|
| 1 | +package backtype.storm.metric; |
| 2 | + |
| 3 | +import backtype.storm.Config; |
| 4 | +import backtype.storm.metric.api.AssignableMetric; |
| 5 | +import backtype.storm.metric.api.IMetric; |
| 6 | +import backtype.storm.task.IBolt; |
| 7 | +import backtype.storm.task.OutputCollector; |
| 8 | +import backtype.storm.task.TopologyContext; |
| 9 | +import backtype.storm.tuple.Tuple; |
| 10 | +import clojure.lang.AFn; |
| 11 | +import clojure.lang.IFn; |
| 12 | +import clojure.lang.RT; |
| 13 | +import com.google.common.collect.ImmutableMap; |
| 14 | +import org.slf4j.Logger; |
| 15 | +import org.slf4j.LoggerFactory; |
| 16 | + |
| 17 | +import java.lang.management.*; |
| 18 | +import java.util.List; |
| 19 | +import java.util.Map; |
| 20 | + |
| 21 | + |
| 22 | +// There is one task inside one executor for each worker of the topology. |
| 23 | +// TaskID is always -1, therefore you can only send-unanchored tuples to co-located SystemBolt. |
| 24 | +// This bolt was conceived to export worker stats via metrics api. |
| 25 | +public class SystemBolt implements IBolt { |
| 26 | + private static Logger LOG = LoggerFactory.getLogger(SystemBolt.class); |
| 27 | + private static boolean _prepareWasCalled = false; |
| 28 | + |
| 29 | + private static class MemoryUsageMetric implements IMetric { |
| 30 | + IFn _getUsage; |
| 31 | + public MemoryUsageMetric(IFn getUsage) { |
| 32 | + _getUsage = getUsage; |
| 33 | + } |
| 34 | + @Override |
| 35 | + public Object getValueAndReset() { |
| 36 | + MemoryUsage memUsage = (MemoryUsage)_getUsage.invoke(); |
| 37 | + return ImmutableMap.builder() |
| 38 | + .put("maxBytes", memUsage.getMax()) |
| 39 | + .put("committedBytes", memUsage.getCommitted()) |
| 40 | + .put("initBytes", memUsage.getInit()) |
| 41 | + .put("usedBytes", memUsage.getUsed()) |
| 42 | + .put("virtualFreeBytes", memUsage.getMax() - memUsage.getUsed()) |
| 43 | + .put("unusedBytes", memUsage.getCommitted() - memUsage.getUsed()) |
| 44 | + .build(); |
| 45 | + } |
| 46 | + } |
| 47 | + |
| 48 | + // canonically the metrics data exported is time bucketed when doing counts. |
| 49 | + // convert the absolute values here into time buckets. |
| 50 | + private static class GarbageCollectorMetric implements IMetric { |
| 51 | + GarbageCollectorMXBean _gcBean; |
| 52 | + Long _collectionCount; |
| 53 | + Long _collectionTime; |
| 54 | + public GarbageCollectorMetric(GarbageCollectorMXBean gcBean) { |
| 55 | + _gcBean = gcBean; |
| 56 | + } |
| 57 | + @Override |
| 58 | + public Object getValueAndReset() { |
| 59 | + Long collectionCountP = _gcBean.getCollectionCount(); |
| 60 | + Long collectionTimeP = _gcBean.getCollectionCount(); |
| 61 | + |
| 62 | + Map ret = null; |
| 63 | + if(_collectionCount!=null && _collectionTime!=null) { |
| 64 | + ret = ImmutableMap.builder() |
| 65 | + .put("count", collectionCountP - _collectionCount) |
| 66 | + .put("timeMs", collectionTimeP - _collectionTime) |
| 67 | + .build(); |
| 68 | + } |
| 69 | + |
| 70 | + _collectionCount = collectionCountP; |
| 71 | + _collectionTime = collectionTimeP; |
| 72 | + return ret; |
| 73 | + } |
| 74 | + } |
| 75 | + |
| 76 | + @Override |
| 77 | + public void prepare(final Map stormConf, TopologyContext context, OutputCollector collector) { |
| 78 | + if(_prepareWasCalled && stormConf.get(Config.STORM_CLUSTER_MODE) != "local") { |
| 79 | + throw new RuntimeException("A single worker should have 1 SystemBolt instance."); |
| 80 | + } |
| 81 | + _prepareWasCalled = true; |
