Scalability, Availability, and Stability Patterns

Overview of scalability, availability and stability patterns, techniques and products.

1. Scalability, Availability & Stability Patterns Jonas Bonér CTO Typesafe twitter: @jboner
2. Outline
3. Outline
4. Outline
5. Outline
6. Outline
7. Introduction
8. Scalability Patterns
9. Managing Overload
10. Scale up vs Scale out?
11. General recommendations • Immutability as the default • Referential Transparency (FP) • Laziness • Think about your data: • Different data need different guarantees
12. Scalability Trade-offs
14. Trade-offs •Performance vs Scalability •Latency vs Throughput •Availability vs Consistency
15. Performance vs Scalability
16. How do I know if I have a performance problem?
17. How do I know if I have a performance problem? If your system is slow for a single user
18. How do I know if I have a scalability problem?
19. How do I know if I have a scalability problem? If your system is fast for a single user but slow under heavy load
20. Latency vs Throughput
21. You should strive for maximal throughput with acceptable latency
22. Availability vs Consistency
23. Brewer’s CAP theorem
24. You can only pick 2 Consistency Availability Partition tolerance At a given point in time
25. Centralized system • In a centralized system (RDBMS etc.) we don’t have network partitions, e.g. P in CAP • So you get both: •Availability •Consistency
26. Atomic Consistent Isolated Durable
27. Distributed system • In a distributed system we (will) have network partitions, e.g. P in CAP • So you get to only pick one: •Availability •Consistency
28. CAP in practice: • ...there are only two types of systems: 1. CP 2. AP • ...there is only one choice to make. In case of a network partition, what do you sacrifice? 1. C: Consistency 2. A: Availability
29. Basically Available Soft state Eventually consistent
30. Eventual Consistency ...is an interesting trade-off
31. Eventual Consistency ...is an interesting trade-off But let’s get back to that later
32. Availability Patterns
33. •Fail-over •Replication • Master-Slave • Tree replication • Master-Master • Buddy Replication Availability Patterns
34. What do we mean with Availability?
35. Fail-over
36. Fail-over Copyright Michael Nygaard
37. Fail-over But fail-over is not always this simple Copyright Michael Nygaard
38. Fail-over Copyright Michael Nygaard
39. Fail-back Copyright Michael Nygaard
40. Network fail-over
41. Replication
42. • Active replication - Push • Passive replication - Pull • Data not available, read from peer, then store it locally • Works well with timeout-based caches Replication
43. • Master-Slave replication • Tree Replication • Master-Master replication • Buddy replication Replication
44. Master-Slave Replication
45. Master-Slave Replication
46. Tree Replication
47. Master-Master Replication
48. Buddy Replication
49. Buddy Replication
50. Scalability Patterns: State
51. •Partitioning •HTTP Caching •RDBMS Sharding •NOSQL •Distributed Caching •Data Grids •Concurrency Scalability Patterns: State
52. Partitioning
53. HTTP Caching Reverse Proxy • Varnish • Squid • rack-cache • Pound • Nginx • Apache mod_proxy • Traffic Server
54. HTTP Caching CDN, Akamai
55. Generate Static Content Precompute content • Homegrown + cron or Quartz • Spring Batch • Gearman • Hadoop • Google Data Protocol • Amazon Elastic MapReduce
56. HTTP Caching First request
57. HTTP Caching Subsequent request
58. Service of Record SoR
59. Service of Record •Relational Databases (RDBMS) •NOSQL Databases
60. How to scale out RDBMS?
61. Sharding •Partitioning •Replication
62. Sharding: Partitioning
63. Sharding: Replication
64. ORM + rich domain model anti-pattern •Attempt: • Read an object from DB •Result: • You sit with your whole database in your lap
65. Think about your data • When do you need ACID? • When is Eventually Consistent a better fit? • Different kinds of data has different needs Think again
66. When is a RDBMS not good enough?
67. Scaling reads to a RDBMS is hard
68. Scaling writes to a RDBMS is impossible
69. Do we really need a RDBMS?
70. Do we really need a RDBMS? Sometimes...
71. Do we really need a RDBMS?
72. Do we really need a RDBMS? But many times we don’t
73. NOSQL (Not Only SQL)
74. •Key-Value databases •Column databases •Document databases •Graph databases •Datastructure databases NOSQL
75. Who’s ACID? • Relational DBs (MySQL, Oracle, Postgres) • Object DBs (Gemstone, db4o) • Clustering products (Coherence, Terracotta) • Most caching products (ehcache)
