Hadoop Summit Europe 2014: Apache Storm Architecture
- 1. © Hortonworks Inc. 2011 P. Taylor Goetz Apache Storm Committer tgoetz@hortonworks.com @ptgoetz Apache Storm Architecture and Integration Real-Time Big Data
- 2. Shedding Light on Data
- 3. Shedding Light on Big Data
- 4. Shedding Light on Big Data In Real Time
- 7. Storm is Streaming Key enabler of the Lamda Architecture
- 9. Storm is Fast Clocked at 1M+ messages per second per node
- 11. Storm is Scalable Thousands of workers per cluster
- 12. Storm is Fault Tolerant
- 13. Storm is Fault Tolerant Failure is expected, and embraced
- 15. Storm is Reliable Guaranteed message delivery
- 16. Storm is Reliable Exactly-once semantics
- 17. Conceptual Model
- 18. Tuple {…}
- 19. Tuple {…} • Core Unit of Data • Immutable Set of Key/Value Pairs
- 20. Streams {…} {…} {…} {…} {…} {…} {…} Unbounded Sequence of Tuples
- 21. Spouts
- 22. Spouts • Source of Streams • Wraps a streaming data source and emits Tuples {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…}
- 23. Spout API public interface ISpout extends Serializable {! ! void open(Map conf, ! ! TopologyContext context, ! ! ! ! SpoutOutputCollector collector);! ! void close();! ! void activate();! ! void deactivate();! ! void nextTuple();! ! void ack(Object msgId);! ! void fail(Object msgId);! } Lifecycle API
- 24. Spout API public interface ISpout extends Serializable {! ! void open(Map conf, ! ! TopologyContext context, ! ! ! ! SpoutOutputCollector collector);! ! void close();! ! void activate();! ! void deactivate();! ! void nextTuple();! ! void ack(Object msgId);! ! void fail(Object msgId);! } Core API
- 25. Spout API public interface ISpout extends Serializable {! ! void open(Map conf, ! ! TopologyContext context, ! ! ! ! SpoutOutputCollector collector);! ! void close();! ! void activate();! ! void deactivate();! ! void nextTuple();! ! void ack(Object msgId);! ! void fail(Object msgId);! } Reliability API
- 26. Bolts
- 27. Bolts • Core functions of a streaming computation • Receive tuples and do stuff • Optionally emit additional tuples
- 28. Bolts • Write to a data store
- 29. Bolts • Read from a data store
- 31. {…} {…} {…} {…} {…} {…} {…} Bolts • (Optionally) Emit additional streams {…} {…} {…} {…} {…} {…} {…}
- 32. Bolt API public interface IBolt extends Serializable {! ! void prepare(Map stormConf, ! TopologyContext context, ! OutputCollector collector);! ! void cleanup();! ! ! void execute(Tuple input);! ! ! } Lifecycle API
- 33. Bolt API public interface IBolt extends Serializable {! ! void prepare(Map stormConf, ! TopologyContext context, ! OutputCollector collector);! ! void cleanup();! ! ! void execute(Tuple input);! ! ! } Core API
- 34. Bolt Output API public interface IOutputCollector extends IErrorReporter {! ! List<Integer> emit(String streamId, ! Collection<Tuple> anchors, ! List<Object> tuple);! ! ! void emitDirect(int taskId, ! String streamId, ! Collection<Tuple> anchors, ! List<Object> tuple);! ! ! void ack(Tuple input);! ! ! void fail(Tuple input);! } Core API
- 35. Bolt Output API public interface IOutputCollector extends IErrorReporter {! ! List<Integer> emit(String streamId, ! Collection<Tuple> anchors, ! List<Object> tuple);! ! ! void emitDirect(int taskId, ! String streamId, ! Collection<Tuple> anchors, ! List<Object> tuple);! ! ! void ack(Tuple input);! ! ! void fail(Tuple input);! } Reliability API
- 36. Topologies
- 37. Topologies
- 38. Topologies • DAG of Spouts and Bolts • Data Flow Representation • Streaming Computation
- 39. Topologies • Storm executes spouts and bolts as individual Tasks that run in parallel on multiple machines.
- 40. Stream Groupings
- 41. Stream Groupings Stream Groupings determine how Storm routes Tuples between tasks in a topology
- 43. Stream Groupings LocalOrShuffle! ! Randomized round-robin. (With a preference for intra-worker Tasks)
- 44. Stream Groupings Fields Grouping! ! Ensures all Tuples with with the same field value(s) are always routed to the same task.
