v0.1.9

Changelog

• bf56287 Merge pull request #7 from lambdadb/chore/update-go-sdk-v0.7.2
• 477ceb5 chore: update LambdaDB Go SDK to v0.7.2
• 14d0cee chore: update LambdaDB Go SDK to v0.8.0 (#11)
• b60a38d ci: gate main pushes and releases on a fixture live smoke (#9)
• de6cddf fix: reject unsupported vector similarities before LambdaDB writes (#10)

v0.1.8

v0.1.8

Upgrade to Go SDK v0.7.0 and support explicitly configured native embedding indexes in migration mappings. Stored-vector migration and existing search defaults remain unchanged.

Changes

• Accept type: vector with embedding under payload.indexConfigs, in JSON or YAML. Omitted managedEmbedding enables native embeddings without adding the flag or embedding similarity/dimension defaults.
• Preserve explicit legacy managedEmbedding: true, legacy SDK helpers and normalized server metadata. Reject false with embedding, misplaced top-level native dimensions/similarity and unsupported mapping options.
• Preserve source text and stored vectors through transformation, upsert and bulk serialization. Keep source-generated mappings as ordinary vector indexes and reject native payload indexes that collide with mapped source vectors.
• Add focused serialization, schema, transport and error-handling tests and native mapping examples. Ordinary dense/sparse query-overlap validation still omits rerank and top-level candidateSize. No Bayesian search or schema-update workflow is added.

Migration notes

• Native embedding generation is explicit and may incur provider calls. Set embedding.sourceField to the destination text field after renaming/normalization.
• Native collections require --migration.write-mode upsert. The pinned backend rejects native bulk at upload-info retrieval, surfaced as a wrapped SDK APIError with HTTP status 400. Existing bulk defaults and error classification are unchanged.
• Native fields reject directly supplied vectors and queryVector. Preserve imported vectors in ordinary vector fields; existing target schemas are not reconciled or updated.
• Retry policy is unchanged. Retried upserts can invoke the provider again; no provider-call idempotency or quality guarantee is added.
• Go SDK v0.7.0's Bayesian support does not change migration verification or make Bayesian the default.
• The Homebrew tap is updated separately after stable release assets are published and verified.

Compatibility and evidence

Release source: fb9c557cb74d9dae7725f6681077ab58c8615de0 (PR #6). SDK: github.com/lambdadb/go-lambdadb@v0.7.0, release commit fd3000bcb4353b3d6ed23fcca4f111955b36b33e.

Backend contract: 9072a1bc8925954369a887f558f1eaf387b7ea0e. Development validation is scoped to the verified dev-aws-apne2-v3 deployment, not production availability. Validation on 2026-10-06:

• Native and legacy migration paths passed with real generated embeddings, preserved stored vectors/text, query overlap 1.0 and completed-checkpoint resume.
• All six maintained Qdrant-to-LambdaDB live scenarios passed, including ordinary bulk ingestion; local mock/checkpoint scenarios also passed without skips.
• The initial native bulk test expectation was corrected after confirming the existing upload-info endpoint returns APIError(400), rather than BadRequestError. Production behavior was unchanged.
• Temporary projects, separately issued keys and collections were removed; zero collections, key revocation and project GET 404 were verified. Persistent CI resources were preserved.
• Go tests/vet/build and post-merge CI passed. Four-platform prepublication archives passed integrity/header checks; macOS arm64 and both Linux architectures executed version checks (Linux amd64 under emulation).

Immediate numDocs aggregate freshness, real Pinecone/Elasticsearch migrations, production availability, macOS amd64 execution and provider quality/usage/idempotency were not established.

Full changes: v0.1.7...v0.1.8

v0.1.7

v0.1.7

Changes

• Upgrade the LambdaDB Go SDK from v0.4.0 to v0.6.0 and support all 49 fixed text analyzer presets in mapping validation and collection schema serialization.
• Preserve Elasticsearch field analyzers and index defaults through inventory and generated mappings, including implicit analyzer: default resolution to standard.
• Report custom or unsupported source analyzers with warnings and invalid mapping markers requiring an explicit choice. Report untranslated search analyzers and Qdrant text tokenizer options; display inventory warnings during migration and dry-run.
• Preserve omitted/null analyzer defaults, explicit empty arrays, exact case, order and duplicates. Distinguish the keyword text analyzer from keyword fields.
• Add JSON/YAML mapping and collection wire tests for every preset, defaults and unsupported options, plus dense/sparse verification-search compatibility coverage. Verification searches do not rerank.
• Document Homebrew installation, updates, uninstall and the manual tap maintenance process.

