jafar-perf

Plugin jafar-perf from marketplace btraceio in btraceio/jafar-perf-box. Commands read from the repo's manifest — 1 month ago.

Skills (9)

• compare

  Decide whether a candidate JFR recording regressed against a baseline, and attribute the change to a frame or a metric. Use for before/after checks, "is this build slower", bisecting a performance regression, verifying that a fix actually helped, or any question involving two recordings of the same

• cpu

  Find where CPU time goes in a JFR recording, pprof profile or OTLP profile, and attribute it to call paths and threads. Use when triage shows high execution-sample counts, when the user asks "why is the CPU pegged", "what is the hot method", "where is the time going", or asks for a flamegraph or pro

• gc

  Analyse garbage collection pressure, pause times, heap sizing and allocation hotspots in a JFR recording. Use when triage reports high GC pressure, when the user asks about GC pauses, heap growth, allocation rate, OutOfMemoryError risk, or which code allocates the most.

• heap-diff

  Compare two or more heap dumps taken at different times to prove memory growth rather than infer it — class-level instance and retained-size deltas, newly appeared clusters, and objects that survived when they should not have. Use whenever two .hprof files of the same application are available, or w

• jfrpath

  Syntax reference for the query languages behind jfr_query, hdump_query, pprof_query and otlp_query — JfrPath, HdumpPath and SamplesPath. Consult before composing any non-trivial query, and whenever a query returns a parse error, so the syntax is right on the first attempt instead of after three fail

• latency

  Investigate response-time problems that are not CPU-bound — lock contention, thread parking, executor queue saturation, blocking I/O, and per-endpoint latency attribution. Use when the user reports slow requests, p99 spikes, timeouts, deadlock suspicion, or when triage shows threads blocked rather t

• memory-leak

  Hunt memory leaks and wasted heap in a Java heap dump (HPROF) — retained sizes, dominator tree, GC root paths, known leak patterns, duplicate strings, collection waste, and correlating retained objects back to their JFR allocation sites. Use for OutOfMemoryError, heap that never comes back after GC,

• report

  The output format and evidence discipline for any performance finding produced with the Jafar tools. Use whenever writing up an analysis, summarising an investigation, answering "what did you find", or handing conclusions to another person or agent.

• triage

  First step for any unfamiliar JFR recording, pprof profile, OTLP profile, or heap dump. Establishes what the artifact contains, what is anomalous, and which specialised playbook to run next. Use when the user says "analyse this recording", "what is wrong with this JVM", "why is this slow", or hands

Subagents (7)

• concurrency-analyst

  Specialist for thread and contention analysis of a JFR recording — thread states, monitor contention, parking, executor queue saturation, and per-endpoint latency attribution. Dispatch when triage shows threads blocked or waiting rather than running, or when the complaint is p99 latency rather than

• cpu-analyst

  Specialist for CPU-bound analysis of a JFR recording or sampling profile — hot methods, call paths, convergence points, and per-thread or per-endpoint attribution of execution samples. Dispatch when triage shows high execution-sample counts or a RUNNABLE-dominated thread state distribution.

• heap-analyst

  Specialist for heap dump analysis — retained sizes, dominator tree, GC root paths, known leak detectors, graph-based clusters, collection waste, duplicate subgraphs, heap-to-heap diffs, and correlating retained objects with JFR allocation sites. Dispatch for any .hprof file, OutOfMemoryError, or mem

• io-analyst

  Specialist for I/O analysis of a JFR recording — slow file and socket operations, per-destination latency and throughput, and separating dependency slowness from JVM problems. Dispatch when USE analysis flags I/O, or when latency correlates with external calls rather than with locks or CPU.

• memory-analyst

  Specialist for GC and allocation analysis of a JFR recording — pause distribution as a fraction of wall clock, heap behaviour over time, allocation rate and allocation hotspots by class and site. Dispatch when triage reports GC pressure, heap growth, or questions about allocation churn.

• perf-engineer

  General-purpose JVM performance analyst. Use for any single-artifact investigation of a JFR recording, heap dump, pprof or OTLP profile when you want one agent to triage, investigate and report end to end. For a broad investigation that should fan out across several dimensions at once, use perf-lead

• perf-lead

  Coordinator for a broad performance investigation. Triages an artifact, dispatches the specialist analysts the evidence justifies, then merges, ranks and de-duplicates their findings into one report. Use when the question is open-ended ("why is this service slow", "review this recording") rather tha

MCP servers (1)

• jafar

  runs: jbang jfr-mcp@btraceio --stdio

Similar plugins