76. Who’s BASE? Distributed databases • Cassandra • Riak • Voldemort • Dynomite, • SimpleDB • etc.
77. • Google: Bigtable • Amazon: Dynamo • Amazon: SimpleDB • Yahoo: HBase • Facebook: Cassandra • LinkedIn: Voldemort NOSQL in the wild
78. But first some background...
79. • Distributed Hash Tables (DHT) • Scalable • Partitioned • Fault-tolerant • Decentralized • Peer to peer • Popularized • Node ring • Consistent Hashing Chord & Pastry
80. Node ring with Consistent Hashing Find data in log(N) jumps
81. “How can we build a DB on top of Google File System?” • Paper: Bigtable: A distributed storage system for structured data, 2006 • Rich data-model, structured storage • Clones: HBase Hypertable Neptune Bigtable
82. “How can we build a distributed hash table for the data center?” • Paper: Dynamo: Amazon’s highly available key- value store, 2007 • Focus: partitioning, replication and availability • Eventually Consistent • Clones: Voldemort Dynomite Dynamo
83. Types of NOSQL stores • Key-Value databases (Voldemort, Dynomite) • Column databases (Cassandra, Vertica, Sybase IQ) • Document databases (MongoDB, CouchDB) • Graph databases (Neo4J, AllegroGraph) • Datastructure databases (Redis, Hazelcast)
84. Distributed Caching
85. •Write-through •Write-behind •Eviction Policies •Replication •Peer-To-Peer (P2P) Distributed Caching
86. Write-through
87. Write-behind
88. Eviction policies • TTL (time to live) • Bounded FIFO (first in first out) • Bounded LIFO (last in first out) • Explicit cache invalidation
89. Peer-To-Peer • Decentralized • No “special” or “blessed” nodes • Nodes can join and leave as they please
90. •EHCache •JBoss Cache •OSCache •memcached Distributed Caching Products
91. memcached • Very fast • Simple • Key-Value (string -­‐>  binary) • Clients for most languages • Distributed • Not replicated - so 1/N chance for local access in cluster
92. Data Grids / Clustering
93. Data Grids/Clustering Parallel data storage • Data replication • Data partitioning • Continuous availability • Data invalidation • Fail-over • C + P in CAP
94. Data Grids/Clustering Products • Coherence • Terracotta • GigaSpaces • GemStone • Tibco Active Matrix • Hazelcast
95. Concurrency
96. •Shared-State Concurrency •Message-Passing Concurrency •Dataflow Concurrency •Software Transactional Memory Concurrency
97. Shared-State Concurrency
98. •Everyone can access anything anytime •Totally indeterministic •Introduce determinism at well-defined places... •...using locks Shared-State Concurrency
99. •Problems with locks: • Locks do not compose • Taking too few locks • Taking too many locks • Taking the wrong locks • Taking locks in the wrong order • Error recovery is hard Shared-State Concurrency
100. Please use java.util.concurrent.* • ConcurrentHashMap • BlockingQueue • ConcurrentQueue   • ExecutorService • ReentrantReadWriteLock • CountDownLatch • ParallelArray • and  much  much  more.. Shared-State Concurrency
101. Message-Passing Concurrency
102. •Originates in a 1973 paper by Carl Hewitt •Implemented in Erlang, Occam, Oz •Encapsulates state and behavior •Closer to the definition of OO than classes Actors
103. Actors • Share NOTHING • Isolated lightweight processes • Communicates through messages • Asynchronous and non-blocking • No shared state … hence, nothing to synchronize. • Each actor has a mailbox (message queue)
104. • Easier to reason about • Raised abstraction level • Easier to avoid –Race conditions –Deadlocks –Starvation –Live locks Actors
105. • Akka (Java/Scala) • scalaz actors (Scala) • Lift Actors (Scala) • Scala Actors (Scala) • Kilim (Java) • Jetlang (Java) • Actor’s Guild (Java) • Actorom (Java) • FunctionalJava (Java) • GPars (Groovy) Actor libs for the JVM
106. Dataflow Concurrency
107. • Declarative • No observable non-determinism • Data-driven – threads block until data is available • On-demand, lazy • No difference between: • Concurrent & • Sequential code • Limitations: can’t have side-effects Dataflow Concurrency
108. STM: Software Transactional Memory
109. STM: overview • See the memory (heap and stack) as a transactional dataset • Similar to a database • begin • commit • abort/rollback • Transactions are retried automatically upon collision • Rolls back the memory on abort
110. • Transactions can nest • Transactions compose (yipee!!) atomic  {              ...              atomic  {                    ...                }        } STM: overview
111. All operations in scope of a transaction: l Need to be idempotent STM: restrictions
112. • Akka (Java/Scala) • Multiverse (Java) • Clojure STM (Clojure) • CCSTM (Scala) • Deuce STM (Java) STM libs for the JVM
113. Scalability Patterns: Behavior
