🗓️ 03112024 1135
📎

FLINK
ABSTRACT
  • How to implement streaming data processing pipelines and why Flink manages state
  • How to use event time to consistently compute accurate analytics
  • How to build event-driven applications on continuous streams
  • How Flink is able to provide fault-tolerant, stateful stream processing with exactly-once semantics

Stream processing​

ParadigmDescription
Batch ProcessingFor processing a bounded data stream
Stream ProcessingProcessing a unbounded data stream (input might never end)

Flink applications composed of:

  1. Streaming data sources (message_queue)
  2. Operators - for transforming data
  3. Send result streams to sinks (applications that need this data)

Parallel dataflows​

INFO

Programs in flink are inherently parallel and distributed

  • During execution, a stream has one or more stream partitions
  • Each operator has one or more operator subtasks
    • The operator subtasks are independent of one another
    • Execute in different threads / machines or containers
IMPORTANT

The number of operator subtasks is the parallelism of that particular operator

Transporting data between operators​

One-to-one forwarding​

Preserves partitioning / ordering of elements

Redistributing streams​

  • Changes partitioning of streams
  • Ordering is preserved within each pair of sending / receiving subtasks
  • Each operator subtask sends data to different target subtasks, depending on the selected transformation
  • Examples
    • keyBy()
    • broadcast()
    • rebalance()

Timely Stream Processing​

TLDR

Consider the timestamp at which the event happened rather than when the event was received

Stateful Stream Processing​

ABSTRACT

Flink’s operations can be stateful > How one event is handled can depend on the accumulated effect of all the events that came before it

State is managed locally on each parallel instance either on

  • JVM Heap
  • On-Disk data structures (if memory too large)

Fault Tolerance via State Snapshots​

  • Flink is able to provide fault-tolerant, exactly-once semantics (stream_processing_semantics) through a combination of state snapshots and stream replay

Snapshots​

  • Capture entire state of the distributed pipeline asynchronously
    • Offsets into input queues
    • State throughout the job graph (up to the point of data ingestion)

Failure​

  1. Sources are rewound
  2. State is restored
  3. Processing resumes

References​