fix(nlp): re-enable the datasetnav integration test in the integration CI job (#20740)

.github/workflows/sep-tests.yml

@@ -725,6 +725,14 @@ jobs:
         run: |
           set -euo pipefail
           COMPOSE_FILE=docker/docker-compose.yml
+          # This teardown runs with `if: always()`. The prepare step exports
+          # COMPOSE_PROJECT_NAME, but if an earlier step failed and skipped the
+          # prepare step there is no stack to remove — and `set -u` would turn
+          # that clean no-op into a hard "unbound variable" error.
+          if [[ -z "${COMPOSE_PROJECT_NAME:-}" ]]; then
+            echo "COMPOSE_PROJECT_NAME is unset; nothing to tear down"
+            exit 0
+          fi
           # Sometimes `docker compose down` fail due to hang container, heavy load etc. Need to remove such containers to release resources(for example, listen ports).
           sudo docker compose -f "${COMPOSE_FILE}" -p ${COMPOSE_PROJECT_NAME} down -v || true
           sudo docker ps -a --filter "label=com.docker.compose.project=${COMPOSE_PROJECT_NAME}" -q | xargs -r sudo docker rm -f
@@ -1254,6 +1262,14 @@ jobs:
         run: |
           set -euo pipefail
           COMPOSE_FILE=docker/docker-compose.yml
+          # This teardown runs with `if: always()`. The prepare step exports
+          # COMPOSE_PROJECT_NAME, but if an earlier step failed and skipped the
+          # prepare step there is no stack to remove — and `set -u` would turn
+          # that clean no-op into a hard "unbound variable" error.
+          if [[ -z "${COMPOSE_PROJECT_NAME:-}" ]]; then
+            echo "COMPOSE_PROJECT_NAME is unset; nothing to tear down"
+            exit 0
+          fi
           # Sometimes `docker compose down` fail due to hang container, heavy load etc. Need to remove such containers to release resources(for example, listen ports).
           sudo docker compose -f "${COMPOSE_FILE}" -p ${COMPOSE_PROJECT_NAME} down -v || true
           sudo docker ps -a --filter "label=com.docker.compose.project=${COMPOSE_PROJECT_NAME}" -q | xargs -r sudo docker rm -f
@@ -1825,16 +1841,10 @@ jobs:
           # a fully-ready MinIO that this CI stack does not reliably provide
           # (the compose /live probe passes while the S3 API still 502s), so they
           # are exercised elsewhere, not here.
-          # internal/service/nlp is also excluded: its DatasetNav integration test
-          # compiles only against source fixes (Infinity 3013/3052, nav
-          # available_int, doc_ids_kwd) that are intentionally out of scope for this
-          # CI-enabling PR, so the package is left out of this job's package list
-          # rather than patched. Those fixes land in their own PR.
           PKGS=$(go list ./... \
             | grep -v '/test/api' \
             | grep -v '^ragflow/internal/deepdoc' \
-            | grep -v '/internal/storage$' \
-            | grep -v '^ragflow/internal/service/nlp$')
+            | grep -v '/internal/storage$')
           ./build.sh --test-integration-go $PKGS
 
       - name: Tear down
@@ -1842,6 +1852,14 @@ jobs:
         run: |
           set -euo pipefail
           COMPOSE_FILE=docker/docker-compose.yml
+          # This teardown runs with `if: always()`. The prepare step exports
+          # COMPOSE_PROJECT_NAME, but if an earlier step failed and skipped the
+          # prepare step there is no stack to remove — and `set -u` would turn
+          # that clean no-op into a hard "unbound variable" error.
+          if [[ -z "${COMPOSE_PROJECT_NAME:-}" ]]; then
+            echo "COMPOSE_PROJECT_NAME is unset; nothing to tear down"
+            exit 0
+          fi
           sudo docker compose -f "${COMPOSE_FILE}" -p ${COMPOSE_PROJECT_NAME} down -v || true
           sudo docker ps -a --filter "label=com.docker.compose.project=${COMPOSE_PROJECT_NAME}" -q | xargs -r sudo docker rm -f
           # Release the per-port docker-network guards reserved by the prepare step.