114. •Event-Driven Architecture •Compute Grids •Load-balancing •Parallel Computing Scalability Patterns: Behavior
115. Event-Driven Architecture “Four years from now, ‘mere mortals’ will begin to adopt an event-driven architecture (EDA) for the sort of complex event processing that has been attempted only by software gurus [until now]” --Roy Schulte (Gartner), 2003
116. • Domain Events • Event Sourcing • Command and Query Responsibility Segregation (CQRS) pattern • Event Stream Processing • Messaging • Enterprise Service Bus • Actors • Enterprise Integration Architecture (EIA) Event-Driven Architecture
117. Domain Events “It's really become clear to me in the last couple of years that we need a new building block and that is the Domain Events” -- Eric Evans, 2009
118. Domain Events “Domain Events represent the state of entities at a given time when an important event occurred and decouple subsystems with event streams. Domain Events give us clearer, more expressive models in those cases.” -- Eric Evans, 2009
119. Domain Events “State transitions are an important part of our problem space and should be modeled within our domain.” -- Greg Young, 2008
120. Event Sourcing • Every state change is materialized in an Event • All Events are sent to an EventProcessor • EventProcessor stores all events in an Event Log • System can be reset and Event Log replayed • No need for ORM, just persist the Events • Many different EventListeners can be added to EventProcessor (or listen directly on the Event log)
121. Event Sourcing
122. “A single model cannot be appropriate for reporting, searching and transactional behavior.” -- Greg Young, 2008 Command and Query Responsibility Segregation (CQRS) pattern
123. Bidirectional Bidirectional
125. Unidirectional Unidirectional Unidirectional
129. CQRS in a nutshell • All state changes are represented by Domain Events • Aggregate roots receive Commands and publish Events • Reporting (query database) is updated as a result of the published Events • All Queries from Presentation go directly to Reporting and the Domain is not involved
130. CQRS Copyright by Axis Framework
131. CQRS: Benefits • Fully encapsulated domain that only exposes behavior • Queries do not use the domain model • No object-relational impedance mismatch • Bullet-proof auditing and historical tracing • Easy integration with external systems • Performance and scalability
132. Event Stream Processing select  *  from   Withdrawal(amount>=200).win:length(5)
133. Event Stream Processing Products • Esper (Open Source) • StreamBase • RuleCast
134. Messaging • Publish-Subscribe • Point-to-Point • Store-forward • Request-Reply
135. Publish-Subscribe
136. Point-to-Point
137. Store-Forward Durability, event log, auditing etc.
138. Request-Reply F.e. AMQP’s ‘replyTo’ header
139. Messaging • Standards: • AMQP • JMS • Products: • RabbitMQ (AMQP) • ActiveMQ (JMS) • Tibco • MQSeries • etc
140. ESB
141. ESB products • ServiceMix (Open Source) • Mule (Open Source) • Open ESB (Open Source) • Sonic ESB • WebSphere ESB • Oracle ESB • Tibco • BizTalk Server
142. Actors • Fire-forget • Async send • Fire-And-Receive-Eventually • Async send + wait on Future for reply
143. Enterprise Integration Patterns
144. Enterprise Integration Patterns Apache Camel • More than 80 endpoints • XML (Spring) DSL • Scala DSL
145. Compute Grids
146. Compute Grids Parallel execution • Divide and conquer 1. Split up job in independent tasks 2. Execute tasks in parallel 3. Aggregate and return result • MapReduce - Master/Worker
147. Compute Grids Parallel execution • Automatic provisioning • Load balancing • Fail-over • Topology resolution
148. Compute Grids Products • Platform • DataSynapse • Google MapReduce • Hadoop • GigaSpaces • GridGain
149. Load balancing
150. • Random allocation • Round robin allocation • Weighted allocation • Dynamic load balancing • Least connections • Least server CPU • etc. Load balancing
151. Load balancing • DNS Round Robin (simplest) • Ask DNS for IP for host • Get a new IP every time • Reverse Proxy (better) • Hardware Load Balancing
152. Load balancing products • Reverse Proxies: • Apache mod_proxy (OSS) • HAProxy (OSS) • Squid (OSS) • Nginx (OSS) • Hardware Load Balancers: • BIG-IP • Cisco
153. Parallel Computing
154. • UE: Unit of Execution • Process • Thread • Coroutine • Actor Parallel Computing • SPMD Pattern • Master/Worker Pattern • Loop Parallelism Pattern • Fork/Join Pattern • MapReduce Pattern
155. SPMD Pattern • Single Program Multiple Data • Very generic pattern, used in many other patterns • Use a single program for all the UEs • Use the UE’s ID to select different pathways through the program. F.e: • Branching on ID • Use ID in loop index to split loops • Keep interactions between UEs explicit
