pinpoint

Pinpoint 4.0.0 is a major release that opens the 4.x line. It has three highlights:

• Service, now on by default, groups applications so that each team can focus on its own service.
• OpenTelemetry Trace Support brings applications instrumented with OpenTelemetry into Pinpoint without the Pinpoint agent.
• Service Alarm is a renewed alarm system built around services. New Alarm Templates make it easy to set up the same alarms for many applications.

It also adds Spring Boot 4.1 / Spring Framework 7.0 support and includes the agent-stability work that shipped in 3.1.1.

This release changes defaults and removes several legacy code paths, so please read the Upgrade guide before upgrading in place from 3.1.x.

Highlights

Service (enabled by default)

Pinpoint 4.0.0 introduces the Service feature. You can group several applications into one service and see how services call each other on the Service Map. Screens that support the service feature show only the data of the applications in the selected service.

• Monitoring: Focus on the data of the service you own and find the cause of a problem faster.
• Operations: As applications grow, group them by service to keep the monitoring targets clear.

Monitoring targets form a Service → Application → Agent hierarchy. The demo-shop-commerce service of the demo shopping mall looks like this:

Service: demo-shop-commerce          products and orders
├─ Application: Shopping-Order       order processing
│  ├─ Agent: shop-order-01
│  └─ Agent: shop-order-02
└─ Application: Shopping-Product     product lookup
   └─ Agent: shop-product-01

Data that is not assigned to a service, such as data collected before the upgrade or sent by agents older than 4.0.0, stays in the DEFAULT service. Creating a new service does not move existing data into it. To use the service feature, create the service in the Web first, then set pinpoint.serviceName on the agents (see Agent). See the Service guide for details.

OpenTelemetry Trace Support

Pinpoint 4.0.0 receives OpenTelemetry traces directly. Applications instrumented with the OpenTelemetry Java agent, any OTel SDK, or the OpenTelemetry Collector appear in the same server map, transaction list and call tree as agent-collected data, without the Pinpoint agent.

• OTLP over gRPC and HTTP Traces are accepted over gRPC and over HTTP in both protobuf and JSON, with gzip and standard OTLP responses. Span time keeps nanosecond precision (#13923).

• URL Statistic and Error Analysis URL Statistic is built from OTel server spans using the route template (#14229). Exceptions recorded on OTel spans flow into Error Analysis with multi-language stacktrace parsing (JVM, Go, Node.js, Python) and grouping by route (#14233). SQL text is shown in the call tree.

• Server map and service map links Caller → callee edges between OTel-instrumented applications are drawn from a pp entry in the W3C tracestate header, written by the sender via pinpoint-otel-extension (OpenTelemetry Java agent extension, plus guides for Java SDK, Python, Go and Node.js). When the caller cannot propagate it, the collector and a batch job join the two sides and fill in the edge (#14453). OTel span links open the linked transaction (

Pinpoint 3.1.1 is a maintenance release of the 3.1.x line. It focuses on agent stability and overhead on reactive and coroutine workloads (Reactor / WebFlux / reactor-netty / Kotlin coroutines), Spring Framework 7 compatibility, and security hardening of the Web and Batch modules. There are no schema changes; collector and web can be upgraded in place from 3.1.0.

Highlights

Agent — Reactor megamorphic-dispatch remediation (JDK 21/25)

Backport of the master track that remediates the JIT megamorphic invokeinterface dispatch observed on reactive applications running on recent JDKs (C2 profile pollution across the generic reactor CoreSubscriber instrumentation).

• The reactor AsyncContext carrier is unified into a single injected field, so the hot interceptors no longer go through polymorphic accessor lookups.
• Each interceptor class now gets a generated monomorphic exception-guard wrapper instead of sharing one reflective guard. The code generation is enabled by default:

  # pinpoint.config (local / release profiles)
  profiler.interceptor.exception.guard.codegen=true   # set false to fall back to the 3.1.0 guard
  

• The generic reactor CoreSubscriber instrumentation can be switched off for a lightweight mode (profiler.reactor.subscriber.instrument=false, default true). Publisher-seam wrapping and scheduler-task tracing stay behind config gates that are off by default (profiler.reactor.trace.scheduler.task, profiler.reactor.trace.scheduler.task.periodic).
• DefaultAsyncTraceContext / DefaultRecorderFactory no longer call the Guice Provider.get() chain on every trace request; the singleton factories are injected or memoized (#14210).

Agent — "Corrupted call stack" storms fixed and throttled

Three layers address the Corrupted call stack found ... call stack is empty log floods (and the CPU spikes that came with them) reported on reactor-netty / WebFlux services:

• DefaultCallStack.peek() returns the disabled instance while overflowed frames are outstanding, instead of null, so a transaction that exceeds profiler.callstack.max.sequence no longer logs a stack dump per span event (#14310).
• OnErrorSubscriberInterceptor (onErrorResume / onErrorReturn / onErrorMap / onErrorComplete) carries its TraceBlock from before() to after() instead of looking the AsyncContext up twice, which closed a frame nobody opened on every failing WebClient response (#14319).
• The corrupted-call-stack WARN and stack dump are throttled to one per 3 s per trace type; suppressed dumps are counted and reported on the next emitted one, and skip the exception construction entirely.