Upgrade notes

• Elasticsearch source credentials now need read access to both /_mapping and /_settings. Unreadable settings fail inventory rather than assuming standard.
• Custom analyzer pipelines/options are not translated. Generated unsupported:<source-name> markers fail validation until reviewed and replaced with an explicit supported target preset.
• Separate search analyzers remain untranslated and are reported. Preset names do not guarantee identical tokenization across engines or versions. Nepali/Tamil/Telugu are LambdaDB Lucene extensions, not shared Elasticsearch/OpenSearch presets; separate OpenSearch compatibility is unverified.
• Confirm that the intended LambdaDB target supports the selected presets before an actual migration. SDK publication and server merge 55d888299fee44466326a9db8016af9811ade13b do not establish deployment.
• Existing dense/sparse verification behavior and checkpoint formats are unchanged. This release performs no live migration and introduces no reranking.
• The Homebrew tap is maintained separately and may lag this GitHub release until its manual formula update is reviewed.

Full changes: v0.1.6...v0.1.7

v0.1.6

v0.1.6

Changes

• Upgrade the LambdaDB Go SDK to v0.4.0 and support the current collection API: HTTP 201 creation, wrapped collection metadata, and no collectionStatus polling.
• Preserve SDK-managed signed bulk-upload headers and Content-Type: application/json, with regression coverage for upload retries, completion retries, and storage 412 responses.
• Fix completed-checkpoint resume: rerunning a completed migration no longer replays the final source page or inflates accepted-record counts.
• Retain validation samples in checkpoints so failed post-migration validation can be retried without uploading the documents again. Checkpoint files now use owner-only permissions.
• Expand regression coverage for partial-write recovery, empty terminal pages, completed reruns, validation retries, restart, and checkpoint cleanup.

Upgrade notes

• This version targets the current LambdaDB API supported by Go SDK v0.4.0. Deployments that still expose the older collection creation/status contract need an API upgrade first.
• Checkpoints add optional sourceDone and validationSamples fields. Older checkpoint files remain readable and resume from their saved cursor.
• A checkpoint from an already completed migration created by an older CLI has no completion marker. Use --migration.restart for a fresh run; completion is not inferred from document counts.
• Use --migration.restart to migrate new source data after a completed run.
• Validation-enabled migrations store bounded sample documents locally. A completed checkpoint without saved samples requires restart for sample validation, or --migration.validation-sample-size=0 for count-only validation.

Full changes: v0.1.5...7256838

v0.1.5

Changelog

• e5b4220 Add Elasticsearch migration source
• bf887aa Document Pinecone namespace migration behavior

v0.1.4

Changelog

• d8ca27d Add Pinecone migration source
• b453216 Validate Pinecone sparse query overlap
• bb52f3b Validate sparse query overlap

v0.1.3

Changelog

• 44e2c96 Require LambdaDB connection settings

v0.1.2

Changelog

• 1e53dca Fix migration create collection flag

v0.1.1

Changelog

• f310b0b Add installer uninstall option
• 6b6b9f7 Improve install quickstart docs

v0.1.0

Changelog

• fcad974 Add CI and checkpoint cleanup coverage
• 7324ff5 Add GitHub release installer
• 98e423e Add Qdrant integration fixture
• ca7906c Add migration mapping validation
• 6ba8e88 Add release workflow and migration examples
• 801bd1a Add structured validation reports
• 8905e47 Add validation reports and query overlap checks
• 1fa8dd4 Expand migration validation and release setup
• d087187 Fix checkpoint cursors and split write batches
• 196d62c Handle Qdrant legacy dense vector output
• 279638c Harden Qdrant to LambdaDB migration
• 5762535 Initial LambdaDB migration scaffold
• a807423 Normalize LambdaDB field names
• d1d5401 Polish docs before publishing
• 41be3d6 Support YAML mapping files
• 248f09a Update handoff after YAML support
• 43f9d1c Update handoff after batch work
• 47e233d Update handoff after integration fixture
• ec30b5a Update handoff after live integration test