←

Engineering Blog

5 min read

DataBuff v0.1.8 Release Notes

ReleaseAPMAIOpsv0.1.8OpenTelemetry

DataBuff v0.1.8 is ready: log detail is pinned to an ingest-generated log_id (GitHub #73); LLM Trace prompt / completion is no longer a string of \uXXXX; the login footer and sidebar show the Maven build-info version. Remote MCP tools run inside the fat-jar. Versus v0.1.7: 32 commits, schema migration V008 (log_id).

Highlights

  • Log detail pinned to log_id (#73): ingest writes a 32-hex id on every log row; detail queries pin that row. time_ns is not unique on a DUPLICATE KEY table, so high-QPS LIMIT 1 could open someone else's row. Existing rows with empty id still use time_ns (+ service_id).
  • GenAI / LLM traces readable: spans whose attributes start with gen_ai. / gen.ai. / llm. decode unicode escapes at query time. Other spans are untouched.
  • Product version: Maven build-info on the login footer and sidebar user menu.
  • Remote MCP: MCP JSON SPI warmed so remote SSE / Streamable HTTP tools actually run from LLM chat in the fat-jar.
  • ResourceIgnoreFilter: ingest.trace.ignore-resources / ignore-resource-regex now bind to the filter bean.
  • SkyWalking SQL normalize: ingest-side SQL standardization aligned with the OTel path.
  • Instance Top 5: rankings group by service_instance.
  • Trace / Logs bar filter: click-to-filter can toggle and reset.
  • Docs: Nginx and eBPF ingest how-tos (how-tos, not new product surfaces).

How to use: log detail

After upgrading to v0.1.8, newly ingested logs carry log_id. Opening a row pins that id, then prunes by a time window. Apply V008 before rolling web / ingest — new Web SELECTs log_id.

Logs analysis list: volume, status charts and table
Fig 1 · Logs analysis list (detail drawer not open yet)
Log detail drawer matches the selected list row
Fig 2 · Drawer body, instance and Trace ID match the clicked row; lookup uses log_id, which is not shown as a column

How to use: LLM traces

If the probe follows the OpenTelemetry GenAI semantic conventions (gen_ai.* / gen.ai.* / llm.*), opening the Trace decodes \uXXXX at query time. There is no separate LLM analysis page in this release.

Trace detail shows gen_ai.prompt in plain text
Fig 3 · LLM Trace: gen_ai.prompt / completion as original text, not unicode escapes

How to use: version and instances

Login footer and sidebar user menu read Maven build-info. On service detail → Basic info → Top 5 by instance, rankings group by service_instance.

Login footer shows product version 0.1.8
Fig 4 · Login footer 0.1.8 comes from Maven build-info
Sidebar shows version 0.1.8
Fig 5 · Same version string in the sidebar after login
Service detail Basic info Top 5 by instance
Fig 6 · Basic info → Top 5 by instance; series are service-a-1 / 2 / 3

Install & upgrade

Fresh install: Docker install. Existing installs: upgrade & uninstall. Schema migration V008 (log_id + structured alert columns, schema_version → 8) runs automatically. Kubernetes does not support in-place upgrade this release; uninstall then reinstall.

# Fresh install (wipes install dir and data/)
curl -fsSL https://databuff.ai/databuff/ai-apm-install.sh | bash

# In-place upgrade (keeps observational data/)
curl -fsSL https://databuff.ai/databuff/ai-apm-update.sh | bash

Try it

GitHub Release v0.1.8 · Repository