internal/service/nlp/datasetnav_integration_test.go

@@ -7,7 +7,6 @@ import (
 	"context"
 	"os"
 	"path/filepath"
-	"strings"
 	"testing"
 
 	"ragflow/internal/common"
@@ -46,9 +45,17 @@ func findNavRow(t *testing.T, tenantID, kbID, docID string) map[string]interface
 		t.Skip("no live document engine")
 	}
 	idx := "ragflow_" + tenantID
+	// The nav row's role marker is type_kwd (nav_doc/nav_cluster). Since the #20587
+	// field-name refactor compile_kwd carries the real compilation kind
+	// ("tree"/"page_index"), so it is no longer the selector that scopes nav rows.
 	req := &types.SearchRequest{
-		IndexNames:   []string{idx},
-		Filter:       map[string]interface{}{"doc_id": []string{docID}, "compile_kwd": []string{"dataset_nav"}},
+		IndexNames: []string{idx},
+		// KbIDs is what lets the engine resolve the per-KB table on Infinity
+		// (ragflow_<tenant>_<kb>); without it the read matches no table and the
+		// row is never found. ES keys on the tenant index alone, so it is inert
+		// there.
+		KbIDs:        []string{kbID},
+		Filter:       map[string]interface{}{"doc_id": []string{docID}, "type_kwd": []string{"nav_doc"}},
 		SelectFields: []string{"available_int", "compile_kwd", "type_kwd"},
 		Limit:        10,
 	}
@@ -71,14 +78,10 @@ func findNavRow(t *testing.T, tenantID, kbID, docID string) map[string]interface
 //
 // Run with: bash build.sh --test-integration ./internal/service/nlp/...
 //
-// Excluded from the integration CI job: this test exercises the DatasetNav flow
-// end-to-end against a live Infinity backend and surfaces several upstream source
-// bugs (Infinity 3013 on pure vector matches, 3052 on empty keyword filters,
-// doc_ids_kwd JSON encoding, NavService.Search available_int pinning,
-// findBestCluster empty-title descent). Those are genuine source fixes that belong
-// in their own PR — this PR only enables the integration-test CI job, so the
-// internal/service/nlp package is excluded from the job's package list until the
-// source fixes land. Re-enable once those are merged.
+// Runs in the integration CI job (infinity). The nav row's role marker is
+// type_kwd (nav_doc/nav_cluster); compile_kwd carries the real compilation kind
+// ("tree"/"page_index") since the #20587 field-name refactor, so the row is
+// located and asserted by type_kwd.
 func TestDatasetNav_AvailableIntZero_Isolation(t *testing.T) {
 	if err := common.InitLogger("info", common.FileOutput{}, ""); err != nil {
 		t.Fatalf("init logger: %v", err)
@@ -115,12 +118,12 @@ func TestDatasetNav_AvailableIntZero_Isolation(t *testing.T) {
 		t.Fatal("NavService.Search returned no hits; nav row is not reachable")
 	}
 
-	// The written nav row must carry compile_kwd=dataset_nav and available_int=0,
-	// so the default retriever (available_int=1 filter) will not surface it.
+	// The written nav row must be a nav_doc with available_int=0, so the default
+	// retriever (available_int=1 filter) will not surface it.
 	row := findNavRow(t, tenantID, kbID, docID)
-	// compile_kwd may come back list-wrapped by the engine, so use firstStrOrSlice.
-	if ck := firstStrOrSlice(row["compile_kwd"]); !strings.Contains(ck, "dataset_nav") {
-		t.Errorf("nav row compile_kwd = %q, want dataset_nav", ck)
+	// type_kwd may come back list-wrapped by the engine, so use firstStrOrSlice.
+	if tk := firstStrOrSlice(row["type_kwd"]); tk != "nav_doc" {
+		t.Errorf("nav row type_kwd = %q, want nav_doc", tk)
 	}
 	avail := intValue(row["available_int"])
 	if avail != 0 {