan best-practices post, Databricks' own docs and two Unravel PDFs. They explain what observability is. None of them tells a platform lead what any of it costs, which is the question that decides the purchase. So this page starts there.

Databricks observability tools compared on price

Sources: databricks.com/product/pricing/lakeflow-jobs (Premium, AWS, US East), the Azure Retail Prices API for Azure Databricks in East US, Databricks data quality monitoring docs, datadoghq.com/pricing/list (US), AWS Marketplace listings for Monte Carlo, Sifflet and Acceldata, dataflint.io/pricing, unraveldata.com/pricing and the databrickslabs/overwatch repository. All read 12 October 2026. List prices, no discounts.

How much does Databricks data quality monitoring cost?

There is no list price per table or per monitor. Databricks' documentation says data quality monitoring "runs on serverless compute and is billed as serverless DBUs under the DATA_QUALITY_MONITORING billing product", and that cost depends on the number and size of the tables and how often they are evaluated. The rate behind those DBUs is the serverless jobs rate: $0.35 per DBU on AWS Premium in US East, compute included, and $0.45 per DBU-hour for Premium Automated Serverless Compute on Azure in East US. Serverless has a one-minute minimum.

In practice that makes it cheap to try and hard to forecast. Every 100 DBUs the monitors consume costs $35 on AWS and $45 on Azure, but nothing on the pricing page tells you how many DBUs a monitor on a 2 TB table will burn. The only way to know is to run it and read the bill, which Databricks documents as a query against system.billing.usage:

SELECT usage_date, SUM(usage_quantity) AS dbus
FROM system.billing.usage
WHERE billing_origin_product
      = 'DATA_QUALITY_MONITORING'
GROUP BY usage_date
ORDER BY usage_date;

Two details matter if you are comparing months. Profiling results created before February 2026 were billed under the jobs serverless SKU with a LakehouseMonitoring tag rather than under the DATA_QUALITY_MONITORING product, so a year-over-year query needs both filters. And anomaly detection, the schema-level part of the feature, is still in Public Preview.

Databricks system tables are free, but only for a year

The cheapest observability on Databricks is the data it already writes. Databricks states that "system tables are free to use. You are only charged for the compute used to query them." The system.lakeflow job tables, system.billing.usage and the lineage tables each keep 365 days free. A configurable-retention option in Beta raises the free window to 395 days and allows up to 3,650, with storage past 395 days to be charged once it reaches general availability.

Lineage has a split worth knowing before an audit. The lineage system tables keep a rolling one-year window, while Catalog Explorer keeps lineage indefinitely but only for events captured after 1 September 2024. If you build a dashboard on the system tables, it will quietly forget anything older than twelve months. Our Unity Catalog lineage guide covers the other limits, including that lineage is not preserved when a table or column is renamed.

One dated change also affects Azure buyers. Microsoft's Azure Databricks pricing page says the Standard tier "will be retired on October 1, 2026". Every Azure workspace is now on Premium rates, which is where Unity Catalog and these features live anyway. The full DBU price list is on Databricks DBU cost.

Is Overwatch still maintained?

No. Overwatch, the Databricks Labs project many older guides still recommend for workspace monitoring, is archived on GitHub. Its README opens with "DEPRECATION NOTICE: This project is now deprecated and will no longer be maintained." The last release, v0.8.2.0, shipped on 26 September 2024, and the only commit to main in the past year is the one that added the notice, on 7 January 2026. If a vendor comparison you are reading still lists Overwatch as a free option, check its date.

Databricks job monitoring and Spark cost tools compared

Job and cluster monitoring is a different purchase from data quality, and on Databricks it is often where the money is. Three tools publish enough to compare.

Datadog lists "Data Observability: Jobs Monitoring (Clusters)" at $0.05 per host per hour on an annual commitment, $0.06 month-to-month and $0.072 on-demand. A workspace running 12 hosts for 10 hours a day is 3,600 host-hours a month, or $180 a month and $2,160 a year. That is inexpensive, and if your platform team already pages through Datadog, a failed Databricks job lands in the same rotation as the infrastructure alert that caused it. Teams that run a separate incident management tool can route the same alert there instead. We cost Datadog's full data bill, quality monitoring included, on Datadog Data Observability pricing.

