Data Stream Processing and Analytics Vasiliki (Vasia) Kalavri
 vkalavri@bu.edu Spring 2020 4/09: Flow control and load shedding ??? Vasiliki Kalavri | Boston University 2020 Keeping up with the producers queue: what if the queue grows larger than available memory? • block the producer (back-pressure, flow control) 2 ??? Vasiliki Kalavri | Boston University 2020 Load management approaches 3 ! Load applications with strict latency constraints that can tolerate approximate results. Slow down the flow of data: • The system buffers excess data for later processing, once input rates stabilize.

Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions paper ‣ A flow-based neural network model from the WaveGlow: A Flow-based Generative Network for Speech Synthesis paper. This model script Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions paper ‣ A flow-based neural network model from the WaveGlow: A Flow-based Generative Network for Speech Synthesis paper. This model script Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions paper ‣ A flow-based neural network model from the WaveGlow: A Flow-based Generative Network for Speech Synthesis paper. This model script

Upstream&C onnpool Pilot SOFARegistry Consul Nacos Consul Others Control Plane MCP Data flow Control flow xDS Discovery Client Metrics Admin REST API Request Request MOSN 做业务扩展 • 扩展非 xDS filter Cluster subset LB Cluster Manager/xDS Discovery 1 2 4 1 2 Envoy MOSN Data flow Control flow GoLang L7 extension SDK GoLang L7 extension filter 5 Other http filter(via GoLang) CGO Manager/xDS Pilot SOFARegistry Antvip Nacos Antvip Others Control Plane MCP Data flow Control flow xDS Discovery Client Admin API REST API Request REST API Callback Callback MOSN

. . . . . . . . . . . . . . . . . . . . 129 6.3.2.3 flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 6.3.2.4 flow_from_directory . . . . . . . . . . . . . . . . . . . 是单个张量（对于单输出 Sequential 模型）。 模型 42 • __**kwargs__: 当使用 Theano/CNTK 后端时，这些参数被传入 K.function。当使用 Tensor- Flow 后端时，这些参数被传递到 tf.Session.run。 异常 • ValueError: 如果 optimizer, loss, metrics 或 sample_weight_mode 这些参数不合法。 是单个张量（单输出模型），张量列表，或一个映射输出名称到目标张量的字典。 • __**kwargs__: 当使用 Theano/CNTK 后端时，这些参数被传入 K.function。当使用 Tensor- Flow 后端时，这些参数被传递到 tf.Session.run。 异常 • ValueError: 如果 optimizer, loss, metrics 或 sample_weight_mode 这些参数不合法。

Istio and manually register the services running #IstioCon V0.2 Mesh Expansion (cont.) ● Traffic flow (VM -> Container) 1. Dnsmasq accepts DNS queries 2. Access the built-in Kube DNS (exposed by ILB) intercepted by the sidecar proxy 5. xDS ■ Traffic forwarded to ingress in the mesh ● Traffic flow (Container -> VM) 1. Manual registration istioctl -n onprem register mysql 1.2.3.4 3306 #IstioCon Network #IstioCon VM Support – Multiple Networks #IstioCon Current State of VM Support ● Traffic flow ○ VM connects up to the Istio control plane through a Gateway ○ WorkloadEntry created ■ VM sidecar

[REJECTREQUEST]system busy, start flow control for a while  too many requests and system thread pool busy, RejectedExecutionExc eption  [PC_SYNCHRONIZED]broker busy, start flow control for a while  [PC [PCBUSY_CLEAN_QUEUE]broker busy, start flow control for a while, period in queue: %sms, size of queue: %d 二、原理解读 在进行消息中间件的选型时，如果待选中间件在功能上、性能上都能满足业务的情况下， 我各个建议把中间件的实现语言这个因素也考虑进去，毕竟选择一门用自己擅长的语言实现 的中间件 offerFirst(byteBuffer); } 其调用栈如下： 从上面的分析看来，并不会随着消息的不断写入而导致内存溢出。 三、现象解答 1. [REJECTREQUEST]system busy, start flow control for a while 其抛出的源码入口点：NettyRemotingAbstract#processRequestCommand，上 面的原理分析部分已经详细介绍其实现原理，总结如下。

results of the computation rather than the execution flow. • Imperative languages are used to describe plans of operators the streams must flow through. • Pattern-based languages specify conditions events in a period during which a user was active 17 Vasiliki Kalavri | Boston University 2020 Flow Management Operators (I) • Join operators merge two streams by matching elements satisfying a condition • it is blocking and must be defined over a window 18 Vasiliki Kalavri | Boston University 2020 Flow Management Operators (II) • Duplicate/Copy Operator replicates a stream, commonly to be used as

Late Data (1/3) ▶ Spark streaming uses watermarks to measure progress in event time. ▶ Watermarks flow as part of the data stream and carry a timestamp t. ▶ A W(t) declares that event time has reached "10 minutes", "5 minutes"), col("word")).count() 67 / 79 Flink 68 / 79 Flink ▶ Distributed data flow processing system ▶ Unified real-time stream and batch processing ▶ Process unbounded and bounded ▶ Periodically, the data sources inject checkpoint barriers into the data stream. ▶ The barriers flow through the data stream, and trigger operators to emit all records that depend only on records before

edge and route traffic to the mesh- managed services inside the cluster. Two-tier Gateway Traffic Flow Cloud Vendor Gateway Consolidation TSB allows service discovery and communication via the NodePort XCP Edge ● Upstream Istio ● XCP Central -> Edge ● TSB CR -> Istio CR TSB Config Data Flow Cluster Onboarding Flow 1. Creating cluster object 2. Deploy Operators: Control plane & data plane 3. Configuring

y,z) • Concurrency ( asynchronousness ) • Shared the same address space • Non blocking for main flow • Ideal for background replication Channel for Communication between Threads • Syntax • No buffering: (logging) • Error handling • Arbitrary exit • Graceful retry • Auto recovery •Really messy in control flow State Machine Comes in to Rescue • Simple is beautiful! • A divide-and-conquer mindset to simplify coordination between goroutines. •State machine pattern simplifies the implementation of control flow. •Try it, love it and contribute to it! https://github.com/vmware/harbor