Add search stage tests

.github/workflows/pr-tests.yml

@@ -61,6 +61,17 @@ jobs:
             --json-report --json-report-file=${{ github.workspace }}/.test-results/${{ matrix.target.name }}-report.json \
             --junitxml=${{ github.workspace }}/.test-results/${{ matrix.target.name }}-results.xml
 
+      - name: Dump container logs
+        if: always()
+        run: |
+          # One file per service in the profile
+          mkdir -p "${{ github.workspace }}/.test-results/container-logs"
+          for svc in $(docker compose -f dev/compose.yaml --profile ${{ matrix.target.profile }} config --services); do
+            docker compose -f dev/compose.yaml --profile ${{ matrix.target.profile }} \
+              logs --no-color --timestamps "$svc" \
+              > "${{ github.workspace }}/.test-results/container-logs/${svc}.log" 2>&1 || true
+          done
+
       - name: Upload test results
         if: always()
         uses: actions/upload-artifact@v7

dev/compose.yaml

@@ -27,7 +27,7 @@
 #     query: <optional connection-string query string, without leading '?'>
 #
 # A service with no `x-test-target` is not a test target and is ignored by the
-# registry.
+# registry (e.g. the mongot sidecar, which is reached only through its mongod).
 #
 # Memory: each mongod caps its WiredTiger cache (--wiredTigerCacheSizeGB). By
 # default a mongod sizes its cache to ~50% of the host/VM RAM; with several
@@ -60,7 +60,26 @@ services:
   mongo-replset:
     image: mongo:8.2.4
     profiles: ["mongo-replset", "all"]
-    command: ["--replSet", "rs0", "--bind_ip_all", "--wiredTigerCacheSizeGB", "1.5"]
+    command:
+      - "--replSet"
+      - "rs0"
+      - "--bind_ip_all"
+      - "--wiredTigerCacheSizeGB"
+      - "1.5"
+      # Point at the mongot search sidecar so this replica set also serves the
+      # search surfaces. mongot is transparent to all other behavior, so the
+      # set behaves identically to a plain replica set apart from gaining
+      # search; it is one target, not two.
+      - "--setParameter"
+      - "mongotHost=mongot:27028"
+      - "--setParameter"
+      - "searchIndexManagementHostAndPort=mongot:27028"
+      - "--setParameter"
+      - "useGrpcForSearch=true"
+      - "--setParameter"
+      - "skipAuthenticationToMongot=true"
+      - "--setParameter"
+      - "skipAuthenticationToSearchIndexManagementServer=true"
     ports:
       - "27018:27017"
     healthcheck:
@@ -71,3 +90,31 @@ services:
     x-test-target:
       engine: mongodb
       query: directConnection=true
+
+  # mongot: the search sidecar for the mongo-replset target. Not a test target
+  # on its own; the suite reaches it only through mongo-replset. mongot is
+  # MongoDB Search Community Edition (SSPL, same license as the server). It
+  # replicates from the replica set as an authenticated sync source and reads
+  # its password from a file, so the entrypoint writes that file (a fixed
+  # local-dev secret, matched by the searchCoordinator user the harness creates
+  # on the replica set) with owner-only permissions before launching. It retries
+  # the connection until that user exists.
+  mongot:
+    image: mongodb/mongodb-community-search:latest
+    profiles: ["mongo-replset", "all"]
+    entrypoint:
+      - "sh"
+      - "-c"
+      - >
+        umask 077 &&
+        mkdir -p /mongot-secrets &&
+        printf '%s' "$$MONGOT_SYNC_PASSWORD" > /mongot-secrets/passwordFile &&
+        exec /mongot-community/mongot --config /mongot-config/mongot.yml
+    environment:
+      # Fixed local-dev secret shared with the searchCoordinator user the
+      # harness provisions on mongo-replset. Not a real credential.
+      MONGOT_SYNC_PASSWORD: "searchSyncPassword"
+      # Cap mongot's JVM heap. Unset, the JVM sizes its max heap to ~25% of host RAM.
+      JAVA_TOOL_OPTIONS: "-Xmx1g"
+    volumes:
+      - ./mongot.yml:/mongot-config/mongot.yml:ro

dev/mongot.yml

@@ -0,0 +1,30 @@
+# mongot configuration for the mongo-replset target (dev/compose.yaml service
+# "mongot"). mongot is MongoDB Search Community Edition (SSPL), the same license
+# as the server. It runs alongside the replica set's mongod and serves the
+# search and vector search surfaces.
+#
+# mongot replicates from the mongod replica set as a sync source. It requires an
+# authenticated connection (it has no unauthenticated mode), so it logs in as a
+# dedicated user holding the searchCoordinator role. That user and its password
+# file are provisioned by the target's startup (see dev/compose.yaml).
+syncSource:
+  replicaSet:
+    hostAndPort: "mongo-replset:27017"
+    username: "searchSyncUser"
+    passwordFile: "/mongot-secrets/passwordFile"
+    authSource: "admin"
+    tls: false
+storage:
+  dataPath: "/var/lib/mongot"
+server:
+  grpc:
+    # mongod reaches mongot here (see mongotHost / searchIndexManagementHostAndPort
+    # on the mongo-replset service). Bound on all interfaces so the mongod
+    # container can connect over the compose network.
+    address: "0.0.0.0:27028"
+    tls:
+      mode: "disabled"
+healthCheck:
+  address: "0.0.0.0:8080"
+logging:
+  verbosity: INFO

documentdb_tests/compatibility/tests/core/operator/stages/search/conftest.py

@@ -0,0 +1,31 @@
+"""Shared fixtures for $search stage tests.
+
+The dynamic-mapping ``indexed_collection`` corpus is queried read-only by most
+$search test files, so it is built once per package here rather than duplicated
+per file. Single-operator corpora (wildcard, equals, in, ...) stay inline in
+their own test file, since each is used by exactly one file."""
+
+from __future__ import annotations
+
+import pytest
+
+from documentdb_tests.compatibility.tests.core.operator.stages.search.utils.search_common import (
+    FIXTURE_DOCS,
+    create_dynamic_search_index,
+)
+from documentdb_tests.framework import fixtures
+
+
+@pytest.fixture(scope="package")
+def indexed_collection(engine_client, worker_id):
+    """A package-scoped collection populated with the fixture docs and a ready
+    dynamic search index, shared read-only across the matching tests so the
+    index is built and polled once rather than per test."""
+    db_name = fixtures.generate_database_name("stages_search_shared", worker_id)
+    fixtures.cleanup_database(engine_client, db_name)
+    db = engine_client[db_name]
+    coll = db["indexed"]
+    coll.insert_many(FIXTURE_DOCS)
+    create_dynamic_search_index(coll)
+    yield coll
+    fixtures.cleanup_database(engine_client, db_name)