DataFlint sells Spark cost optimization rather than alerting. Its Team plan is $2,000 a month flat, billed monthly, for up to $1M a year of monitored Spark spend. Enterprise is priced at 1.5% of retail Spark spend, which on a $2M Databricks bill is $30,000 a year, though the footnote adds "floor commitments and volume discounts tailored to your organization". It also maintains an Apache-2.0 replacement for the Spark UI.

Unravel publishes no dollar figures. Its pricing page says pricing is "based on DBU consumption per month" and offers a free health check. Acceldata's AWS listing prices Spend Intelligence for Snowflake or Databricks at $100,000 a year per account or workspace, separate from its $5,000 Data Reliability package.

Data observability for Databricks tables

For the tables themselves, the published packages are annual contracts in the tens of thousands. Monte Carlo's AWS Marketplace listing sells a $50,000 credit package for 12 months and Sifflet's sells $48,000 in platform credits. Datadog's Quality Monitoring is $16 per monitored table a month, so 250 tables cost $48,000 a year at list, the same as the Sifflet package. All three monitor freshness, volume and schema across many tables and use lineage to explain an alert after it fires.

What none of them is built around is the question a Databricks team asks before a change: if I drop or rename this column, which notebooks, jobs and dashboards break? Unity Catalog lineage answers it inside Databricks, for tables registered in Unity Catalog, as long as nothing was renamed. It stops at the workspace edge, so a Snowflake table fed from Databricks or a Tableau workbook built on it falls outside the graph. Databricks lineage in DataTrail reads query history read-only and connects Databricks, the other warehouses and the BI layer at column level, with freshness and schema-change alerts that list what each failure reaches.

Which Databricks observability tool should you buy?

Start with what you already pay for. If the workspace is on Unity Catalog, turn on data quality monitoring on your ten most important tables, query system.billing.usage after a month, and you will know what native monitoring costs you. Build job dashboards on the free system tables.

Add Datadog if your platform team lives in it and job failures are the pain. At $0.05 per host-hour the Databricks line is small; watch the per-table line if you add quality monitoring.

Add DataFlint if the Databricks bill is the pain. A flat $24,000 a year is easy to justify against a seven-figure Spark spend if it finds even a few percent.

Buy Monte Carlo or Sifflet if you run hundreds of business-critical tables across several platforms and need an incident process with owners and SLAs.

Add DataTrail if the incidents that hurt are the ones nobody saw coming downstream: a column change that broke a dashboard in another tool. Plans are flat, $708 to $5,748 a year on Starter, Team and Scale, with Enterprise above that, and the full field is compared in data observability tools.

Frequently asked questions

Does Databricks have built-in data observability?

Partly. Unity Catalog includes data quality monitoring for table freshness, completeness and profiles, lineage for registered tables, and free system tables for jobs and billing. It does not cover assets outside Databricks, and monitoring is billed as serverless DBUs with no published per-table price, so cost has to be measured from system.billing.usage.

How much does Databricks Lakehouse Monitoring cost?

Lakehouse Monitoring, now called data quality monitoring, is billed as serverless DBUs. On AWS Premium in US East the serverless jobs rate is $0.35 per DBU including compute, and on Azure Premium in East US it is $0.45 per DBU-hour. Databricks publishes no per-table or per-monitor price, so the total depends on table size and refresh frequency.

Does Datadog monitor Databricks jobs?

Yes. Datadog Data Observability Jobs Monitoring covers Databricks clusters and Spark, showing run health, durations and stage metrics such as idle executor CPU and disk spill. It is billed at $0.05 per host per hour on annual terms, $0.06 month-to-month and $0.072 on-demand, per Datadog's US price list read on 12 October 2026.

What is the best observability tool for Databricks?

It depends on the failure you keep having. For data quality inside one workspace, Databricks' native monitoring is the cheapest start. For job failures in a Datadog shop, Datadog. For Spark cost, DataFlint. For many critical tables across platforms, Monte Carlo or Sifflet. For column-level impact across Databricks and BI tools, DataTrail.