Skip to content
Maintained by AxonOps — production-grade documentation from engineers who operate distributed databases at scale Get Cassandra Help Get Kafka Help

nodetool sjk

Invokes the Swiss Java Knife (SJK) diagnostic tool for JVM troubleshooting and performance analysis.

Synopsis

nodetool [connection_options] sjk [--] <args>

See connection options for connection options.

Arguments:

Argument Description
-- Optional separator to distinguish nodetool options from SJK arguments
<args> Arguments passed verbatim to Swiss Java Knife

The -- separator is useful when SJK arguments might be mistaken for nodetool options.

Description

nodetool sjk provides access to Swiss Java Knife (SJK), a collection of command-line tools for JVM diagnostics, monitoring, and profiling.

Background

SJK was integrated into Cassandra nodetool via CASSANDRA-12197. This integration allows operators to perform JVM diagnostics directly through nodetool without needing to install separate tools or manage additional JAR files on Cassandra nodes.

How It Works

When nodetool sjk is invoked:

  1. Nodetool establishes a JMX connection to the target Cassandra node using standard nodetool connection options (-h, -p, -u, -pw)
  2. The SJK arguments are passed to the embedded SJK library
  3. SJK commands that require JVM access use the JMX connection and process ID provided by nodetool
  4. Results are returned to the console

Connection Handling

Cassandra's nodetool integration manages the JMX connection. SJK's own connection options (--pid, --socket, etc.) are not available through nodetool sjk. The connection is always established through nodetool's standard mechanism.

Getting Help

To see available SJK commands for the installed Cassandra version:

# List all available commands
nodetool sjk --commands

# Get general help
nodetool sjk --help

# Get help for a specific command
nodetool sjk <command> --help

Version-Specific Commands

The available commands and their options depend on the SJK version bundled with the Cassandra distribution. Always use nodetool sjk --commands to see what is available in the current environment.

Common Commands

The following commands are typically available. Run nodetool sjk --commands to confirm availability and nodetool sjk <command> --help for exact option names.

ttop - Thread CPU Usage

Displays JVM threads sorted by CPU usage, similar to the Linux top command but for threads within a single JVM process.

# Show thread CPU usage
nodetool sjk ttop

# Show top N threads (check --help for exact option name)
nodetool sjk ttop --help

Use cases:

  • Identify threads consuming excessive CPU
  • Detect runaway compaction or streaming threads
  • Find threads blocked in contention
  • Monitor allocation rates to diagnose GC pressure

hh - Heap Histogram

Prints a histogram of objects on the heap, similar to jmap -histo. This is useful for identifying memory leaks or unexpected object retention.

Stop-the-World Pause

The heap histogram operation may cause a stop-the-world pause. Some options require a full garbage collection. Use with caution on production systems.

# Show heap histogram
nodetool sjk hh

# Show top N classes with live objects only
nodetool sjk -- hh --top-number 20 --live

# Get help for available options
nodetool sjk hh --help

Use cases:

  • Identify classes with unexpectedly high instance counts
  • Detect memory leaks (growing instance counts over time)
  • Analyze object retention patterns
  • Compare live vs dead objects to understand allocation patterns

gc - Garbage Collection Tracker

Reports garbage collection statistics in real time.

# Monitor GC activity
nodetool sjk gc

# Get available options
nodetool sjk gc --help

Use cases:

  • Monitor GC behavior under load
  • Detect long GC pauses affecting latency
  • Verify GC tuning changes
  • Correlate GC activity with application performance

stcap - Stack Trace Capture

Captures thread stack traces at regular intervals for profiling and analysis.

# Capture stack traces (check --help for options)
nodetool sjk stcap --help

# Example: capture to file for specified duration
nodetool sjk -- stcap --output /tmp/stacks.std --timeout 30000

Use cases:

  • Profile CPU-intensive operations
  • Identify hot code paths
  • Analyze thread behavior over time
  • Create data for flame graph generation

ssa - Stack Trace Analyzer

Analyzes stack trace files created by stcap.

# Analyze captured file
nodetool sjk -- ssa --file /tmp/stacks.std --help

# Generate histogram
nodetool sjk -- ssa --file /tmp/stacks.std --histo

mx - MBean Operations

Queries and invokes JMX MBean operations.

# Get help for MBean operations
nodetool sjk mx --help

# Example: list MBean info
nodetool sjk -- mx --bean "java.lang:type=Memory" --info

Practical Examples

Diagnosing High CPU Usage

# Step 1: Identify which threads are using CPU
nodetool sjk ttop

# Step 2: If specific threads are high, capture stack traces
nodetool sjk -- stcap --output /tmp/high-cpu.std --timeout 60000

# Step 3: Analyze the captured data
nodetool sjk -- ssa --file /tmp/high-cpu.std --histo

Investigating Memory Issues

# Step 1: Get heap histogram
nodetool sjk -- hh --top-number 30

# Step 2: Check for live objects only (triggers GC)
nodetool sjk -- hh --top-number 30 --live

# Step 3: Monitor GC behavior
nodetool sjk gc

Collecting Diagnostic Data

# Create diagnostic directory
mkdir -p /tmp/cassandra-diag

# Capture thread CPU usage
nodetool sjk ttop > /tmp/cassandra-diag/ttop.txt

# Capture heap histogram
nodetool sjk -- hh --top-number 50 > /tmp/cassandra-diag/heap.txt

# Capture stack traces for 60 seconds
nodetool sjk -- stcap --output /tmp/cassandra-diag/stacks.std --timeout 60000

Common Cassandra Thread Patterns

When analyzing thread output, these patterns indicate Cassandra-specific activity:

Thread Name Pattern Purpose
ReadStage-* Processing read requests
MutationStage-* Processing write requests
CompactionExecutor-* Running compactions
MemtableFlushWriter-* Flushing memtables
GossipStage-* Cluster gossip protocol
Native-Transport-* CQL client connections
HintsDispatcher-* Delivering stored hints

Best Practices

Production Usage

  • Use ttop and gc freely—they have minimal overhead
  • Use hh sparingly as it may cause stop-the-world pauses
  • Capture stack traces (stcap) during off-peak hours when possible
  • Always check --help for exact option names before scripting

Using the Separator

When SJK options conflict with nodetool parsing, use -- to separate them:

# Without separator (may cause parsing issues)
nodetool sjk hh --top-number 10

# With separator (explicit argument passing)
nodetool sjk -- hh --top-number 10

Troubleshooting

Command Not Recognized

If a command is not recognized, verify it is available:

nodetool sjk --commands

Connection Issues

SJK commands use the nodetool JMX connection. If commands fail:

# Verify nodetool connectivity first
nodetool status

# Check JMX settings
grep JMX /etc/cassandra/cassandra-env.sh

Option Errors

Option names may vary between SJK versions. Always check:

nodetool sjk <command> --help
Command Relationship
tpstats Thread pool statistics (Cassandra-specific)
gcstats GC statistics (Cassandra-specific format)
info General node information including memory
tablestats Table-level metrics

External Resources