splunk-analysis

name: splunk-analysis description: Splunk log analysis using SPL (Search Processing Language). Use when investigating issues via Splunk logs, saved searches, or alerts. allowed-tools: Bash(python *)

Splunk Analysis

Authentication

IMPORTANT: Credentials are injected automatically by a proxy layer. Do NOT check for SPLUNK_HOST, SPLUNK_TOKEN, or other credentials in environment variables - they won't be visible to you. Just run the scripts directly; authentication is handled transparently.

MANDATORY: Statistics-First Investigation

NEVER dump raw logs. Always follow this pattern:

STATISTICS → SAMPLE → PATTERNS → CORRELATE

1. Statistics First - Know volume, error rate, and top patterns before sampling
2. Strategic Sampling - Choose the right strategy based on statistics
3. Pattern Extraction - Cluster similar errors to find root causes
4. Context Correlation - Investigate around anomaly timestamps

Available Scripts

All scripts are in .claude/skills/observability-splunk/scripts/

PRIMARY INVESTIGATION SCRIPTS

get_statistics.py - ALWAYS START HERE

Comprehensive statistics with pattern extraction.

python .claude/skills/observability-splunk/scripts/get_statistics.py [--index INDEX] [--sourcetype SOURCETYPE] [--time-range MINUTES]

# Examples:
python .claude/skills/observability-splunk/scripts/get_statistics.py --time-range 60
python .claude/skills/observability-splunk/scripts/get_statistics.py --index main
python .claude/skills/observability-splunk/scripts/get_statistics.py --sourcetype access_combined

Output includes:

• Total count, error count, error rate percentage
• Status distribution (info, warn, error)
• Top sourcetypes and hosts by log volume
• Top error patterns (crucial for quick triage)
• Actionable recommendation

sample_logs.py - Strategic Sampling

Choose the right sampling strategy based on statistics.

python .claude/skills/observability-splunk/scripts/sample_logs.py --strategy STRATEGY [--index INDEX] [--sourcetype SOURCETYPE] [--limit N]

# Strategies:
#   errors_only   - Only error logs (default for incidents)
#   warnings_up   - Warning and error logs
#   around_time   - Logs around a specific timestamp
#   all           - All log levels

# Examples:
python .claude/skills/observability-splunk/scripts/sample_logs.py --strategy errors_only --index main
python .claude/skills/observability-splunk/scripts/sample_logs.py --strategy around_time --timestamp "2026-01-27T05:00:00" --window 5
python .claude/skills/observability-splunk/scripts/sample_logs.py --strategy all --sourcetype access_combined --limit 20

SPL (Search Processing Language)

Basic Search

# Simple keyword search
error

# Index specific search (ALWAYS specify index for performance)
index=main error

# Multiple keywords (implicit AND)
index=main error connection

# Exact phrase
index=main "connection refused"

Field Searches

# Exact field match
index=main host=web-01

# Wildcard
index=main host=web-*

# Numeric comparison
index=main status>=400

# NOT operator
index=main NOT status=200

# OR operator
index=main (status=500 OR status=503)

Time Range

# Relative time (in tool call)
earliest=-15m latest=now

# Absolute time
earliest="01/15/2024:10:00:00" latest="01/15/2024:11:00:00"

# Natural time modifiers
earliest=-1h@h  # 1 hour ago, rounded to hour
earliest=-1d@d  # 1 day ago, rounded to day

Investigation Workflow

Standard Incident Investigation

┌─────────────────────────────────────────────────────────────┐
│ 1. STATISTICS FIRST (mandatory)                              │
│    python get_statistics.py --index <index>                  │
│    → Know volume, error rate, top patterns                   │
└─────────────────────────────────────────────────────────────┘
                             │
                             ▼
                     High Error Rate?
               ┌─────────────┴─────────────┐
               │                           │
       YES (>5%)                           NO
               │                           │
               ▼                           ▼
┌─────────────────────────────┐  ┌───────────────────────────────────────────┐
│ 2. FAST PATH                │  │ 2. TARGETED INVESTIGATION                 │
│    Sample errors directly   │  │    Filter by specific criteria            │
│    python sample_logs.py    │  │    python sample_logs.py --strategy all   │
│    --strategy errors_only   │  │    → Look for anomalies                   │
└─────────────────────────────┘  └───────────────────────────────────────────┘

Quick Commands Reference

SPL Commands Reference

Filtering Commands

Transformation Commands

Field Operations

Common Query Patterns

Error Rate Analysis

# Error count per 5 minutes
index=main | timechart span=5m count(eval(level="ERROR")) as errors, count as total

# Error percentage over time
index=main
| timechart span=5m count(eval(level="ERROR")) as errors, count as total
| eval error_rate=errors/total*100

Top Errors by Service

index=main level=ERROR
| stats count by service, message
| sort -count
| head 20

Response Time Analysis

index=main sourcetype=access_combined
| stats avg(response_time) as avg_rt,
        p95(response_time) as p95_rt,
        max(response_time) as max_rt
    by uri_path
| sort -avg_rt

Anomaly Detection

# Sudden spike detection
index=main
| timechart span=5m count as events
| eventstats avg(events) as avg_events, stdev(events) as stdev_events
| eval anomaly=if(events > avg_events + 2*stdev_events, 1, 0)
| where anomaly=1

Anti-Patterns to Avoid

1. ❌ NEVER skip statistics - get_statistics.py is MANDATORY first step
2. ❌ No index specified - Always use index=X for performance
3. ❌ Unbounded time range - Always specify time ranges
4. ❌ Fetching all logs - Use sampling strategies, not unbounded searches
5. ❌ Ignoring error rate - High error rate means immediate investigation
6. ❌ Complex rex on all events - Filter first, then extract