DAOS-14010 rebuild: reint from stable epoch

Signed-off-by: Xuezhao Liu <[email protected]>

src/include/daos_srv/vos.h

@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2015-2024 Intel Corporation.
+ * (C) Copyright 2015-2025 Intel Corporation.
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -1712,4 +1712,16 @@ vos_unpin_objects(daos_handle_t coh, struct vos_pin_handle *hdl);
 int
 vos_pin_objects(daos_handle_t coh, daos_unit_oid_t oids[], int count, struct vos_pin_handle **hdl);
 
+/**
+ * Check if the oid exist in current vos.
+ *
+ * \param[in]	coh	container open handle.
+ * \param[in]	oid	oid to be checked.
+ *
+ * \return	true	exist.
+ *		false	does not exist.
+ */
+bool
+vos_oi_exist(daos_handle_t coh, daos_unit_oid_t oid);
+
 #endif /* __VOS_API_H */

src/object/srv_internal.h

@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2016-2024 Intel Corporation.
+ * (C) Copyright 2016-2025 Intel Corporation.
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -101,7 +101,9 @@ struct migrate_pool_tls {
 	/* migrate leader ULT */
 	unsigned int		mpt_ult_running:1,
 				mpt_init_tls:1,
-				mpt_fini:1;
+				mpt_fini:1,
+				mpt_reintegrating_check:1,
+				mpt_reintegrating:1;
 
 	/* migration init error */
 	int			mpt_init_err;

src/object/srv_obj_migrate.c

@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2019-2024 Intel Corporation.
+ * (C) Copyright 2019-2025 Intel Corporation.
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -3241,6 +3241,7 @@ migrate_obj_ult(void *data)
 	struct iter_obj_arg	*arg = data;
 	struct migrate_pool_tls	*tls = NULL;
 	daos_epoch_range_t	epr;
+	daos_epoch_t		stable_epoch = 0;
 	int			i;
 	int			rc = 0;
 
@@ -3253,7 +3254,7 @@ migrate_obj_ult(void *data)
 	/* Only reintegrating targets/pool needs to discard the object,
 	 * if sp_need_discard is 0, either the target does not need to
 	 * discard, or discard has been done. spc_discard_done means
-	 * discarding has been done in the current VOS target.
+	 * discarding has been done in the current VOS target.  lxzlxz
 	 */
 	if (tls->mpt_pool->spc_pool->sp_need_discard) {
 		while(!tls->mpt_pool->spc_discard_done) {
@@ -3271,8 +3272,70 @@ migrate_obj_ult(void *data)
 		}
 	}
 
