Imported from kestra-io/plugin-opensearch-lib (AGENTS.md). Install upstream with npx skills add kestra-io/plugin-opensearch-lib. Copyright stays with the author.

AGENTS.md

What

• Shared kernel library published as io.kestra.plugin:plugin-opensearch-lib, consumed by plugin-opensearch (OSS) and plugin-ee-opensearch (EE). It is not a loadable Kestra plugin — it ships no task, no trigger, no plugin docs or icons, and cannot be registered on a running Kestra instance on its own.
• Provides classes under io.kestra.plugin.opensearch.shared: OpensearchConnection (hosts, basic auth, headers, TLS trust, path prefix, strict deprecation mode) and BulkService (buffered bulk indexing with requests.count / records / requests.duration metrics).

Why

• OSS and EE each shipped a near-verbatim, but drifted, copy of the connection-building and bulk-indexing code. This library removes that duplication so both consumers evolve the connection and bulk logic in one place instead of drifting further apart.

Local rules

• This library is shared between OSS and EE only — it is not a general-purpose OpenSearch client wrapper. Do not add task/trigger classes, plugin docs, or plugin icons here; each consumer keeps its own.
• It is a plain library, not a plugin: no package-info.java with @PluginSubGroup, no shadowJar, no X-Kestra-* jar manifest, no Dockerfile/docker-compose.yml.
• This library is the single source of the OpenSearch client version (inherited from the io.kestra:platform BOM); neither consumer declares it anymore. Both inherit it transitively through the api dependency, so bumping it here changes the connection schema for OSS and EE simultaneously — bump deliberately and QA both.
• Preserve metric names (requests.count, records, requests.duration) byte-for-byte, and keep the stable, user-facing shape of error/log messages (e.g. the Indexer failed bulk: prefix, the header/host validation wording) — both consumers' users template outputs and grep logs on these. This does not forbid bounded improvements to error content: BulkService.logError intentionally caps the per-item errors it renders (MAX_LOGGED_ERRORS) and appends ... (N more errors omitted) rather than building an unbounded string for a large failed bulk.
• Must be released to Maven Central before either consumer can pin a released (non-SNAPSHOT) version; until then, consumers resolve it from mavenLocal()/the Sonatype snapshots repository.

References

• https://kestra.io/docs/plugin-developer-guide
• https://kestra.io/docs/plugin-developer-guide/contribution-guidelines