auditing-table-statistics

Auditing Table Statistics

Audits optimizer table statistics for staleness, missing column coverage, and row count drift to diagnose poor query performance caused by outdated or incomplete statistics. Uses SHOW STATISTICS for read-only SQL analysis of table-level and column-level statistics freshness, entirely without requiring DB Console access.

Complement to profiling-statement-fingerprints: This skill diagnoses optimizer statistics issues; for identifying historically slow queries, see profiling-statement-fingerprints.

Prerequisites

SQL connection with any privilege on target tables
Automatic statistics collection enabled (default): sql.stats.automatic_collection.enabled = true

Related skills: profiling-statement-fingerprints for historical query analysis, triaging-live-sql-activity for live triage.

Core Concepts

CockroachDB-specific defaults:

Automatic collection triggers at ~20% row count change (sql.stats.automatic_collection.fraction_stale_rows)
Auto-collection covers index column groups (when sql.stats.multi_column_collection.enabled = true, the default); ad-hoc multi-column stats on non-indexed columns require manual CREATE STATISTICS
Large tables (>10M rows) may have delayed auto-collection
Staleness thresholds: refresh if >7 days (OLTP) or >30 days (OLAP), or >20-30% row count drift

See references/statistics-thresholds.md for workload-specific guidance.

Core Diagnostic Queries

Query 1: Identify Tables with Stale or Missing Statistics

Finds tables with outdated statistics or no statistics at all, ranked by staleness.

WITH table_stats AS (
  SELECT
    table_catalog,
    table_schema,
    table_name,
    column_names,
    row_count,
    created,
    now() - created AS stats_age
  FROM [SHOW STATISTICS FOR TABLE database_name.*]  -- Replace database_name
  WHERE column_names = '{}'  -- Table-level stats only (empty array)
)
SELECT
  table_schema || '.' || table_name AS full_table_name,
  row_count,
  created AS stats_created_at,
  stats_age,
  CASE
    WHEN created IS NULL THEN 'Missing statistics'
    WHEN stats_age > INTERVAL '30 days' THEN 'Very stale (>30d)'
    WHEN stats_age > INTERVAL '7 days' THEN 'Stale (>7d)'
    ELSE 'Fresh'
  END AS staleness_status
FROM table_stats
WHERE stats_age > INTERVAL '7 days' OR created IS NULL  -- Adjust threshold
ORDER BY stats_age DESC NULLS FIRST
LIMIT 50;

Customization:

Replace database_name.* with specific schema pattern (e.g., mydb.public.*)
Adjust staleness threshold: INTERVAL '7 days' for OLTP, '30 days' for OLAP
Increase LIMIT to see more tables

Key columns:

staleness_status: Quick classification of statistics freshness
stats_age: Exact time since last collection
row_count: Last known table size

Query 2: Audit Statistics for Specific Table

Shows all statistics for a single table, including table-level and per-column details.

SELECT
  column_names,
  row_count,
  distinct_count,
  null_count,
  created,
  now() - created AS stats_age,
  CASE
    WHEN histogram_id IS NOT NULL THEN 'Yes'
    ELSE 'No'
  END AS has_histogram
FROM [SHOW STATISTICS FOR TABLE database_name.schema_name.table_name]
ORDER BY
  CASE WHEN column_names = '{}' THEN 0 ELSE 1 END,  -- Table-level first
  created DESC;

Customization:

Replace database_name.schema_name.table_name with fully-qualified table name

Key columns:

column_names: Empty {} = table-level, single element = column-level
distinct_count: Cardinality for selectivity estimates
null_count: NULL value count for IS NULL predicates
has_histogram: Distribution data availability

Interpretation:

First row (column_names = '{}') shows table-level row_count
Subsequent rows show per-column statistics
Missing columns indicate no statistics collected yet

Query 3: Detect Row Count Drift

Compares current table row count against cached statistics to identify significant drift.

