DAOS-15391 pool: destroy container during reintegration discard (#13958…

…) (#14542)

Destroy the container in reintegration discard phase,
otherwise the orphan container might be left after reintegration
and causing trouble in reclaim phase.

Signed-off-by: Di Wang <di.wang@intel.com>
Co-authored-by: wangdi <di.wang@intel.com>

src/container/srv_target.c

@@ -1165,7 +1165,7 @@ cont_destroy_wait(struct ds_pool_child *child, uuid_t co_uuid)
  * Called via dss_collective() to destroy the ds_cont object as well as the vos
  * container.
  */
-static int
+int
 cont_child_destroy_one(void *vin)
 {
 	struct dsm_tls		       *tls = dsm_tls_get();
@@ -1260,6 +1260,19 @@ cont_child_destroy_one(void *vin)
 	return rc;
 }
 
+int
+ds_cont_child_destroy(uuid_t pool_uuid, uuid_t cont_uuid)
+{
+	struct cont_tgt_destroy_in  destroy_in;
+	int rc;
+
+	uuid_copy(destroy_in.tdi_pool_uuid, pool_uuid);
+	uuid_copy(destroy_in.tdi_uuid, cont_uuid);
+	rc = cont_child_destroy_one(&destroy_in);
+
+	return rc;
+}
+
 int
 ds_cont_tgt_destroy(uuid_t pool_uuid, uuid_t cont_uuid)
 {

src/include/daos_srv/container.h

@@ -191,6 +191,9 @@ void ds_cont_child_stop_all(struct ds_pool_child *pool_child);
 
 int ds_cont_child_lookup(uuid_t pool_uuid, uuid_t cont_uuid,
 			 struct ds_cont_child **ds_cont);
+int
+ds_cont_child_destroy(uuid_t pool_uuid, uuid_t cont_uuid);
+
 void
 ds_cont_child_reset_ec_agg_eph_all(struct ds_pool_child *pool_child);
 /** initialize a csummer based on container properties. Will retrieve the

src/pool/srv_target.c

@@ -2270,6 +2270,8 @@ cont_discard_cb(daos_handle_t ih, vos_iter_entry_t *entry,
 
 put:
 	ds_cont_child_put(cont);
+	if (rc == 0)
+		rc = ds_cont_child_destroy(arg->tgt_discard->pool_uuid, entry->ie_couuid);
 	return rc;
 }
 

src/tests/suite/daos_rebuild.c

@@ -1534,6 +1534,34 @@ rebuild_delay_and_extend(void **state)
 	rebuild_io_verify(arg, update_oids, OBJ_NR);
 }
 
+static void
+rebuild_cont_destroy_and_reintegrate(void **state)
+{
+	test_arg_t	*arg = *state;
+	daos_obj_id_t	oids[OBJ_NR];
+	int		rc;
+	int		i;
+
+	if (!test_runable(arg, 6))
+		return;
+
+	for (i = 0; i < OBJ_NR; i++) {
+		oids[i] = daos_test_oid_gen(arg->coh, DAOS_OC_R3S_SPEC_RANK, 0, 0, arg->myrank);
+		oids[i] = dts_oid_set_rank(oids[i], 5);
+	}
+
+	rebuild_io(arg, oids, OBJ_NR);
+	rebuild_single_pool_rank(arg, 5, true);
+
+	rc = daos_cont_close(arg->coh, NULL);
+	assert_rc_equal(rc, 0);
+	arg->coh = DAOS_HDL_INVAL;
+
+	daos_cont_destroy(arg->pool.poh, arg->co_str, 0 /* force */, NULL /* ev */);
+	uuid_clear(arg->co_uuid);
+	reintegrate_single_pool_rank(arg, 5, true);
+}
+
 /** create a new pool/container for each test */
 static const struct CMUnitTest rebuild_tests[] = {
 	{"REBUILD0: drop rebuild scan reply",
@@ -1623,6 +1651,9 @@ static const struct CMUnitTest rebuild_tests[] = {
 	  rebuild_delay_and_reintegrate, rebuild_sub_setup, rebuild_sub_teardown},
 	{"REBUILD34: delay rebuild and extend",
 	  rebuild_delay_and_extend, rebuild_sub_6nodes_rf1_setup, rebuild_sub_teardown},
+	{"REBUILD35: destroy container then reintegrate",
+	  rebuild_cont_destroy_and_reintegrate, rebuild_sub_6nodes_rf1_setup,
+	  rebuild_sub_teardown},
 };
 
 /* TODO: Enable aggregation once stable view rebuild is done. */