Core.async

Introduction to core.async with some remarks on concurrency models in general

Transcript

1. CORE.ASYNC WITH SOME REMARKS ON CONCURRENCY MODELS IN GENERAL / Peter Brachwitz @p_brc
2. AGENDA 1. Popular concurrency models in comparison [15 min.] 2. A taste of core.async [15 min]
3. WHY? We want to be more 'reactive' (we can discuss whether 'reactive' is a 'thing' later) We want to make use of our multi-core hardware Low-level concurrency is hard
4. WHAT? Definitions: Concurrency vs. parallelism and what about multi-core Here: 'processes' working together in a meaningful manner
5. APPROACHES TO CONCURRENT PROGRAMMING 1. Shared memory: Threads and locks * 2. Message passing: 1. Actors 2. CSP
6. WHAT NOT (TODAY)? Refs Atoms Agents Futures, Delays
7. SHARED MEMORY, THREADS AND LOCKS 'Location based programming' (R. Hickey) Naive approaches don't scale very well. Think: web server using a thread per request. Hard to reason about, error-prone
8. MITIGATION: EVENT-LOOP ARCHITECTURE Single-threaded, non-blocking Handle blocking or time-consuming computations in different execution contexts Handle the completion of these computations via callbacks (callback hell)
9. THE ACTOR MODEL [originally] Hewitt, Bishop, Steiger 1973
10. 'Objects' communicating via messages Sequencing of messages via actor-local queues ('mailbox') You need to know your peer No inherent back-pressure (can be implemented on application level)
11. CSP [originally] Hoare 1978
12. CSP in core.async flavour Channels as conveyor belts between 'processes' Sequential looking asynchrony via go blocks Channels are bounded: back pressure Not for distributed computing: everything within one VM
13. COMPARISON WITH ACTOR SYSTEMS Anonymity vs. Identity Rendezvous vs. Asynchrony Explicit vs. implicit 'channels' Local vs. Distributed
14. LET'S LOOK AT SOME CODE!
15. SOURCES core.async on Github Tim Baldrige's Clojure Conj 2013 talk Rich Hickey speaking at Strange Loop 2013 Gul Agha's dissertation