Investigate the path to replace olivere/elastic driver

[yurishkuro]

The olivere/elastic library is deprecated, but our Elasticsearch/OpenSearch storage implementation is heavily dependent on it. Bugs like this #2192 (comment) cannot be resolved.

Objective: Investigate available options for Go drivers for ES/OS and compile a report so that we can build a roadmap around it.

• Which libraries are available for use
• What are their version compatibility profiles for the ES/OS versions we currently support
• Is there one library that can continue to work across both ES and OS
• What are the differences in how that library works compared to olivere/elastic
• How much changes would we need to make internally

Worth noting that our use of olivere/elastic is indirect, we have a shim layer internal/storage/elasticsearch/client/interfaces.go that attempts (only partially) to abstract away the underlying driver.

[thc1006]

I investigated the available Go client libraries for replacing olivere/elastic. Here's what I found:

Available Libraries

Two official clients are actively maintained:

• elastic/go-elasticsearch (v9.1.0) - Elasticsearch only
• opensearch-go (v4.5.0) - OpenSearch only

Version Compatibility

The go-elasticsearch client supports ES 6.x through 9.x, while opensearch-go supports OS 1.x through 3.x. Unfortunately, Elastic added product detection in v7.13.2 (December 2021) that explicitly rejects OpenSearch connections.

Single Library for Both?

No. The product detection header check makes it impossible to use go-elasticsearch with OpenSearch. Downgrading to v7.13.1 would leave issue #2192 unfixed.

API Differences

The main changes from olivere/elastic:

• BulkProcessor → BulkIndexer (with proper FlushBytes hard limits - fixes Request Entity Too Large when connecting to AWS ElasticSearch #2192)
• Fluent DSL → Typed API structs (go-elasticsearch) or JSON queries (opensearch-go)
• Different query construction patterns

I found 41 query construction sites that would need updating.

Internal Changes Required

I analyzed the codebase and found Jaeger already partially migrated to go-elasticsearch/v9:

• Template creation for ES 8+ already uses v9 (config.go:283-289)
• Dual-client infrastructure exists (wrapper.go:62-69)
• Bulk operations still use olivere/elastic (issue Request Entity Too Large when connecting to AWS ElasticSearch #2192 remains unfixed)

Migration would affect approximately 38 files, mainly query DSL changes in:

• config/config.go - BulkProcessor migration (priority 1)
• wrapper/wrapper.go - Shim layer updates
• v1/spanstore/reader.go - Query DSL (~1800 lines)
• metricstore/query_builder.go - Aggregations (~600 lines)

Recommendation

Dual-client strategy:

• Use go-elasticsearch/v9 for Elasticsearch
• Use opensearch-go/v4 for OpenSearch
• Runtime detection (already partially implemented)

Estimated effort: 8-12 weeks with 1 developer, prioritizing BulkIndexer migration to fix #2192 first.

I've documented the full investigation at https://github.com/thc1006/jaeger/blob/research/jaeger-7612/docs/jaeger-7612/ES_OS_CLIENT_INVESTIGATION.md