test(#74): E2E Karate test — PDF → chunks indexed in OpenSearch

Full ingestion workflow: upload → analyze → rechunk → POST /api/ingestion
→ assert chunksIndexed > 0 and embeddingDimension > 0. Idempotency verified
with a second ingest. Failure scenario for non-COMPLETED job. @ingestion tag
for selective CI execution.

e2e/api/src/test/java/E2ERunner.java

@@ -4,7 +4,14 @@ class E2ERunner {
 
     @Karate.Test
     Karate testAll() {
-        return Karate.run("classpath:health", "classpath:documents", "classpath:analyses", "classpath:workflows")
+        return Karate.run("classpath:health", "classpath:documents", "classpath:analyses", "classpath:workflows", "classpath:ingestion")
+                .relativeTo(getClass());
+    }
+
+    @Karate.Test
+    Karate testIngestion() {
+        return Karate.run("classpath:ingestion")
+                .tags("@ingestion")
                 .relativeTo(getClass());
     }
 

e2e/api/src/test/resources/ingestion/ingestion-pipeline.feature

@@ -0,0 +1,125 @@
+@e2e @ingestion
+Feature: Ingestion pipeline — PDF → chunks indexed in OpenSearch
+
+  Tests the complete ingestion workflow:
+  upload PDF → analyze → chunk → ingest → verify in OpenSearch.
+
+  Requires the full dev stack (OpenSearch + embedding service running).
+  Skip by excluding the @ingestion tag when OpenSearch is not available.
+
+  Background:
+    * url baseUrl
+
+  Scenario: Full ingestion workflow — PDF becomes searchable chunks in OpenSearch
+
+    # ---- Step 1: Check ingestion pipeline is available ----
+    Given path '/api/ingestion/status'
+    When method GET
+    Then status 200
+    And match response.available == true
+
+    # ---- Step 2: Upload a PDF ----
+    Given path '/api/documents/upload'
+    And multipart file file = { read: 'classpath:common/data/generated/medium.pdf', filename: 'ingest-test.pdf', contentType: 'application/pdf' }
+    When method POST
+    Then status 200
+    And match response.id == '#string'
+    And match response.filename == 'ingest-test.pdf'
+    * def docId = response.id
+
+    # ---- Step 3: Run analysis (with document JSON for chunking) ----
+    Given path '/api/analyses'
+    And request { documentId: '#(docId)', pipelineOptions: { doOcr: false, tableMode: 'fast' } }
+    When method POST
+    Then status 200
+    And match response.status == 'PENDING'
+    * def jobId = response.id
+
+    # ---- Step 4: Poll until analysis is COMPLETED ----
+    Given path '/api/analyses', jobId
+    And retry until response.status == 'COMPLETED' || response.status == 'FAILED'
+    When method GET
+    Then status 200
+    And match response.status == 'COMPLETED'
+    And match response.hasDocumentJson == true
+
+    # ---- Step 5: Run chunking to produce chunk list ----
+    Given path '/api/analyses', jobId, 'rechunk'
+    And request { chunkingOptions: { chunkerType: 'hybrid', maxTokens: 512, mergePeers: true } }
+    When method POST
+    Then status 200
+    And match response == '#[_ > 0]'
+    * def chunkCount = response.length
+
+    # ---- Step 6: Trigger ingestion ----
+    Given path '/api/ingestion', jobId
+    When method POST
+    Then status 200
+    And match response.docId == '#string'
+    And match response.chunksIndexed == '#number'
+    And match response.embeddingDimension == '#number'
+    And match response.chunksIndexed > 0
+    And match response.embeddingDimension > 0
+    * def indexedCount = response.chunksIndexed
+    * def embDim = response.embeddingDimension
+
+    # Verify at least as many chunks as returned by rechunk
+    * assert indexedCount >= 1
+
+    # Embedding dimension must be a valid vector size (e.g. 384 for all-MiniLM-L6-v2)
+    * assert embDim >= 128
+
+    # ---- Step 7: Re-ingest is idempotent ----
+    Given path '/api/ingestion', jobId
+    When method POST
+    Then status 200
+    And match response.chunksIndexed == '#number'
+    And match response.chunksIndexed > 0
+
+    # ---- Step 8: Delete indexed chunks ----
+    Given path '/api/ingestion', docId
+    When method DELETE
+    Then status 204
+
+    # ---- Step 9: Cleanup — delete analysis and document ----
+    Given path '/api/analyses', jobId
+    When method DELETE
+    Then status 204
+
+    Given path '/api/documents', docId
+    When method DELETE
+    Then status 204
+
+  Scenario: Ingestion fails gracefully when job is not COMPLETED
+
+    # Upload a document
+    Given path '/api/documents/upload'
+    And multipart file file = { read: 'classpath:common/data/generated/medium.pdf', filename: 'ingest-fail-test.pdf', contentType: 'application/pdf' }
+    When method POST
+    Then status 200
+    * def docId = response.id
+
+    # Create analysis but do NOT wait for completion
+    Given path '/api/analyses'
+    And request { documentId: '#(docId)' }
+    When method POST
+    Then status 200
+    * def jobId = response.id
+
+    # Immediately attempt ingestion — should fail with 422 (job not COMPLETED or no chunks)
+    Given path '/api/ingestion', jobId
+    When method POST
+    * def sc = responseStatus
+    Then assert sc == 422 || sc == 503
+
+    # Cleanup
+    Given path '/api/documents', docId
+    When method DELETE
+    * def ignored = responseStatus
+
+  Scenario: Ingestion status — available when OpenSearch is configured
+
+    Given path '/api/ingestion/status'
+    When method GET
+    Then status 200
+    And match response == { available: '#boolean', reason: '#string' }