private void removeTask(String topologyId, int taskId, StormClusterState clusterState) throws Exception { Set<Integer> taskIds = new HashSet<Integer>(taskId); clusterState.remove_task(topologyid, taskIds); Map<Integer, TkHbCacheTime> TkHbs = data.getTaskHeartbeatsCache(topologyid, false); if (TkHbs != null) { TkHbs.remove(taskId); } }
private void setTaskInfo(StormTopology oldTopology, StormTopology newTopology) throws Exception { StormClusterState clusterState = data.getStormClusterState(); // Retrieve the max task ID TreeSet<Integer> taskIds = new TreeSet<Integer>(clusterState.task_ids(topologyid)); int cnt = taskIds.descendingIterator().next(); cnt = setBoltInfo(oldTopology, newTopology, cnt, clusterState); cnt = setSpoutInfo(oldTopology, newTopology, cnt, clusterState); }
@Override public <T> Object execute(T... args) { boolean isSetTaskInfo = false; try { Boolean reassign = (Boolean) args[1]; Map<Object, Object> conf = (Map<Object, Object>) args[2]; // args[0]: // delay, // args[1]: // reassign_flag, // args[2]: // conf if (conf != null) { boolean isConfUpdate = false; Map stormConf = data.getConf(); // Update topology code Map topoConf = StormConfig.read_nimbus_topology_conf(stormConf, topologyid); StormTopology rawOldTopology = StormConfig.read_nimbus_topology_code(stormConf, topologyid); StormTopology rawNewTopology = NimbusUtils.normalizeTopology(conf, rawOldTopology, true); StormTopology sysOldTopology = rawOldTopology.deepCopy(); StormTopology sysNewTopology = rawNewTopology.deepCopy(); if (conf.get(Config.TOPOLOGY_ACKER_EXECUTORS) != null) { Common.add_acker(topoConf, sysOldTopology); Common.add_acker(conf, sysNewTopology); int ackerNum = JStormUtils.parseInt(conf.get(Config.TOPOLOGY_ACKER_EXECUTORS)); int oldAckerNum = JStormUtils.parseInt(topoConf.get(Config.TOPOLOGY_ACKER_EXECUTORS)); LOG.info("Update acker from oldAckerNum=" + oldAckerNum + " to ackerNum=" + ackerNum); topoConf.put(Config.TOPOLOGY_ACKER_EXECUTORS, ackerNum); isConfUpdate = true; } // If scale-out, setup task info for new added tasks setTaskInfo(sysOldTopology, sysNewTopology); isSetTaskInfo = true; // If everything is OK, write topology code into disk StormConfig.write_nimbus_topology_code( stormConf, topologyid, Utils.serialize(rawNewTopology)); // Update topology conf if worker num has been updated Set<Object> keys = conf.keySet(); Integer workerNum = JStormUtils.parseInt(conf.get(Config.TOPOLOGY_WORKERS)); if (workerNum != null) { Integer oldWorkerNum = JStormUtils.parseInt(topoConf.get(Config.TOPOLOGY_WORKERS)); topoConf.put(Config.TOPOLOGY_WORKERS, workerNum); isConfUpdate = true; LOG.info("Update worker num from " + oldWorkerNum + " to " + workerNum); } if (keys.contains(Config.ISOLATION_SCHEDULER_MACHINES)) { topoConf.put( Config.ISOLATION_SCHEDULER_MACHINES, conf.get(Config.ISOLATION_SCHEDULER_MACHINES)); } if (isConfUpdate) { StormConfig.write_nimbus_topology_conf(stormConf, topologyid, topoConf); } } TopologyAssignEvent event = new TopologyAssignEvent(); event.setTopologyId(topologyid); event.setScratch(true); event.setOldStatus(oldStatus); event.setReassign(reassign); if (conf != null) event.setScaleTopology(true); TopologyAssign.push(event); event.waitFinish(); } catch (Exception e) { LOG.error("do-rebalance error!", e); // Rollback the changes on ZK if (isSetTaskInfo) { try { StormClusterState clusterState = data.getStormClusterState(); clusterState.remove_task(topologyid, newTasks); } catch (Exception e1) { LOG.error("Failed to rollback the changes on ZK for task-" + newTasks, e); } } } DelayStatusTransitionCallback delayCallback = new DelayStatusTransitionCallback( data, topologyid, oldStatus, StatusType.rebalancing, StatusType.done_rebalance); return delayCallback.execute(); }