DAOS-13896 pool: reintegration discard after rebuild (#12577)

Pool reintegration discard should be done after rebuild
to make sure all DOWNOUT targets being discarded before
reintegration.

Required-githooks: true

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

src/pool/srv_pool.c

@@ -5787,7 +5787,6 @@ pool_svc_update_map_internal(struct pool_svc *svc, unsigned int opc,
 	struct pool_map	       *map;
 	uint32_t		map_version_before;
 	uint32_t		map_version;
-	uint32_t		rebuild_ver = 0;
 	struct pool_buf	       *map_buf = NULL;
 	bool			updated = false;
 	int			rc;
@@ -5796,19 +5795,6 @@ pool_svc_update_map_internal(struct pool_svc *svc, unsigned int opc,
 		DP_UUID(svc->ps_uuid), opc, exclude_rank, tgts->pti_number,
 		tgt_addrs == NULL ? 0 : tgt_addrs->pta_number);
 
-	/* Check if there are ongoing rebuild jobs for extend/reint/drain, to avoid
-	 * updating the pool map during rebuild, which might screw the object layout.
-	 */
-	if (opc == POOL_EXTEND || opc == POOL_REINT || opc == POOL_DRAIN) {
-		ds_rebuild_running_query(svc->ps_uuid, &rebuild_ver, NULL, NULL);
-		if (rebuild_ver != 0) {
-			D_ERROR(DF_UUID": other rebuild job rebuild ver %u is ongoing,"
-				" so current opc %d can not be done: %d\n",
-				DP_UUID(svc->ps_uuid), rebuild_ver, opc, DER_BUSY);
-			D_GOTO(out, rc = -DER_BUSY);
-		}
-	}
-
 	rc = rdb_tx_begin(svc->ps_rsvc.s_db, svc->ps_rsvc.s_term, &tx);
 	if (rc != 0)
 		goto out;
@@ -6146,6 +6132,7 @@ ds_pool_extend_handler(crt_rpc_t *rpc)
 	d_rank_list_t		rank_list;
 	uint32_t		ndomains;
 	uint32_t		*domains;
+	uint32_t		rebuild_ver = 0;
 	int			rc;
 
 	D_DEBUG(DB_MD, DF_UUID": processing rpc %p\n", DP_UUID(in->pei_op.pi_uuid), rpc);
@@ -6160,12 +6147,24 @@ ds_pool_extend_handler(crt_rpc_t *rpc)
 	if (rc != 0)
 		goto out;
 
+	/**
+	 * Check if there are ongoing rebuild jobs for extend/reint/drain, to avoid
+	 * updating the pool map during rebuild, which might screw the object layout.
+	 */
+	ds_rebuild_running_query(svc->ps_uuid, &rebuild_ver, NULL, NULL);
+	if (rebuild_ver != 0) {
+		D_ERROR(DF_UUID": other rebuild job rebuild ver %u is ongoing,"
+			"extending can not be done: %d\n",
+			DP_UUID(svc->ps_uuid), rebuild_ver, -DER_BUSY);
+		D_GOTO(put, rc = -DER_BUSY);
+	}
+
 	rc = pool_svc_update_map(svc, opc_get(rpc->cr_opc),
 				 false /* exclude_rank */,
 				 &rank_list, domains, ndomains,
 				 NULL, NULL, &out->peo_op.po_map_version,
 				 &out->peo_op.po_hint);
-
+put:
 	pool_svc_put_leader(svc);
 out:
 	out->peo_op.po_rc = rc;
@@ -6239,6 +6238,7 @@ ds_pool_update_handler(crt_rpc_t *rpc)
 	struct pool_tgt_update_in      *in = crt_req_get(rpc);
 	struct pool_tgt_update_out     *out = crt_reply_get(rpc);
 	struct pool_svc		       *svc;
+	uint32_t			rebuild_ver = 0;
 	struct pool_target_addr_list	list = { 0 };
 	struct pool_target_addr_list	inval_list_out = { 0 };
 	int				rc;
@@ -6255,6 +6255,21 @@ ds_pool_update_handler(crt_rpc_t *rpc)
 	if (rc != 0)
 		goto out;
 
+	if (opc_get(rpc->cr_opc) == POOL_REINT || opc_get(rpc->cr_opc) == POOL_DRAIN) {
+		/**
+		 * Check if there are ongoing rebuild jobs for extend/reint/drain, to avoid
+		 * updating the pool map during rebuild, which might screw the object layout.
+		 */
+		ds_rebuild_running_query(svc->ps_uuid, &rebuild_ver, NULL, NULL);
+		if (rebuild_ver != 0) {
+			D_ERROR(DF_UUID": other rebuild job rebuild ver %u is ongoing,"
+				" so current opc %d can not be done: %d\n",
+				DP_UUID(svc->ps_uuid), rebuild_ver, opc_get(rpc->cr_opc),
+				-DER_BUSY);
+			D_GOTO(out_svc, rc = -DER_BUSY);
+		}
+	}
+
 	list.pta_number = in->pti_addr_list.ca_count;
 	list.pta_addrs = in->pti_addr_list.ca_arrays;