WITH current_count AS (
  SELECT count(*) AS actual_rows
  FROM database_name.schema_name.table_name  -- Replace with target table
),
stats_count AS (
  SELECT row_count, created
  FROM [SHOW STATISTICS FOR TABLE database_name.schema_name.table_name]
  WHERE column_names = '{}'  -- Table-level stats
  ORDER BY created DESC
  LIMIT 1
)
SELECT
  c.actual_rows,
  s.row_count AS stats_rows,
  s.created AS stats_created_at,
  now() - s.created AS stats_age,
  ABS(c.actual_rows - s.row_count) AS drift_absolute,
  ROUND(
    ABS(c.actual_rows - s.row_count)::NUMERIC /
    NULLIF(s.row_count, 0) * 100,
    2
  ) AS drift_pct,
  CASE
    WHEN ABS(c.actual_rows - s.row_count)::NUMERIC / NULLIF(s.row_count, 0) > 0.30 THEN 'High drift (>30%)'
    WHEN ABS(c.actual_rows - s.row_count)::NUMERIC / NULLIF(s.row_count, 0) > 0.20 THEN 'Medium drift (>20%)'
    WHEN ABS(c.actual_rows - s.row_count)::NUMERIC / NULLIF(s.row_count, 0) > 0.10 THEN 'Low drift (>10%)'
    ELSE 'Minimal drift (<10%)'
  END AS drift_status
FROM current_count c, stats_count s;

Customization:

Replace table name in both CTEs
Adjust drift thresholds (30%, 20%, 10%) based on workload tolerance

Key columns:

drift_pct: Percentage difference between current and cached row count
drift_status: Classification for prioritization
stats_age: Time since statistics last refreshed

Interpretation:

>30% drift: Urgent refresh recommended, optimizer estimates likely very inaccurate
20-30% drift: Consider refresh if experiencing performance issues
10-20% drift: Monitor for trends, may trigger automatic collection soon
<10% drift: Normal variance, no action needed

Query 4: Identify Missing Column-Level Statistics

Finds table columns without statistics, focusing on columns frequently used in WHERE/JOIN clauses.

WITH table_columns AS (
  SELECT column_name
  FROM information_schema.columns
  WHERE table_schema = 'schema_name'  -- Replace
    AND table_name = 'table_name'    -- Replace
    AND is_hidden = 'NO'             -- Exclude internal columns
),
stats_columns AS (
  SELECT UNNEST(column_names) AS column_name
  FROM [SHOW STATISTICS FOR TABLE database_name.schema_name.table_name]
  WHERE column_names != '{}'  -- Exclude table-level stats
)
SELECT
  tc.column_name AS missing_column,
  'No statistics available' AS status
FROM table_columns tc
WHERE tc.column_name NOT IN (SELECT column_name FROM stats_columns)
ORDER BY tc.column_name;

Customization:

Replace schema_name, table_name, and database_name with target table

Interpretation:

Columns returned have no optimizer statistics
Prioritize creating statistics for columns used in:
- WHERE clause predicates (WHERE user_id = 123)
- JOIN conditions (JOIN orders ON users.id = orders.user_id)
- GROUP BY / ORDER BY expressions

Action: Generate CREATE STATISTICS commands (see Query 7)

Query 5: Histogram Coverage Analysis

Identifies columns with/without histogram data for range query optimization.

SELECT
  UNNEST(column_names) AS column_name,
  created,
  now() - created AS stats_age,
  CASE
    WHEN histogram_id IS NOT NULL THEN 'Has histogram'
    ELSE 'Missing histogram'
  END AS histogram_status
FROM [SHOW STATISTICS FOR TABLE database_name.schema_name.table_name]
WHERE column_names != '{}'  -- Exclude table-level stats
ORDER BY
  CASE WHEN histogram_id IS NULL THEN 0 ELSE 1 END,  -- Missing first
  created DESC;

Customization:

Replace database_name.schema_name.table_name

Key columns:

histogram_status: Indicates distribution data availability
stats_age: Time since histogram last updated

Interpretation:

Has histogram: Optimizer can estimate range scan selectivity (BETWEEN, >, <)
Missing histogram: Optimizer uses uniform distribution assumption (less accurate)
Automatic collection creates histograms; missing indicates very new column or disabled collection

Query 6: Multi-Column Statistics Detection

Identifies existing multi-column (composite) statistics for correlated columns.