- 45. Stream Groupings Fields Grouping! ! Ensures all Tuples with with the same field value(s) are always routed to the same task. ! (this is a simple hash of the field values, modulo the number of tasks)
- 46. Physical View
- 47. Physical View ZooKeeperNimbus Supervisor Supervisor Supervisor Supervisor Worker* Worker* Worker* Worker*
- 48. Topology Deployment ZooKeeperNimbus Supervisor Supervisor Supervisor Supervisor Topology Submitter Topology Submitter uploads topology:! • topology.jar! • topology.ser! • conf.ser $ bin/storm jar
- 49. Topology Deployment Nimbus calculates assignments and sends to Zookeeper ZooKeeperNimbus Supervisor Supervisor Supervisor Supervisor Topology Submitter
- 50. Topology Deployment Supervisor nodes receive assignment information ! via Zookeeper watches. ZooKeeperNimbus Supervisor Supervisor Supervisor Supervisor Topology Submitter
- 51. Topology Deployment Supervisor nodes download topology from Nimbus:! • topology.jar! • topology.ser! • conf.ser ZooKeeperNimbus Supervisor Supervisor Supervisor Supervisor Topology Submitter
- 52. Topology Deployment Supervisors spawn workers (JVM processes) to start the topology ZooKeeperNimbus Supervisor Supervisor Supervisor Supervisor Topology Submitter Worker Worker Worker Worker
- 53. Fault Tolerance
- 54. Fault Tolerance Workers heartbeat back to Supervisors and Nimbus via ZooKeeper, ! as well as locally. ZooKeeperNimbus Supervisor Supervisor Supervisor Supervisor Topology Submitter Worker Worker Worker Worker
- 55. Fault Tolerance If a worker dies (fails to heartbeat), the Supervisor will restart it ZooKeeperNimbus Supervisor Supervisor Supervisor Supervisor Topology Submitter Worker Worker Worker Worker X
- 56. Fault Tolerance If a worker dies repeatedly, Nimbus will reassign the work to other! nodes in the cluster. ZooKeeperNimbus Supervisor Supervisor Supervisor Supervisor Topology Submitter Worker Worker Worker Worker X
- 57. Fault Tolerance If a supervisor node dies, Nimbus will reassign the work to other nodes. ZooKeeperNimbus Supervisor Supervisor Supervisor Supervisor Topology Submitter Worker Worker Worker Worker X X
- 58. Fault Tolerance If Nimbus dies, topologies will continue to function normally,! but won’t be able to perform reassignments. ZooKeeperNimbus Supervisor Supervisor Supervisor Supervisor Topology Submitter Worker Worker Worker Worker X
- 59. Parallelism Scaling a Distributed Computation
- 60. Parallelism Worker (JVM) Executor (Thread) Executor (Thread) Executor (Thread) Task Task Task 1 Worker, Parallelism = 1
- 61. Parallelism Worker (JVM) Executor (Thread) Executor (Thread) Executor (Thread) Task Task Task Executor (Thread) Task 1 Worker, Parallelism = 2
- 62. Parallelism Worker (JVM) Executor (Thread) Executor (Thread) Task Task Executor (Thread) Task Task 1 Worker, Parallelism = 2, NumTasks = 2
- 63. Parallelism 3 Workers, Parallelism = 1, NumTasks = 1 Worker (JVM)Worker (JVM)Worker (JVM) Executor (Thread) Executor (Thread) Executor (Thread) Task Task Task
- 66. Worker Internal Messaging Worker Receive Thread Worker Port List<List<Tuple>> Receive Buffer Executor Thread * Inbound Queue Outbound Queue Router Send Thread Worker Transfer Thread List<List<Tuple>> Transfer Buffer To Other Workers Task (Spout/Bolt) Task (Spout/Bolt) Task(s) (Spout/Bolt)
- 67. Reliable Processing At Least Once
- 68. Reliable Processing Bolts may emit Tuples Anchored to one received. Tuple “B” is a descendant of Tuple “A” {A} {B}
- 69. Reliable Processing Multiple Anchorings form a Tuple tree (bolts not shown) {A} {B} {C} {D} {E} {F} {G} {H}
- 70. Reliable Processing Bolts can Acknowledge that a tuple has been processed successfully. {A} {B} ACK
- 71. Reliable Processing Acks are delivered via a system-level bolt ACK {A} {B} Acker Bolt ackack
- 72. Reliable Processing Bolts can also Fail a tuple to trigger a spout to replay the original. FAIL {A} {B} Acker Bolt failfail
- 73. Reliable Processing Any failure in the Tuple tree will trigger a replay of the original tuple {A} {B} {C} {D} {E} {F} {G} {H} X X
- 74. Reliable Processing How to track a large-scale tuple tree efficiently?
- 75. Reliable Processing A single 64-bit integer.
- 76. XOR Magic Long a, b, c = Random.nextLong();
- 77. XOR Magic Long a, b, c = Random.nextLong();! ! a ^ a == 0
- 78. XOR Magic Long a, b, c = Random.nextLong();! ! a ^ a == 0! ! a ^ a ^ b != 0
- 79. XOR Magic Long a, b, c = Random.nextLong();! ! a ^ a == 0! ! a ^ a ^ b != 0! ! a ^ a ^ b ^ b == 0
- 80. XOR Magic Long a, b, c = Random.nextLong();! ! a ^ (a ^ b) ^ c ^ (b ^ c) == 0
- 81. XOR Magic Long a, b, c = Random.nextLong();! ! a ^ (a ^ b) ^ c ^ (b ^ c) == 0 Acks can arrive asynchronously, in any order
- 82. Trident
- 83. Trident High-level abstraction built on Storm’s core primitives.
