← All release posts
Release By SkyWalking Team

Release Apache SkyWalking GraalVM Distro 0.5.0

SkyWalking GraalVM Distro 0.5.0 is released. Go to downloads page to find release tars.

This release is built on the Apache SkyWalking v11.1.0 release tag.

Upstream Sync

  • Sync SkyWalking submodule to the upstream v11.1.0 release tag (d34eff29ba).
  • TAL (Trace Analysis Language) on the native image. Wire upstream’s new trace-analyzer module (on by default, SW_TRACE_ANALYZER=- removes it) and pre-compile its rules like LAL: the precompiler compiles tal/otlp-general.yaml (otlp-general, otlp-slow-sql) and generates the 28 otlp_* meter classes at build time; the replacement TalDSL loads the pre-compiled class and creates its metrics, and upstream’s SpanFilterListener, pairing and receiver path run unchanged.
  • TAL rules are user-editable, so the pre-compilation is generic: every tal/*.yaml is compiled, whatever it declares (any metric function, labeled metrics with their labelLimit, percentile buckets / ranks, inline layerDefinitions used across files, alias or class-name input and output types), and files are installed in upstream’s order, which decides which rule runs first on a span.
  • Wire the two new analyzer modules: ai-evaluation (SWIP-16 LLM-as-judge over sampled GenAI spans; disabled by default, configured through the new config/ai-evaluation.yml) and ai-agent-conversation (AI agent conversations landed by the AI Sessionizer; on by default because the GraphQL query module requires it, turned off with its none provider).
  • Server clusters (apache/skywalking#14183). server-clusters.yml, and its dynamic configuration item core.default.serverClusters, name a message broker, database or cache that clients reach at several addresses as one service. The native trace analysis renames such peers, and on the OTLP path the new TAL function findServerClusterName(address[, kind]) does, which the shipped otlp-general and otlp-slow-sql rules call.
  • The shipped otlp-general rule counts a CONSUMER receive span that has a parent (the .NET Kafka instrumentation’s only record of a message) as the consumer’s request, reads messaging.system.name before messaging.system, reads aws_sqs / aws_sns as aws.sqs / aws.sns, and drops a CLIENT settle span as it drops a CLIENT receive.
  • HTTP alarm hook delivery self-observability: alarm_http_delivery_total, alarm_http_delivery_status_total and alarm_http_delivery_latency_seconds on telemetry.prometheus, with three new otel-rules/oap.yaml rules. TraceQL accepts the All wildcard on status and kind.
  • Adopt every rule file upstream added or changed, all enabled with the upstream defaults: lal/ai-agent.yaml, meter-analyzer-config/gen-ai-model.yaml (judge scores), otel-rules/ai-agent/runtime-{instance,service}.yaml and mcp_endpoint.yaml, otel-rules/banyandb/banyandb-trace-sampling.yaml, OpenTelemetry-native host and process monitoring (otel-rules/process-hostmetrics-{linux,windows}.yaml, reworked vm.yaml / windows.yaml), the Kafka, RabbitMQ, PostgreSQL, MySQL, Kong, APISIX, Zabbix and AI agent rule fixes, the AI-evaluation and alarm-delivery counters in otel-rules/oap.yaml, and oal/mesh.oal.
  • Mirror the new application.yml options: native OTLP trace storage (otlpTraceStorage: otlp is the new default; zipkin keeps the converted path), OTLP service grouping, searchable tags, head sampling and rate cap, the TraceQL /otlp datasource, the Kubernetes metadata read settings, the alarm hooks’ httpTimeout, the ai-agent-conversation read windows and hide, BanyanDB credential / trust-CA hot reload (secretsManagementFile), the recordsAIAgent BanyanDB group, the Zipkin cold-stage query, and the ports opening once boot completes.

GraalVM Native Image Compatibility

  • Embed every query-protocol/*.graphqls schema with one glob instead of a per-file list; upstream’s new gen-ai-evaluation-record and ai-agent-conversation schemas were missing and the native OAP died at boot.
  • The precompiler’s Armeria scan collects @Decorator targets (the AI agent conversation view’s CompressResponse): without the metadata the native OAP died at boot with “cannot inject the dependency”.
  • Register the admin-server family’s Jackson response POJOs for reflection: /inspect/* answered HTTP 400 (No serializer found for class ...MetricsResponse) in the native image since 0.4.0.
  • Register the protobuf descriptor closure of the OTLP/HTTP JSON receivers’ requests and of LogData: OTLP/HTTP JSON requests failed natively with Generated message class ... missing method getResourceLogsList. The same for ExportTraceServiceResponse (now JSON-printed by the OTLP/HTTP trace receiver) and Tempo’s TraceByIDResponse (the TraceQL /otlp trace-by-id route).
  • Register the ConversationFile LAL output builder for reflection, and package ai-evaluation.yml and server-clusters.yml in both distributions.
  • Register for reflection the TAL expression and meter classes, their configs, every input type, input provider and output builder the TAL rules resolve through, and PairingFact (TAL pairing halves cross OAP nodes as stream data) and SpanMetadata; a native rule calling toJson(meta) and toJson(span) returns the same JSON as the JVM OAP.
  • Register the status module’s Debugging*Rsp DTOs (trace, OTLP, Zipkin, log, MQE, topology): /debugging/* serializes them to YAML through Jackson, and none were covered since the 11.0.0 relocation.
  • Bundle ios-analyzer: the distro never shipped it, so the iOS HTTP and MetricKit span listeners (SWIP-11) did not run and an iOS app’s service_cpm, endpoint and instance traffic stayed empty.
  • Reflection metadata for three JSON paths: the admin /status/cluster/nodes answer (HTTP 500 without it), the /status/alarm/* answers ({} without them) and the AWS Firehose receiver’s request and answer (every post answered 400 without them).
  • Register the UI template REST bodies (DashboardSetting, DashboardConfiguration, TemplateChangeStatus) for reflection: on 0.4.0 POST / PUT /ui-management/templates failed natively.

Documentation

  • Document the new ai-evaluation, ai-agent-conversation, trace-analyzer and traceQL modules and TAL in distro-policy.md, configuration.md and internals/dsl-immigration.md (TAL rules are not live-debuggable because upstream emits no TAL probes), list server-clusters.yml among the config/ files, and refresh the build-time counts.
  • Correct the TLS docs: supported-features.md and the FAQ still called gRPC TLS unsupported, which 0.3.0 fixed. They now list what the native image serves: agent gRPC TLS, receiver gRPC TLS and mTLS, REST HTTPS, OAP-to-BanyanDB TLS.

All issues and pull requests are here