SELECT
  column_names,
  created,
  now() - created AS stats_age,
  row_count,
  ARRAY_LENGTH(column_names, 1) AS column_count
FROM [SHOW STATISTICS FOR TABLE database_name.schema_name.table_name]
WHERE ARRAY_LENGTH(column_names, 1) > 1  -- Multi-column only
ORDER BY created DESC;

Customization:

Replace database_name.schema_name.table_name

Key columns:

column_names: Array of correlated columns
column_count: Number of columns in composite statistic

Interpretation:

Present: Multi-column statistics exist — either auto-collected for an index column group or manually created
Absent: No multi-column stats yet; may need manual creation for correlated non-indexed columns (e.g., city + state + zip)
Common use case: Manual stats on correlated columns that aren't covered by an index

See references/create-statistics-examples.md for multi-column creation patterns.

Query 7: Generate CREATE STATISTICS Recommendations

Produces ready-to-run CREATE STATISTICS commands for tables with stale or missing statistics.

WITH stale_tables AS (
  SELECT
    table_schema,
    table_name,
    created,
    now() - created AS stats_age
  FROM [SHOW STATISTICS FOR TABLE database_name.*]
  WHERE column_names = '{}'
    AND (created IS NULL OR now() - created > INTERVAL '7 days')  -- Adjust threshold
)
SELECT
  table_schema || '.' || table_name AS full_table_name,
  stats_age,
  'CREATE STATISTICS __auto__ FROM ' || table_schema || '.' || table_name || ';' AS create_command
FROM stale_tables
ORDER BY stats_age DESC NULLS FIRST
LIMIT 50;

Customization:

Replace database_name.* with schema pattern
Adjust INTERVAL '7 days' staleness threshold
Increase LIMIT for more recommendations

Output:

create_command: Copy-paste ready SQL command
__auto__: Uses automatic column selection (recommended default)

Execution:

Review generated commands before execution
Run during low-traffic periods for large tables (>10M rows)
Monitor job progress (see Query 6 for job monitoring)

Common Workflows

Workflow 1: Post-Bulk-Load Statistics Audit

Scenario: After bulk INSERT/COPY/IMPORT operation, validate statistics are current.

Steps:

Identify affected tables:

-- List tables modified in last 24 hours
SELECT DISTINCT table_schema || '.' || table_name AS full_table_name
FROM [SHOW TABLES]
WHERE table_schema = 'target_schema';  -- Replace

Check row count drift (Query 3): Run drift detection query for each affected table.

Generate and execute refresh commands (Query 7):

CREATE STATISTICS __auto__ FROM schema_name.table_name;  -- From Query 7 output

Monitor collection job:

SELECT job_id, status, fraction_completed, running_status
FROM [SHOW JOBS]
WHERE job_type = 'CREATE STATS'
  AND created > now() - INTERVAL '1 hour'
ORDER BY created DESC
LIMIT 10;

Verify refresh (Query 2): Re-run statistics audit to confirm created timestamp updated.

Expected outcome: Statistics age <1 hour, drift_pct <5%.

Workflow 2: Diagnose Unexpected Query Plan Changes

Scenario: Query performance suddenly degrades; EXPLAIN shows different plan.

Steps:

Identify affected query from profiling-statement-fingerprints: Find query with latency spike or plan hash change.
Extract table references: Parse query text to identify tables in FROM/JOIN clauses.
Audit statistics for each table (Query 2): Check staleness and row count currency.

