RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.
Go agent-harnessagent-native-databaseagentic-aiagentic-retrievalagentic-searchaiai-agentscontext-enginecontext-engineeringcontext-managementharness-engineeringknowledge-compilationragretrieval-augmented-generationretrieval-harnesssearch-harness
fix(nlp): re-enable the datasetnav integration test in the integration CI job (#20740)
Caobaining18eca005563e3
diff --git a/.github/workflows/sep-tests.yml b/.github/workflows/sep-tests.ymlindex 6f99f148cf..29961eb3c8 100644--- a/.github/workflows/sep-tests.yml+++ b/.github/workflows/sep-tests.yml@@ -725,6 +725,14 @@ jobs:run: |set -euo pipefailCOMPOSE_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 ${COMPOSE_PROJECT_NAME} down -v || truesudo docker ps -a --filter "label=com.docker.compose.project=${COMPOSE_PROJECT_NAME}" -q | xargs -r sudo docker rm -f@@ -1254,6 +1262,14 @@ jobs:run: |set -euo pipefailCOMPOSE_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 ${COMPOSE_PROJECT_NAME} down -v || truesudo docker ps -a --filter "label=com.docker.compose.project=${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 $PKGS- name: Tear down@@ -1842,6 +1852,14 @@ jobs:run: |set -euo pipefailCOMPOSE_FILE=docker/docker-compose.yml+ # This teardown runs with `if: always()`. The prepare step exports+ # COMPOSE_PROJECT_NAME, but if an earlier step failed and skipped the+ # prepare step there is no stack to remove — and `set -u` would turn+ # that clean no-op into a hard "unbound variable" error.+ if [[ -z "${COMPOSE_PROJECT_NAME:-}" ]]; then+ echo "COMPOSE_PROJECT_NAME is unset; nothing to tear down"+ exit 0+ fisudo docker compose -f "${COMPOSE_FILE}" -p ${COMPOSE_PROJECT_NAME} down -v || truesudo docker ps -a --filter "label=com.docker.compose.project=${COMPOSE_PROJECT_NAME}" -q | xargs -r sudo docker rm -f# Release the per-port docker-network guards reserved by the prepare step.diff --git a/internal/service/nlp/datasetnav_integration_test.go b/internal/service/nlp/datasetnav_integration_test.goindex a814334ca5..1f8567ac49 100644--- a/internal/service/nlp/datasetnav_integration_test.go+++ b/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]interfacet.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 {