JVM introspection

Install

emisar pack install fetches this pack, re-validates it, and verifies its content hash against the --hash below — the exact bytes this page was rendered against, so a tampered copy is rejected — before copying it into the runner's packs dir. The command reloads a running daemon itself; no manual restart.

sudo emisar pack install java-jvm --hash sha256:a8eacb2f1978bdadf7121e5869bee29bd0a461d3f36d50f241d64a2cfbdae03e

Setup

Attaches to local JVM processes on the runner host with the JDK tools (jcmd/jstack/jmap/jstat/JFR) — no credentials needed; the target JVM's pid is passed as an argument.

sudo emisar pack verify java-jvm

Install and configure a pack walks through the whole sequence on a host.

• jvm.classloader_stats exec low Low — read-only or trivially reversible

  jcmd VM.classloader_stats

  Show per-classloader statistics — instance count, loaded classes, and chunk/block bytes. Use to spot a class-loader leak. Uses the attach API, so it works on JVMs running -XX:+PerfDisableSharedMem (the Cassandra default) where jvm.jstat_class fails.

  View source on GitHub
• jvm.heap_dump exec critical Critical — data loss or irreversible

  jmap -dump:live (heap dump)

  Trigger a full GC, then dump the live heap to /tmp/<pid>-heap.hprof. Can be many GB. Causes a long STW pause (seconds). Use only when you have analyst capacity to load the .hprof and disk space to hold it.

  View source on GitHub
• jvm.heap_summary exec low Low — read-only or trivially reversible

  jcmd GC.heap_info

  Show heap geometry — used/committed per generation (Eden, Survivor, Old, Metaspace) via the attach API. Works on JDK 9+ where `jmap -heap` was removed, and on JVMs running -XX:+PerfDisableSharedMem (e.g. Cassandra).

  View source on GitHub
• jvm.jcmd_gc_run exec medium Medium — changes state, easily reversible

  jcmd <pid> GC.run

  Trigger a full GC. Use to compare heap usage before/after, or to force collection ahead of a heap dump for cleaner output. Causes a pause whose length depends on heap size + collector tuning.

  View source on GitHub
• jvm.jfr_dump exec medium Medium — changes state, easily reversible

  Dump active JFR recording

  Dump the active "emisar" recording to its current file. Use mid-flight to grab a snapshot without stopping the recording.

  View source on GitHub
• jvm.jfr_start exec medium Medium — changes state, easily reversible

  Start a Java Flight Recording

  Start a 60-second Java Flight Recording named "emisar". JFR is low-overhead (~1%). Pair with `jvm.jfr_dump` to retrieve the recording, then `jvm.jfr_stop` to end it.

  View source on GitHub
• jvm.jfr_stop exec medium Medium — changes state, easily reversible

  Stop active JFR recording

  Stop the "emisar" recording and finalize the .jfr file.

  View source on GitHub
• jvm.jmap_histo_live exec medium Medium — changes state, easily reversible

  jmap -histo:live (top 50)

  List top classes by retained-heap size after a forced full GC. Use to find the leak candidate.

  View source on GitHub
• jvm.jstack exec low Low — read-only or trivially reversible

  jstack -l (thread dump)

  Dump full thread state with lock info. The canonical "what are all my threads doing?" diagnostic.

  View source on GitHub
• jvm.jstack_blocked exec low Low — read-only or trivially reversible

  jstack filtered to BLOCKED threads

  Run jstack and filter output to threads in BLOCKED state plus their lock owners. Use to find deadlocks fast.

  View source on GitHub
• jvm.jstat_class exec low Low — read-only or trivially reversible

  jstat -class

  Show class-loader stats — loaded class count and bytes. Use to spot a class-loading leak. Needs jvmstat perf data: fails with "<pid> not found" on JVMs running -XX:+PerfDisableSharedMem (the Cassandra default) — there use jvm.jmap_histo_live (live-object histogram) or JFR instead.

  View source on GitHub
• jvm.jstat_gc exec low Low — read-only or trivially reversible

  jstat -gc (10 samples)

  Show ten 1-second samples of GC stats. Surfaces young + old GC time, survivor sizes, metaspace usage. Use to spot a GC storm. Needs jvmstat perf data: fails with "<pid> not found" on JVMs running -XX:+PerfDisableSharedMem (the Cassandra default) — there use jvm.heap_summary for heap state, or JFR (jvm.jfr_start / jvm.jfr_dump) for GC behavior over time.

  View source on GitHub
• jvm.perfdata exec low Low — read-only or trivially reversible

  jcmd PerfCounter.print

  Show JVM performance counters — class loading, JIT, GC. Cheaper than jstat for one-shot. Reads the same jvmstat perf data as jstat: fails on JVMs running -XX:+PerfDisableSharedMem (the Cassandra default) — there use the attach-based reads (jvm.heap_summary, jvm.vm_flags, jvm.jstack) or JFR.

  View source on GitHub
• jvm.thread_count exec low Low — read-only or trivially reversible

  OS thread count for a JVM

  Count /proc/<pid>/task entries — fastest way to spot "do we have an unbounded thread pool?"

  View source on GitHub
• jvm.vm_flags exec low Low — read-only or trivially reversible

  jcmd VM.flags

  Return the JVM command-line and tuning flags. Use to confirm "is -Xmx set to what we expect?"

  View source on GitHub
• jvm.vm_system_properties exec high High — service-affecting

  jcmd VM.system_properties

  Dump every -Dkey=value system property the JVM was started with (jcmd VM.system_properties = System.getProperties()). Use to confirm config overrides took effect — but note this deliberately surfaces every launch -D property, which routinely carries secrets (spring.datasource.password, javax.net.ssl.keyStorePassword, JDBC / cloud credentials). Scope it by policy. The runner's redaction is a fail-closed backstop, not a guarantee — it is pattern-bound and can miss a bespoke secret whose name and value match no rule.

  View source on GitHub
• jvm.vm_uptime exec low Low — read-only or trivially reversible

  jcmd VM.uptime

  Show how long this JVM has been running.

  View source on GitHub