+	if (!tls->mpt_reintegrating_check) {
+		struct pool_target *target;
+		struct ds_pool *pool = tls->mpt_pool->spc_pool;
+
+		ABT_rwlock_rdlock(pool->sp_lock);
+		rc = pool_map_find_target_by_rank_idx(pool->sp_map, dss_self_rank(),
+						      dss_get_module_info()->dmi_tgt_id,
+						      &target);
+		ABT_rwlock_unlock(pool->sp_lock);
+		D_ASSERT(rc == 1);
+
+		if (target->ta_comp.co_status == PO_COMP_ST_UP &&
+		    target->ta_comp.co_in_ver <= tls->mpt_version)
+			tls->mpt_reintegrating = 1;
+		D_DEBUG(DB_REBUILD, "status %u in version %u, mpt_reintegrating %d\n",
+			target->ta_comp.co_status, target->ta_comp.co_in_ver,
+			tls->mpt_reintegrating);
+
+		tls->mpt_reintegrating_check = 1;
+	}
+
+	if (tls->mpt_reintegrating) {
+		struct ds_cont_child *cont_child = NULL;
+
+		/* check again to see if the container is being destroyed. */
+		migrate_get_cont_child(tls, arg->cont_uuid, &cont_child, false);
+		if (cont_child != NULL && !cont_child->sc_stopping) {
+			if (vos_oi_exist(cont_child->sc_hdl, arg->oid)) {
+				rc = vos_dtx_cmt_reindex(cont_child->sc_hdl);
+				if (rc < 0) {
+					ds_cont_child_put(cont_child);
+					D_GOTO(out, rc);
+				}
+				stable_epoch = vos_cont_get_global_stable_epoch(cont_child->sc_hdl);
+			} else {
+				/* If the object does not exist on the local target at all,
+				 * it is either created after stable epoch or migrated to this
+				 * target due to co-location. let's set stable epoch to 0,
+				 * migrate the whole object.
+				 */
+				stable_epoch = 0;
+			}
+		}
+
+		if (cont_child)
+			ds_cont_child_put(cont_child);
+	}
+
 	for (i = 0; i < arg->snap_cnt; i++) {
-		epr.epr_lo = i > 0 ? arg->snaps[i - 1] + 1 : 0;
+		daos_epoch_t lower_epoch = 0;
+
+		if (arg->snaps[i] < stable_epoch) {
+			D_DEBUG(DB_REBUILD, DF_CONT " obj " DF_UOID " skip snap "DF_X64
+				" < stable " DF_X64 "\n", DP_CONT(arg->pool_uuid, arg->cont_uuid),
+				DP_UOID(arg->oid), arg->snaps[i], stable_epoch);
+			continue;
+		} else {
+			if (i == 0)
+				lower_epoch = stable_epoch;
+			else
+				lower_epoch = max(stable_epoch, arg->snaps[i - 1]);
+		}
+
+		epr.epr_lo = lower_epoch;
 		epr.epr_hi = arg->snaps[i];
 		D_DEBUG(DB_REBUILD, DF_RB ": rebuild_snap %d " DF_X64 "-" DF_X64 "\n",
 			DP_RB_MPT(tls), i, epr.epr_lo, epr.epr_hi);
@@ -3281,13 +3344,14 @@ migrate_obj_ult(void *data)
 			D_GOTO(free, rc);
 	}
 
-	if (arg->snap_cnt > 0 && arg->punched_epoch != 0) {
+	if (arg->snap_cnt > 0 && arg->punched_epoch != 0 && arg->punched_epoch > stable_epoch) {
 		rc = migrate_obj_punch(arg);
 		if (rc)
 			D_GOTO(free, rc);
 	}
 
 	epr.epr_lo = arg->snaps ? arg->snaps[arg->snap_cnt - 1] + 1 : 0;
+	epr.epr_lo = max(epr.epr_lo, stable_epoch);
 	D_ASSERT(tls->mpt_max_eph != 0);
 	epr.epr_hi = tls->mpt_max_eph;
 	if (arg->epoch > 0) {

src/vos/vos_container.c

@@ -925,7 +925,7 @@ vos_cont_get_local_stable_epoch(daos_handle_t coh)
  * epoch when reintegrate into the system.
  */
 daos_epoch_t
-vos_cont_get_globla_stable_epoch(daos_handle_t coh)
+vos_cont_get_global_stable_epoch(daos_handle_t coh)
 {
 	struct vos_container	*cont;
 	struct vos_cont_ext_df	*cont_ext;

src/vos/vos_obj_index.c

@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2016-2024 Intel Corporation.
+ * (C) Copyright 2016-2025 Intel Corporation.
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -227,6 +227,22 @@ static btr_ops_t oi_btr_ops = {
 	.to_node_alloc		= oi_node_alloc,
 };
 
+bool
+vos_oi_exist(daos_handle_t coh, daos_unit_oid_t oid)
+{
+	struct vos_container	*cont = vos_hdl2cont(coh);
+	d_iov_t			 key_iov;
+	d_iov_t			 val_iov;
+	int			 rc;
+
+	d_iov_set(&key_iov, &oid, sizeof(oid));
+	d_iov_set(&val_iov, NULL, 0);
+
+	rc = dbtree_fetch(cont->vc_btr_hdl, BTR_PROBE_EQ,
+			  DAOS_INTENT_DEFAULT, &key_iov, NULL, &val_iov);
+	return rc == 0;
+}
+
 /**
  * Locate a durable object in OI table.
  */