Compare historical vs current row counts:

-- Example: Check if table grew significantly
SELECT row_count, created
FROM [SHOW STATISTICS FOR TABLE users]
WHERE column_names = '{}'
ORDER BY created DESC
LIMIT 5;  -- Last 5 collections

Refresh stale statistics (Query 7): Execute CREATE STATISTICS for tables with high drift.
Validate plan stability: Re-run EXPLAIN to verify plan returns to expected structure.

Expected outcome: Plan hash stabilizes, latency returns to baseline after statistics refresh.

Workflow 3: Routine Statistics Health Check

Scenario: Periodic audit to proactively identify statistics issues before performance degrades.

Steps:

Run cluster-wide staleness scan (Query 1):

-- All databases
SHOW STATISTICS FOR TABLE *.*;  -- Warning: May be slow on large clusters

Prioritize critical tables: Focus on high-traffic tables from profiling-statement-fingerprints.

Check automatic collection is enabled:

SHOW CLUSTER SETTING sql.stats.automatic_collection.enabled;  -- Should be true

Review pending auto-collection jobs:

SELECT job_id, description, status, fraction_completed
FROM [SHOW JOBS]
WHERE job_type = 'AUTO CREATE STATS'
  AND status IN ('pending', 'running')
ORDER BY created DESC;

Generate batch refresh script (Query 7): Save output to file for scheduled execution.
Schedule refresh during maintenance window: Execute generated CREATE STATISTICS commands during low-traffic period.

Frequency: Weekly for OLTP, monthly for OLAP.

Safety Considerations

Production-Safe Operations

SHOW STATISTICS:

Impact: Read-only, no cluster impact
Safe for production: Yes, run anytime without restrictions

CREATE STATISTICS:

Impact: CPU/IO-intensive, non-blocking table scans
Safety:
- Does NOT lock table or block writes
- Consumes resources (CPU, network, disk I/O)
- May impact query performance during collection on large tables
Best practices:
- Run during low-traffic periods for tables >10M rows
- Stagger execution (avoid creating statistics on many tables simultaneously)
- Monitor job progress and resource utilization

Resource Consumption

Small tables (<10K rows): Negligible impact, safe anytime

Medium tables (10K-10M rows): Seconds to minutes, minor impact

Large tables (>10M rows): Minutes to hours, plan accordingly:

Schedule during maintenance windows
Monitor cluster metrics (CPU, disk I/O) during collection
Use SHOW JOBS to track progress

Cancellation (if needed):

-- Find job ID
SELECT job_id, status, fraction_completed
FROM [SHOW JOBS]
WHERE job_type = 'CREATE STATS' AND status = 'running';

-- Cancel job (non-destructive, existing statistics remain)
CANCEL JOB 123456789012345678;

Batch Collection Best Practices

Avoid overwhelming cluster:

Collect statistics for 3-5 tables concurrently maximum
Wait for completion before starting next batch
Monitor cluster health metrics between batches

Example staggered script:

# Collect statistics in batches with delays
for table in table1 table2 table3; do
  cockroach sql -e "CREATE STATISTICS __auto__ FROM $table;" &
done
wait  # Wait for batch to complete

sleep 60  # Delay between batches

for table in table4 table5 table6; do
  cockroach sql -e "CREATE STATISTICS __auto__ FROM $table;" &
done
wait

See references/create-statistics-examples.md for detailed batch patterns.

Additional details

Further sections for this skill are in references/additional-details.md.

About this skill

Prerequisites

Core Concepts

Core Diagnostic Queries

Query 1: Identify Tables with Stale or Missing Statistics

Query 2: Audit Statistics for Specific Table

Query 3: Detect Row Count Drift

Query 4: Identify Missing Column-Level Statistics

Query 5: Histogram Coverage Analysis

Query 6: Multi-Column Statistics Detection

Query 7: Generate CREATE STATISTICS Recommendations

Common Workflows

Workflow 1: Post-Bulk-Load Statistics Audit

Workflow 2: Diagnose Unexpected Query Plan Changes

Workflow 3: Routine Statistics Health Check

Safety Considerations

Production-Safe Operations

Resource Consumption

Batch Collection Best Practices

Additional details