| 82 | + |
| 83 | + int bucketSize = RT.intCast(stormConf.get(Config.TOPOLOGY_BUILTIN_METRICS_BUCKET_SIZE_SECS)); |
| 84 | + |
| 85 | + final RuntimeMXBean jvmRT = ManagementFactory.getRuntimeMXBean(); |
| 86 | + |
| 87 | + context.registerMetric("uptimeSecs", new IMetric() { |
| 88 | + @Override |
| 89 | + public Object getValueAndReset() { |
| 90 | + return jvmRT.getUptime()/1000.0; |
| 91 | + } |
| 92 | + }, bucketSize); |
| 93 | + |
| 94 | + // You can calculate topology percent uptime between T_0 to T_1 using this metric data: |
| 95 | + // let s = sum topologyPartialUptimeSecs for each worker for each time buckets between T_0 and T_1 |
| 96 | + // topology percent uptime = s/(T_1-T_0) |
| 97 | + // Even if the number of workers change over time the value is still correct because I divide by TOPOLOGY_WORKERS. |
| 98 | + context.registerMetric("topologyPartialUptimeSecs", new IMetric() { |
| 99 | + private long _prevUptime = jvmRT.getUptime(); |
| 100 | + private final double NUM_WORKERS = RT.doubleCast(stormConf.get(Config.TOPOLOGY_WORKERS)); |
| 101 | + @Override |
| 102 | + public Object getValueAndReset() { |
| 103 | + long _nowUptime = jvmRT.getUptime(); |
| 104 | + double ret = (_nowUptime - _prevUptime)/1000.0/NUM_WORKERS; |
| 105 | + _prevUptime = _nowUptime; |
| 106 | + return ret; |
| 107 | + } |
| 108 | + }, bucketSize); |
| 109 | + |
| 110 | + context.registerMetric("startTimeSecs", new IMetric() { |
| 111 | + @Override |
| 112 | + public Object getValueAndReset() { |
| 113 | + return jvmRT.getStartTime()/1000.0; |
| 114 | + } |
| 115 | + }, bucketSize); |
| 116 | + |
| 117 | + context.registerMetric("newWorkerEvent", new IMetric() { |
| 118 | + boolean doEvent = true; |
| 119 | + |
| 120 | + @Override |
| 121 | + public Object getValueAndReset() { |
| 122 | + if (doEvent) { |
| 123 | + doEvent = false; |
| 124 | + return 1; |
| 125 | + } else return 0; |
| 126 | + } |
| 127 | + }, bucketSize); |
| 128 | + |
| 129 | + // This is metric data global to topology, but we don't support this concept, so put it here. |
| 130 | + // It's very useful to have time series of TOPOLOGY_WORKERS to compare actual worker count. |
| 131 | + context.registerMetric("configTopologyWorkers", new IMetric() { |
| 132 | + @Override |
| 133 | + public Object getValueAndReset() { |
| 134 | + return stormConf.get(Config.TOPOLOGY_WORKERS); |
| 135 | + } |
| 136 | + }, bucketSize); |
| 137 | + |
| 138 | + final MemoryMXBean jvmMemRT = ManagementFactory.getMemoryMXBean(); |
| 139 | + |
| 140 | + context.registerMetric("memory/heap", new MemoryUsageMetric(new AFn() { |
| 141 | + public Object invoke() { |
| 142 | + return jvmMemRT.getHeapMemoryUsage(); |
| 143 | + } |
| 144 | + }), bucketSize); |
| 145 | + context.registerMetric("memory/nonHeap", new MemoryUsageMetric(new AFn() { |
| 146 | + public Object invoke() { |
| 147 | + return jvmMemRT.getNonHeapMemoryUsage(); |
| 148 | + } |
| 149 | + }), bucketSize); |
| 150 | + |
| 151 | + for(GarbageCollectorMXBean b : ManagementFactory.getGarbageCollectorMXBeans()) { |
| 152 | + context.registerMetric("GC/" + b.getName().replaceAll("\\W", ""), new GarbageCollectorMetric(b), bucketSize); |
| 153 | + } |
| 154 | + } |
| 155 | + |
| 156 | + @Override |
| 157 | + public void execute(Tuple input) { |
| 158 | + throw new RuntimeException("Non-system tuples should never be sent to __system bolt."); |
| 159 | + } |
| 160 | + |
| 161 | + @Override |
| 162 | + public void cleanup() { |
| 163 | + } |
| 164 | +} |
0 commit comments