156. Master/Worker
157. Master/Worker • Good scalability • Automatic load-balancing • How to detect termination? • Bag of tasks is empty • Poison pill • If we bottleneck on single queue? • Use multiple work queues • Work stealing • What about fault tolerance? • Use “in-progress” queue
158. Loop Parallelism •Workflow 1.Find the loops that are bottlenecks 2.Eliminate coupling between loop iterations 3.Parallelize the loop •If too few iterations to pull its weight • Merge loops • Coalesce nested loops •OpenMP • omp  parallel  for
159. What if task creation can’t be handled by: • parallelizing loops (Loop Parallelism) • putting them on work queues (Master/Worker)
160. What if task creation can’t be handled by: • parallelizing loops (Loop Parallelism) • putting them on work queues (Master/Worker) Enter Fork/Join
161. •Use when relationship between tasks is simple •Good for recursive data processing •Can use work-stealing 1. Fork: Tasks are dynamically created 2. Join: Tasks are later terminated and data aggregated Fork/Join
162. Fork/Join •Direct task/UE mapping • 1-1 mapping between Task/UE • Problem: Dynamic UE creation is expensive •Indirect task/UE mapping • Pool the UE • Control (constrain) the resource allocation • Automatic load balancing
163. Java 7 ParallelArray (Fork/Join DSL) Fork/Join
164. Java 7 ParallelArray (Fork/Join DSL) ParallelArray  students  =      new  ParallelArray(fjPool,  data); double  bestGpa  =  students.withFilter(isSenior)                                                    .withMapping(selectGpa)                                                    .max(); Fork/Join
165. • Origin from Google paper 2004 • Used internally @ Google • Variation of Fork/Join • Work divided upfront not dynamically • Usually distributed • Normally used for massive data crunching MapReduce
166. • Hadoop (OSS), used @ Yahoo • Amazon Elastic MapReduce • Many NOSQL DBs utilizes it for searching/querying MapReduce Products
167. MapReduce
168. Parallel Computing products • MPI • OpenMP • JSR166 Fork/Join • java.util.concurrent • ExecutorService, BlockingQueue etc. • ProActive Parallel Suite • CommonJ WorkManager (JEE)
169. Stability Patterns
170. •Timeouts •Circuit Breaker •Let-it-crash •Fail fast •Bulkheads •Steady State •Throttling Stability Patterns
171. Timeouts Always use timeouts (if possible): • Thread.wait(timeout) • reentrantLock.tryLock • blockingQueue.poll(timeout,  timeUnit)/ offer(..) • futureTask.get(timeout,  timeUnit) • socket.setSoTimeOut(timeout) • etc.
172. Circuit Breaker
173. Let it crash • Embrace failure as a natural state in the life-cycle of the application • Instead of trying to prevent it; manage it • Process supervision • Supervisor hierarchies (from Erlang)
174. Restart Strategy OneForOne
175. Restart Strategy OneForOne
176. Restart Strategy OneForOne
177. Restart Strategy AllForOne
178. Restart Strategy AllForOne
179. Restart Strategy AllForOne
180. Restart Strategy AllForOne
181. Supervisor Hierarchies
182. Supervisor Hierarchies
183. Supervisor Hierarchies
184. Supervisor Hierarchies
185. Fail fast • Avoid “slow responses” • Separate: • SystemError - resources not available • ApplicationError - bad user input etc • Verify resource availability before starting expensive task • Input validation immediately
186. Bulkheads
187. Bulkheads • Partition and tolerate failure in one part • Redundancy • Applies to threads as well: • One pool for admin tasks to be able to perform tasks even though all threads are blocked
188. Steady State • Clean up after you • Logging: • RollingFileAppender (log4j) • logrotate (Unix) • Scribe - server for aggregating streaming log data • Always put logs on separate disk
189. Throttling • Maintain a steady pace • Count requests • If limit reached, back-off (drop, raise error) • Queue requests • Used in for example Staged Event-Driven Architecture (SEDA)
190. ?
191. thanks for listening
192. Extra material
193. Client-side consistency • Strong consistency • Weak consistency • Eventually consistent • Never consistent
194. Client-side Eventual Consistency levels • Casual consistency • Read-your-writes consistency (important) • Session consistency • Monotonic read consistency (important) • Monotonic write consistency
195. Server-side consistency N = the number of nodes that store replicas of the data W = the number of replicas that need to acknowledge the receipt of the update before the update completes R = the number of replicas that are contacted when a data object is accessed through a read operation
196. Server-side consistency W + R > N strong consistency W + R <= N eventual consistency