Agent — Kotlin coroutines

ResumeWithInterceptor keeps the span event opened in before() and closes exactly that one in after(), so a nested Reactor callback swapping the current trace during BaseContinuationImpl.resumeWith(...) no longer leaves unbalanced blocks (previously surfacing as corrupted call stacks on the parent ChildTrace).

Agent — Spring Framework 7 compatibility

• WebFlux and RestTemplate header adaptors no longer throw NoSuchMethodError on Spring Framework 7, where HttpHeaders stopped implementing MultiValueMap (#14278).
• The RestTemplate plugin matches the new RestTemplate(Iterable<HttpMessageConverter<?>>) constructor, so RestTemplates built by Spring Boot 4 / Framework 7 are traced again.

Agent — Plugin fixes

• Redisson: reactive command tracing works on Redisson 3.17+ and 3.19+ (ReactiveProxyBuilder callback signature changes); it had been silently inactive on modern versions.
• reactor-netty / netty HTTP client: HttpClientRequestWrapper.toRemoteHost() and HttpClientHandlerRequestWithBodyInterceptor no longer cast a non-inet remote address (e.g. Unix domain sockets) to InetSocketAddress, which threw ClassCastException inside the interceptor.
• Duplicate accessor member injection is skipped in `ASMC

Highlights

ApplicationName: 254-Character Support

-Dpinpoint.modules.uid.version=v3
# This property must be added when starting the agent.
# It is compatible with collector versions 3.1.0 and higher.

The maximum length of applicationName has been expanded from 24 to 254 characters, aligning it with SERVICE_NAME_MAX_LEN.

Impact: Users with long, structured application identifiers (e.g., team.service.environment.region) can now register them as-is, removing the prior naming constraint that forced abbreviations.

ServerMap V3 — Performance Improvements

ServerMap has been rebuilt as ServerMap V3 with a redesigned storage layout and query path.

• New Map Tables: MapAppSelf, MapAgentSelf, MapAppOut, MapAppIn, MapAppHost
• Schema: hbase-create-snappy.hbase

Impact: Faster initial render and time-range navigation on large topologies, lower HBase read amplification, and improved support for 254-character ApplicationName values.

GrpcBatchDataSender — Unary Span Transport

A new span transport, SpanBatchGrpcDataSender, replaces the long-lived gRPC bidirectional stream (SpanGrpcDataSender) with a unary SendSpanList RPC that ships spans in independent batches.

Starting from 3.1.0, BATCH is the default in both local and release profiles:

# pinpoint-root.config
profiler.transport.grpc.span.sender.type=BATCH    # BATCH | STREAM

Collector-side support: the receiver-side SpanService exposes the new SendSpanList RPC and returns batch-level results. Requires a 3.1.0+ collector. Older collectors should keep sender.type=STREAM.

Impact: simpler operational model (no stream restarts on partial failures), better resilience behind L7 proxies and rolling collector deploys, and tunable batching that lets operators trade latency versus RPC overhead per workload.

OpenTelemetry Metric Collection

Pinpoint Collector now supports receiving metrics via the OTLP protocol. Applications can send metrics using the OpenTelemetry SDK or OpenTelemetry Collector without requiring the Pinpoint Agent.

📖 Manual

Heatmap View

A new Heatmap view is now available alongside Scatter Chart. While Scatter Chart plots individual transactions, Heatmap provides a statistics-based visualization for faster data overview.

📖 Manual

Error Reason Categorization

Errors are now categorized (e.g. exception, http-status, sql) and the category is shown on the call stack in the web UI, so you can tell at a glance why a span was marked as an error.

Opt-in via agent configuration:

profiler.error.enable=true

Choose which categories should mark a span as an error (default: all enabled):

# Allow-list
profiler.error.mark=http-status,sql
# Or deny-list
profiler.error.mark.exclude=exception

JDBC SQL Cache Expiration

The SQL cache now expires entries after a configurable time, causing the agent to re-send the metadata periodically.

# Default: 168 hours (7 days). Set 0 or negative to disable.
profiler.jdbc.sqlcacheexpirehours=168

TraceIndex — Scatter Chart and Transaction List Performance Improvements

Scatter Chart and Transaction List now use the new TraceIndex table for transacti

What's Changed

• [#13054] Migrate to Central Portal Publisher by @intr3p1d in https://github.com/pinpoint-apm/pinpoint/pull/13111
• [#13054] Prepare 3.0.5-SNAPSHOT by @intr3p1d in https://github.com/pinpoint-apm/pinpoint/pull/13113
• [#13275] Backport: Update agent interceptable target by @jaehong-kim in https://github.com/pinpoint-apm/pinpoint/pull/13276
• [#13355] Backport: Update s3 http auth header by @jaehong-kim in https://github.com/pinpoint-apm/pinpoint/pull/13356
• [#13483] Backport: Fix grpc-plugin ServerStreamGetter by @jaehong-kim in https://github.com/pinpoint-apm/pinpoint/pull/13484
• [#13496] Backport: Update agent grpcSpanSender state init by @jaehong-kim in https://github.com/pinpoint-apm/pinpoint/pull/13498
• [#13511] Backport: Update to reduce the occurrence of span events rel… by @jaehong-kim in https://github.com/pinpoint-apm/pinpoint/pull/13513
• [#13546] 3.0.5 release by @intr3p1d in https://github.com/pinpoint-apm/pinpoint/pull/13548

Full Changelog: https://github.com/pinpoint-apm/pinpoint/compare/v3.0.4...v3.0.5