Soda Cloud Data Quality

Soda Cloud data testing

Soda Cloud tests SQL tables and DataFrames, reconciles tables between sources by metrics or row by row, and stores failing rows in your own warehouse. Data contracts are its core testing primitive.

Compare: data contracts tools · data testing tools · data comparison tools

• SQL Table Tests: Contract checks cover missing count/percent (completeness), duplicate count (uniqueness), invalid values and regex (validity), numeric ranges (accuracy), freshness assertions, referential integrity via custom SQL (consistency), and arbitrary custom SQL checks (docs).

• DataFrame Tests: Soda Core is explicitly required for in-memory sources (Spark, DataFrames); contracts can be verified against Pandas and Spark DataFrames in pipelines (docs).

• Table Comparison Tests: The soda-reconciliation commercial extension compares aggregates (row counts, sums) between two tables, used for cloud migrations and replication validation (docs).

• Data Diff: Record reconciliation checks perform row-to-row comparison of column contents between two data sources, surfacing rows that differ, are missing, or were added — with simple (batched) and deepdiff (full in-memory) strategies (docs).

• Data Contracts: Data contracts are the core testing primitive in Soda v4 — YAML-defined producer-consumer agreements covering schema, quality checks, and freshness, authored in UI or Git and verified on schedule or in CI/CD (docs).

• Inspect Failed Rows: The Diagnostics Warehouse (Record-Level Diagnostics) stores and surfaces the actual rows that triggered failing checks in the user's own data warehouse for investigation (docs).

• Unstructured Data Tests: No LLM-based semantic validation of free-text warehouse columns found in documentation or product pages (docs).

Soda Cloud data observability

Soda Cloud profiles columns while you write contracts and suggests checks. Its anomaly detection learns metric baselines without thresholds. Checks live in YAML contracts in Git or in the UI.

Compare: data observability tools · data profiling tools

• Data Profiling: Soda Cloud profiles datasets during contract authoring to suggest checks; the observability engine profiles column-level statistics (null %, distinct count, distributions) as continuous metrics (docs).

• Anomaly Detection: Proprietary anomaly detection algorithm learns from historical metric baselines and flags deviations without requiring hardcoded thresholds; 70% more accurate than Facebook Prophet in benchmarks (docs).

• Freshness Monitors: Metric Monitoring tracks most-recent timestamps to detect stale or delayed data with anomaly detection adapted to each dataset's load schedule (docs).

• Volume Monitors: Row count metrics are tracked over time with anomaly detection to detect unexpected volume drops or spikes without fixed thresholds (docs).

• Schema Monitors: Schema change monitoring is listed as a key observability metric, alerting on structural changes to tracked datasets (docs).

• Field Monitors: Metric Monitoring tracks missing value rates, averages, and distributions per column as continuous time-series with anomaly detection (docs).

• Distribution Monitors: Distribution shifts are monitored as column-level metric time series; the proprietary algorithm detects abnormal changes in value distributions (docs).

• Segment Monitors: Group-by configuration runs checks broken down by a categorical column (up to 1000 groups) with anomaly detection per segment — equivalent to segment-level monitoring (docs).

• Custom SQL Monitors: Contract checks support arbitrary user-defined SQL expressions tracked as continuous metric monitors with anomaly detection over time (docs).

• Partition-Based Monitors: Dataset filters scope scans to a recent time slice (e.g. WHERE ts > NOW() - INTERVAL '1d') to reduce compute cost — available when running Soda Library against Soda Cloud; not available in Soda Cloud Agreement-managed scans (docs).

• UI-Based Monitors: Soda Cloud provides a no-code editor for authoring contracts and configuring observability monitors without writing YAML or CLI commands (docs).

• Monitors-as-Code: Contracts are YAML-defined and version-controlled in Git; Soda Core CLI verifies them in CI/CD pipelines for GitOps-style data quality (docs).

• Window-Based Monitors: Configurable tumbling/sliding time windows for intra-day metric evaluation are not documented in Soda v4; scans run at dataset level on a schedule (docs).

• Pipeline Monitoring: Pipeline job runtime tracking (dbt model run times, Airflow DAG durations) is not a documented feature in Soda v4 (docs).

Soda Cloud incident management

Soda Cloud has a full incident workflow: alerts, triage, feedback, resolution, ticketing and on-call integrations.

• Alert Notifications: Notifications sent via Slack, Microsoft Teams, email, Jira, and ServiceNow when contract verifications fail or anomalies fire (docs).