- 84. Trident Built-in support for: • Merges and Joins • Aggregations • Groupings • Functions • Filters
- 85. Trident Stateful, incremental processing on top of any persistence store.
- 88. Trident Fluent, Stream-Oriented API TridentTopology topology = new TridentTopology();! FixedBatchSpout spout = new FixedBatchSpout(…);! Stream stream = topology.newStream("words", spout);! ! stream.each(…, new MyFunction())! .groupBy()! .each(…, new MyFilter())! .persistentAggregate(…);! User-defined functions
- 89. Trident Micro-Batch Oriented Tuple Micro-Batch {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…}
- 90. Trident Trident Batches are Ordered Tuple Micro-Batch {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} Tuple Micro-Batch {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} Batch #1 Batch #2
- 91. Trident Trident Batches can be Partitioned Tuple Micro-Batch {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…}
- 92. Trident Trident Batches can be Partitioned Tuple Micro-Batch {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} Partition Operation Partition A {…} {…} {…}{…} Partition B {…} {…} {…}{…} Partition C {…} {…} {…}{…} Partition D {…} {…} {…}{…}
- 93. Trident Operation Types 1. Local Operations (Functions/Filters) 2. Repartitioning Operations (Stream Groupings, etc.) 3. Aggregations 4. Merges/Joins
- 95. Trident Toplogies Partitioning operations define the boundaries between bolts, and thus network transfer and parallelism
- 96. Trident Topologies each each shuffle Function Filter partition persist Bolt 1 Bolt 2 shuffleGrouping() Partitioning! Operation
- 98. Trident Batch Coordination Trident SpoutMaster Batch Coordinator User Logic next batch {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} {…} commit
- 100. Controlling Deployment How do you control where spouts and bolts get deployed in a cluster?
- 101. Controlling Deployment How do you control where spouts and bolts get deployed in a cluster? Plug-able Schedulers
- 102. Controlling Deployment How do you control where spouts and bolts get deployed in a cluster? Isolation Scheduler
- 103. Wait… Nimbus, Supervisor, Schedulers… ! Doesn’t that sound kind of like resource negotiation?
- 104. Storm on YARN HDFS2 (redundant, reliable storage) YARN (cluster resource management) MapReduce (batch) Apache STORM (streaming) HADOOP 2.0 Tez (interactive) Multi Use Data Platform Batch, Interactive, Online, Streaming, …
- 105. Storm on YARN HDFS2 (redundant, reliable storage) YARN (cluster resource management) MapReduce (batch) Apache STORM (streaming) HADOOP 2.0 Tez (interactive) Multi Use Data Platform Batch, Interactive, Online, Streaming, … Batch and real-time on the same cluster
- 106. Storm on YARN HDFS2 (redundant, reliable storage) YARN (cluster resource management) MapReduce (batch) Apache STORM (streaming) HADOOP 2.0 Tez (interactive) Multi Use Data Platform Batch, Interactive, Online, Streaming, … Security and Multi-tenancy
- 107. Storm on YARN HDFS2 (redundant, reliable storage) YARN (cluster resource management) MapReduce (batch) Apache STORM (streaming) HADOOP 2.0 Tez (interactive) Multi Use Data Platform Batch, Interactive, Online, Streaming, … Elasticity
- 108. Storm on YARN Nimbus Resource Management, Scheduling Supervisor Node and Process management Workers Runs topology tasks YARN RM Resource Management Storm AM Manage Topology Containers Runs topology tasks YARN NM Process Management Storm’s resource management system maps very naturally to the YARN model.
- 109. Storm on YARN Nimbus Resource Management, Scheduling Supervisor Node and Process management Workers Runs topology tasks YARN RM Resource Management Storm AM Manage Topology Containers Runs topology tasks YARN NM Process Management High Availability
- 110. Storm on YARN Nimbus Resource Management, Scheduling Supervisor Node and Process management Workers Runs topology tasks YARN RM Resource Management Storm AM Manage Topology Containers Runs topology tasks YARN NM Process Management Detect and scale around bottlenecks
- 111. Storm on YARN Nimbus Resource Management, Scheduling Supervisor Node and Process management Workers Runs topology tasks YARN RM Resource Management Storm AM Manage Topology Containers Runs topology tasks YARN NM Process Management Optimize for available resources
- 113. Thank You! Contributions welcome. Join the storm community at: http://storm.incubator.apache.org P. Taylor Goetz tgoetz@hortonworks.com @ptgoetz