DAOS-15498 rebuild: reprobe in migrate_obj_iter_cb (#14458) (#14494)

migrate_system_enter possibly will yield, so need to do a re-probe before delete. Signed-off-by: Xuezhao Liu <[email protected]>
daos-stack · Jun 3, 2024 · 533ce74 · 533ce74
1 parent 2b519e9
commit 533ce74
Showing 1 changed file with 16 additions and 2 deletions.
diff --git a/src/object/srv_obj_migrate.c b/src/object/srv_obj_migrate.c
@@ -1869,7 +1869,7 @@ enum {
 
 /* Check if there are enough resource for the migration to proceed. */
 static int
-migrate_system_enter(struct migrate_pool_tls *tls, int tgt_idx)
+migrate_system_enter(struct migrate_pool_tls *tls, int tgt_idx, bool *yielded)
 {
 	uint32_t tgt_cnt = 0;
 	int	 rc = 0;
@@ -1883,6 +1883,7 @@ migrate_system_enter(struct migrate_pool_tls *tls, int tgt_idx)
 	while ((tls->mpt_inflight_max_ult / dss_tgt_nr) <= tgt_cnt) {
 		D_DEBUG(DB_REBUILD, "tgt%d:%u max %u\n",
 			tgt_idx, tgt_cnt, tls->mpt_inflight_max_ult / dss_tgt_nr);
+		*yielded = true;
 		ABT_mutex_lock(tls->mpt_inflight_mutex);
 		ABT_cond_wait(tls->mpt_inflight_cond, tls->mpt_inflight_mutex);
 		ABT_mutex_unlock(tls->mpt_inflight_mutex);
@@ -3378,6 +3379,8 @@ migrate_obj_iter_cb(daos_handle_t ih, d_iov_t *key_iov, d_iov_t *val_iov, void *
 	daos_epoch_t			punched_epoch = obj_val->punched_epoch;
 	unsigned int			tgt_idx = obj_val->tgt_idx;
 	unsigned int			shard = obj_val->shard;
+	d_iov_t				tmp_iov;
+	bool				yielded = false;
 	int				rc;
 
 	if (arg->pool_tls->mpt_fini)
@@ -3387,7 +3390,7 @@ migrate_obj_iter_cb(daos_handle_t ih, d_iov_t *key_iov, d_iov_t *val_iov, void *
 		" eph "DF_U64" start\n", DP_UUID(arg->cont_uuid), DP_UOID(*oid),
 		ih.cookie, epoch);
 
-	rc = migrate_system_enter(arg->pool_tls, tgt_idx);
+	rc = migrate_system_enter(arg->pool_tls, tgt_idx, &yielded);
 	if (rc != 0) {
 		DL_ERROR(rc, DF_UUID" enter migrate failed.", DP_UUID(arg->cont_uuid));
 		return rc;
@@ -3401,6 +3404,17 @@ migrate_obj_iter_cb(daos_handle_t ih, d_iov_t *key_iov, d_iov_t *val_iov, void *
 		return rc;
 	}
 
+	/* migrate_system_enter possibly yielded the ULT, let's re-probe before delete  */
+	if (yielded) {
+		d_iov_set(&tmp_iov, oid, sizeof(*oid));
+		rc = dbtree_iter_probe(ih, BTR_PROBE_EQ, DAOS_INTENT_MIGRATION, &tmp_iov, NULL);
+		if (rc) {
+			D_ASSERT(rc != -DER_NONEXIST);
+			D_ERROR("obj "DF_UOID" probe failed: "DF_RC"\n", DP_UOID(*oid), DP_RC(rc));
+			return rc;
+		}
+	}
+
 	rc = dbtree_iter_delete(ih, NULL);
 	if (rc) {
 		D_ERROR("dbtree_iter_delete failed: "DF_RC"\n", DP_RC(rc));