DAOS-18296 rdb: Call vos_pool_create/open in ULTs (#17302)

Call vos_pool_create and vos_pool_open in new deep-stack ULTs on the
xstream to avoid pmemobj_create and pmemobj_open from potentially
overflowing caller stacks.

Signed-off-by: Li Wei <liwei@hpe.com>

Author: liw

src/engine/ult.c

@@ -1,6 +1,6 @@
 /**
  * (C) Copyright 2016-2024 Intel Corporation.
- * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
+ * (C) Copyright 2025-2026 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -9,6 +9,7 @@
 
 #include <abt.h>
 #include <daos/common.h>
+#include <daos_srv/vos.h>
 #include <daos_errno.h>
 #include "srv_internal.h"
 
@@ -931,3 +932,80 @@ dss_chore_queue_fini(struct dss_xstream *dx)
 	ABT_cond_free(&queue->chq_cond);
 	ABT_mutex_free(&queue->chq_mutex);
 }
+
+struct dss_vos_pool_create_args {
+	const char    *spc_path;
+	unsigned char *spc_uuid;
+	daos_size_t    spc_scm_size;
+	daos_size_t    spc_data_sz;
+	daos_size_t    spc_meta_sz;
+	unsigned int   spc_flags;
+	uint32_t       spc_version;
+	daos_handle_t *spc_pool;
+};
+
+static int
+dss_vos_pool_create_ult(void *varg)
+{
+	struct dss_vos_pool_create_args *arg = varg;
+
+	return vos_pool_create(arg->spc_path, arg->spc_uuid, arg->spc_scm_size, arg->spc_data_sz,
+			       arg->spc_meta_sz, arg->spc_flags, arg->spc_version, arg->spc_pool);
+}
+
+/**
+ * Call vos_pool_create in a new deep-stack ULT on the same xstream. This is to
+ * avoid pmemobj_create or SPDK from overflowing the stack of the calling ULT.
+ */
+int
+dss_vos_pool_create(const char *path, unsigned char *uuid, daos_size_t scm_size,
+		    daos_size_t data_sz, daos_size_t meta_sz, unsigned int flags, uint32_t version,
+		    daos_handle_t *pool)
+{
+	struct dss_vos_pool_create_args args;
+
+	args.spc_path     = path;
+	args.spc_uuid     = uuid;
+	args.spc_scm_size = scm_size;
+	args.spc_data_sz  = data_sz;
+	args.spc_meta_sz  = meta_sz;
+	args.spc_flags    = flags;
+	args.spc_version  = version;
+	args.spc_pool     = pool;
+
+	return dss_ult_execute(dss_vos_pool_create_ult, &args, NULL /* user_cb */,
+			       NULL /* cb_args */, DSS_XS_SELF, 0 /* tgt_id */, DSS_DEEP_STACK_SZ);
+}
+
+struct dss_vos_pool_open_args {
+	const char    *spo_path;
+	unsigned char *spo_uuid;
+	unsigned int   spo_flags;
+	daos_handle_t *spo_pool;
+};
+
+static int
+dss_vos_pool_open_ult(void *varg)
+{
+	struct dss_vos_pool_open_args *arg = varg;
+
+	return vos_pool_open(arg->spo_path, arg->spo_uuid, arg->spo_flags, arg->spo_pool);
+}
+
+/**
+ * Call vos_pool_open in a new deep-stack ULT on the same xstream. This is to
+ * avoid pmemobj_open or SPDK from overflowing the stack of the calling ULT.
+ */
+int
+dss_vos_pool_open(const char *path, unsigned char *uuid, unsigned int flags, daos_handle_t *pool)
+{
+	struct dss_vos_pool_open_args args;
+
+	args.spo_path  = path;
+	args.spo_uuid  = uuid;
+	args.spo_flags = flags;
+	args.spo_pool  = pool;
+
+	return dss_ult_execute(dss_vos_pool_open_ult, &args, NULL /* user_cb */, NULL /* cb_args */,
+			       DSS_XS_SELF, 0 /* tgt_id */, DSS_DEEP_STACK_SZ);
+}

src/include/daos_srv/daos_engine.h

@@ -1,6 +1,6 @@
 /**
  * (C) Copyright 2016-2024 Intel Corporation.
- * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
+ * (C) Copyright 2025-2026 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -846,4 +846,11 @@ dss_select_module_version(int module_id, uint8_t *module_ver)
 		return dss_select_module_version(module_id, version);                              \
 	}
 
+int
+dss_vos_pool_create(const char *path, unsigned char *uuid, daos_size_t scm_size,
+		    daos_size_t data_sz, daos_size_t meta_sz, unsigned int flags, uint32_t version,
+		    daos_handle_t *pool);
+int
+dss_vos_pool_open(const char *path, unsigned char *uuid, unsigned int flags, daos_handle_t *pool);
+
 #endif /* __DSS_API_H__ */

src/rdb/rdb.c

@@ -1,6 +1,6 @@
 /**
  * (C) Copyright 2017-2023 Intel Corporation.
- * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
+ * (C) Copyright 2025-2026 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -57,13 +57,12 @@ rdb_create(const char *path, const uuid_t uuid, uint64_t caller_term,
 	 * basic system memory reservation and VOS_POF_EXCL for concurrent
 	 * access protection.
 	 */
-	rc = vos_pool_create(path, (unsigned char *)uuid, params->rcp_size, 0 /* data_sz */,
-			     0 /* meta_sz */,
-			     VOS_POF_SMALL | VOS_POF_EXCL | VOS_POF_RDB | VOS_POF_EXTERNAL_CHKPT,
-			     params->rcp_vos_df_version, &pool);
+	rc = dss_vos_pool_create(
+	    path, (unsigned char *)uuid, params->rcp_size, 0 /* data_sz */, 0 /* meta_sz */,
+	    VOS_POF_SMALL | VOS_POF_EXCL | VOS_POF_RDB | VOS_POF_EXTERNAL_CHKPT,
+	    params->rcp_vos_df_version, &pool);
 	if (rc != 0)
 		goto out;
-	ABT_thread_yield();
 
 	/* Create and open the metadata container. */
 	rc = vos_cont_create(pool, (unsigned char *)uuid);
@@ -427,9 +426,9 @@ rdb_open(const char *path, const uuid_t uuid, uint64_t caller_term, struct rdb_c
 	 * RDB pools specify VOS_POF_SMALL for basic system memory reservation
 	 * and VOS_POF_EXCL for concurrent access protection.
 	 */
-	rc = vos_pool_open(path, (unsigned char *)uuid,
-			   VOS_POF_SMALL | VOS_POF_EXCL | VOS_POF_RDB | VOS_POF_EXTERNAL_CHKPT,
-			   &pool);
+	rc = dss_vos_pool_open(path, (unsigned char *)uuid,
+			       VOS_POF_SMALL | VOS_POF_EXCL | VOS_POF_RDB | VOS_POF_EXTERNAL_CHKPT,
+			       &pool);
 	if (rc == -DER_ID_MISMATCH) {
 		ds_notify_ras_eventf(RAS_RDB_DF_INCOMPAT, RAS_TYPE_INFO, RAS_SEV_ERROR,
 				     NULL /* hwid */, NULL /* rank */, NULL /* inc */,
@@ -442,7 +441,6 @@ rdb_open(const char *path, const uuid_t uuid, uint64_t caller_term, struct rdb_c
 			path, DP_RC(rc));
 		goto err;
 	}
-	ABT_thread_yield();
 
 	rc = vos_cont_open(pool, (unsigned char *)uuid, &mc);
 	if (rc != 0) {