DAOS-18694 tests: test for container recovery when inc-reint

src/container/cli.c

@@ -3590,5 +3590,8 @@ cont_mark_slave(struct d_hlink *link, void *arg)
 int
 dc_cont_mark_all_slave(void)
 {
+	if (DAOS_FAIL_CHECK(DAOS_CONT_DESTROY_AFTER_FORK))
+		return 0;
+
 	return daos_hhash_traverse(DAOS_HTYPE_CO, cont_mark_slave, NULL);
 }

src/include/daos/common.h

@@ -918,6 +918,7 @@ enum {
 #define DAOS_POOL_RFCHECK_FAIL		(DAOS_FAIL_UNIT_TEST_GROUP_LOC | 0xa1)
 #define DAOS_POOL_REINT_SLOW            (DAOS_FAIL_UNIT_TEST_GROUP_LOC | 0xa2)
 #define DAOS_CONT_DESTROY_FAIL_POST        (DAOS_FAIL_UNIT_TEST_GROUP_LOC | 0xa3)
+#define DAOS_CONT_DESTROY_AFTER_FORK       (DAOS_FAIL_UNIT_TEST_GROUP_LOC | 0xa4)
 
 #define DAOS_CHK_CONT_ORPHAN		(DAOS_FAIL_UNIT_TEST_GROUP_LOC | 0xb0)
 #define DAOS_CHK_CONT_BAD_LABEL		(DAOS_FAIL_UNIT_TEST_GROUP_LOC | 0xb1)

src/tests/ftest/rebuild/inc_reint.py

@@ -0,0 +1,30 @@
+"""
+  (C) Copyright 2026 Hewlett Packard Enterprise Development LP
+
+  SPDX-License-Identifier: BSD-2-Clause-Patent
+"""
+from daos_core_base import DaosCoreBase
+
+
+class IncReintContRecovTest(DaosCoreBase):
+    # pylint: disable=too-many-ancestors,too-many-public-methods
+    """Runs daos incremental reintegration core tests.
+
+    :avocado: recursive
+    """
+
+    def test_daos_inc_reint_cont_recov(self):
+        """Jira ID: DAOS-17857
+
+        Test Description:
+            Run daos_test -Y
+
+        Use cases:
+            Core tests for daos container recovery when incremental reintegration
+
+        :avocado: tags=all,daily_regression
+        :avocado: tags=hw,medium,provider
+        :avocado: tags=inc_reint,cont_recov
+        :avocado: tags=IncReintContRecovTest,test_daos_inc_reint_cont_recov
+        """
+        self.run_subtest()

src/tests/ftest/rebuild/inc_reint.yaml

@@ -0,0 +1,63 @@
+hosts:
+  test_servers: 4
+
+timeout: 1H
+
+server_config:
+  name: daos_server
+  engines_per_host: 2
+  engines:
+    0:
+      targets: 4
+      pinned_numa_node: 0
+      nr_xs_helpers: 0
+      log_file: daos_server0.log
+      log_mask: DEBUG,MEM=ERR
+      env_vars:
+        - DD_MASK=mgmt,md,any
+        - D_LOG_FILE_APPEND_PID=1
+        - D_LOG_FILE_APPEND_RANK=1
+        - D_LOG_FLUSH=DEBUG
+        - FI_LOG_LEVEL=warn
+        - D_LOG_STDERR_IN_LOG=1
+      storage: auto
+
+    1:
+      targets: 4
+      pinned_numa_node: 1
+      nr_xs_helpers: 0
+      log_file: daos_server1.log
+      log_mask: DEBUG,MEM=ERR
+      env_vars:
+        - DD_MASK=mgmt,md,any
+        - D_LOG_FILE_APPEND_PID=1
+        - D_LOG_FILE_APPEND_RANK=1
+        - D_LOG_FLUSH=DEBUG
+        - FI_LOG_LEVEL=warn
+        - D_LOG_STDERR_IN_LOG=1
+      storage: auto
+
+  transport_config:
+    allow_insecure: true
+
+pool:
+  scm_size: 6G
+  nvme_size: 80G
+
+agent_config:
+  transport_config:
+    allow_insecure: true
+
+dmg:
+  transport_config:
+    allow_insecure: true
+
+daos_tests:
+  num_clients:
+    test_daos_inc_reint_cont_recov: 1
+  test_name:
+    test_daos_inc_reint_cont_recov: DAOS_Inc_Reint_Cont_Recov
+  daos_test:
+    test_daos_inc_reint_cont_recov: Y
+  args:
+    test_daos_inc_reint_cont_recov: -s3

src/tests/suite/daos_inc_reint.c

@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
+ * (C) Copyright 2025-2026 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -253,19 +253,32 @@ ir_race(test_arg_t *arg, bool create)
 		/* Wait for ir_rank_reint can be run firstly on parent. */
 		sleep(3);
 
+		if (!create)
+			daos_fail_loc_set(DAOS_CONT_DESTROY_AFTER_FORK | DAOS_FAIL_ALWAYS);
+
+		rc = daos_reinit();
+		assert_rc_equal(rc, 0);
+
 		if (create)
 			ir_cont_create(arg, &conts[3]);
 		else
 			ir_cont_destroy(arg, &conts[1]);
 
 		/* Do NOT exit immediately, otherwise, the pipeline for parent may be broken. */
 		sleep(15);
+		print_message("Child process complete and exit\n");
+
+		if (!create)
+			daos_fail_loc_reset();
+
+		daos_fini();
 		exit(0);
 	} else {
 		/* Incremental reintegration will internally repeat to handle the race. */
 		rc = ir_rank_reint(arg, 1, false);
 		assert_rc_equal(rc, 0);
 		waitpid(pid, &rc, 0);
+		print_message("Parent process wait child %d, got %d\n", pid, rc);
 	}
 
 	rc = daos_debug_set_params(arg->group, -1, DMG_KEY_FAIL_LOC, 0, 0, NULL);