← Back to all sparks
A

Apache SkyWalking

ANALYTICS
Velocity0.0

Observability platform for distributed systems and microservices

SkyWalking is rebuilding its own foundations — its own database, its own runtime, and now GenAI traces

observabilitybanyandbgenai-tracingapmbreaking-changes
Current state
Apache SkyWalking ships roughly one major a year. 10.4.0 added GenAI observability, replaced the Groovy-dependent runtime with a new OAL V2 engine, and became compatible with Grafana Tempo. Before it, 10.3.0 landed a new trace model in BanyanDB, and 10.2.0 removed the H2 storage option permanently while deepening BanyanDB support.
Where it's heading
Two rewrites are running at once. Storage is consolidating onto BanyanDB, SkyWalking's purpose-built database, with alternatives being removed rather than deprecated. The query and aggregation layer is moving off Groovy onto a typed, immutable OAL V2 engine with real error locations. GenAI observability arriving on top of that suggests the foundations work was clearing room for new telemetry types.
Prediction
Expect the next major to extend GenAI observability and continue narrowing supported storage backends toward BanyanDB, with further OAL V2 migration on the way.

Recent moves

  1. 4mo ago

    10.4.0 - GenAI Observability, Groovy-Free Runtime and Grafana Tempo Compatible

    ⚡ SPARK

    10.4.0 pairs a new telemetry domain — GenAI observability — with an OAL V2 engine that replaces string-based filters and Groovy-era behaviour with immutable ASTs, typed enums and precise error locations. Grafana Tempo compatibility opens it to teams already standardised elsewhere.

    View source ↗
  2. 7mo ago

    10.3.0 - New Trace Model in BanyanDB

    A new trace model in BanyanDB, presented with deployment evidence at around 200 services. This is the release where SkyWalking's own storage engine is claimed to be ready for large-scale use rather than being the experimental option.

    View source ↗
  3. 1y ago

    10.2.0 - No H2, More BanyanDB

    ⚡ SPARK

    10.2.0 removed H2 storage permanently and added BanyanDB 0.8.0 support, closing off the easy default that small deployments relied on. It also added doc_values only for Elasticsearch fields that need sorting or aggregation.

    View source ↗