• Incident Triage: Soda Cloud routes alerts to the right teams based on dataset attributes and provides an incident workflow with owner assignment and status tracking (docs).

• Alert Feedback: Users can add feedback to individual anomaly data points in the check history graph ('mark as expected') and adjust detection sensitivity to suppress known patterns (docs).

• Incident Resolution: Soda Cloud tracks incidents through triage to resolution, logs root-cause context, and integrates bidirectionally with Jira and ServiceNow for lifecycle management (docs).

• Ticketing Integrations: Native Jira and ServiceNow integrations are listed as supported messaging and collaboration integrations (docs).

• On-Call Integrations: PagerDuty and Opsgenie are confirmed as supported notification targets via the Soda Cloud webhook integration — alert payloads can trigger on-call pages (docs).

Soda Cloud data quality analytics

Soda Cloud's dashboards and health scores track data quality.

• Data Quality Dashboards: Soda Cloud provides cross-organizational dashboards with health scores, scan activity, failing check counts, and trend views filterable by domain, owner, and criticality (docs).

• Data Health Scores: Soda Cloud computes per-dataset health scores derived from check pass/fail results and surfaces them in dashboards for stakeholder communication (docs).

• Incident Resolution Metrics: Soda Incidents support status tracking and resolution notes per incident, and the built-in Organization Dashboard shows only failing-checks counts and an overall health score — no MTTD/MTTR or resolution-time trend chart out of the box; that would require building custom KPI dashboards via the Reporting API (docs).

Soda Cloud data governance

Soda Cloud syncs data quality results to catalogs such as Alation, Atlan, Collibra, Microsoft Purview and data.world.

• Data Catalog Integration: Soda integrates with Alation, Atlan, Collibra, data.world, Metaphor, and Purview to push quality scores and contract metadata into external catalog platforms (docs).

• Data Catalog: Soda does not include a built-in searchable data catalog; it integrates with external catalogs (Alation, Atlan, Collibra) to push quality metadata (docs).

• Data Classification: Soda focuses on data quality testing and observability; no data classification or sensitive column detection feature is documented (docs).

• Data Products: Soda offers generic 'dataset attributes' for tagging datasets (with 'business domain' given only as an example attribute value) used for filtering/reporting, but there is no first-class Data Product/domain entity that groups monitors, tests, and quality scores at a data-product level (docs).

Soda Cloud workflows

Soda Cloud has a REST API, a Python SDK, webhooks, an MCP server and a dbt integration. Contracts live in Git, and the Soda Core CLI verifies them in CI/CD, failing the build when checks fail.

• Scheduling: Contract verifications and observability scans are scheduled via Soda Cloud; the Soda Runner executes them on a configurable cadence (docs).

• CI/CD Integrations: Soda Core CLI runs checks in GitHub Actions and other CI pipelines and exits non-zero on failure, blocking deployments on contract violations (docs).

• Webhooks: A dedicated Webhook API sends HTTP callbacks to external endpoints when check results or incidents occur (docs).

• REST / GraphQL API: Soda exposes a public REST API for managing datasets, running verifications, and retrieving results programmatically (docs).

• MCP Server: Soda MCP is a Model Context Protocol server listed in the reference documentation, enabling AI coding agents to query Soda context (docs).

• Python SDK: The Soda Python API (pip-installable soda-core library) enables programmatic contract verification in pipelines and in-memory DataFrame scanning (docs).

• dbt Integration: Soda Cloud ingests dbt test results by reading manifest.json and run_results.json via 'soda ingest', surfacing dbt outcomes alongside native Soda checks with alerting and incident tracking (docs).

Soda Cloud AI agents

Soda Cloud's Ask AI turns plain-language descriptions into checks, suggests checks from profiling, and triages issues. An MCP server is also available.

• AI Chat Assistant: Ask AI (formerly SodaGPT) is a built-in Soda Cloud chat assistant that converts natural language descriptions into fully-formed, syntax-correct SodaCL YAML checks (docs).

• Natural Language Test Generation: Ask AI converts plain-English instructions directly into fully-formed, syntax-correct SodaCL checks, and Contract Copilot lets users request changes like "Add freshness checks" in plain English to generate/update actual data contract checks (docs).

• Automated Test Generation: Contract Autopilot automatically generates a first-draft data contract with checks based on the dataset's schema and profiling data (docs).

• AI Issue Triaging: Record-level Anomaly Detection (RAD) includes built-in root cause analysis to identify what changed and why an anomaly fired (docs).

• AI Issue Resolution: AI-powered remediation is listed as 'coming soon' on the Soda website; not yet available in the product (website).