Working with Asynchrony Generically: A Tour of C++ Executors ## ERIC NIEBLER 20 21 October 24-29 ## TALK OUTLINE ## Part 1: 1. Goals for the Executors proposal 2. Some simple examples, intro to senders ## Part 2: 1. Structured concurrency 2. Cancellation 3. An extended example ## GOALS FOR THE EXECUTORS PROPOSAL The vision: “An asynchronous analog of the STL” • A full suite of standard async algorithms com/facebookexperimental/libunifex ## +21 ## Working with Asynchrony Generically: A Tour of C++ Executors ## ERIC NIEBLER 20 21 October 24-29 ## SUMMARY FROM PART 1 1. Vision: “An asynchronous analogue

# Design patterns for error handling in C++ programs using parallel algorithms and executors Mark Hoemmen $ ^{*} $ mhoemmen@stellarscience.com CppCon 2020 * hou'màn, or hœm'màn; he/him ## exception in task gets captured – Waiting on result throws passed-along exception ## • P0443 executors + P1897 asynchronous algorithms — Separate path for handling ancestor task’s uncaught exception Uncaught exception in task gets captured — Waiting on result throws passed-along exception • P0443 executors + P1897 asynchronous algorithms — Separate path for handling ancestor task’s uncaught exception

FROM SCRATCH ## I /O Objects Robert Leahy - rleahy@rleahy.ca ISO C++ Networking is blocked on Executors (P0443) Target for Networking is currently C++23 (P0592) Used standalone Asio (1.18.0) to prepare

for computing resources is not the same for those queries. It is better for Spark to release some executors when either the query is lightweight, or the SQL engine is being idled. ## Tuning You can specify and sharing. On the other hand, we need to enable Spark's DRA feature for the engines' executors' elastic scaling. ## The Basics of Dynamic Resource Allocation Spark provides a mechanism engine has a backlog of pending tasks, it can request executors via ExecutorAllocationManager. When the engine has executors that become idle, the executors are released, and the occupied resources are given

for computing resources is not the same for those queries. It is better for Spark to release some executors when either the query is lightweight, or the SQL engine is being idled. ## Tuning You can specify and sharing. On the other hand, we need to enable Spark's DRA feature for the engines' executors' elastic scaling. ## The Basics of Dynamic Resource Allocation Spark provides a mechanism engine has a backlog of pending tasks, it can request executors via ExecutorAllocationManager. When the engine has executors that become idle, the executors are released, and the occupied resources are given

executor take. |Name|Default|Meaning| |---|---|---| |spark.executor.instances|1|The number of executors for static allocation| |spark.executor.cores|1|The number of cores to use on each executor| |Name|Default|Meaning| for computing resources is not the same for those queries. It is better for Spark to release some executors when either the query is lightweight, or the SQL engine is being idled. ## Tuning You can specify isolation, and sharing. On the other hand, we need to enable Spark’s DRA feature for the engines’ executors’ elastic scaling. ##### 2.1.1. The Basics of Dynamic Resource Allocation Spark provides a mechanism

executor take. |Name|Default|Meaning| |---|---|---| |spark.executor.instances|1|The number of executors for static allocation| |spark.executor.cores|1|The number of cores to use on each executor| |Name|Default|Meaning| for computing resources is not the same for those queries. It is better for Spark to release some executors when either the query is lightweight, or the SQL engine is being idled. ## Tuning You can specify isolation, and sharing. On the other hand, we need to enable Spark’s DRA feature for the engines’ executors’ elastic scaling. #### 1.1. The Basics of Dynamic Resource Allocation Spark provides a mechanism

for computing resources is not the same for those queries. It is better for Spark to release some executors when either the query is lightweight, or the SQL engine is being idled. ## Tuning You can specify and sharing. On the other hand, we need to enable Spark's DRA feature for the engines' executors' elastic scaling. ## The Basics of Dynamic Resource Allocation Spark provides a mechanism to engine has a backlog of pending tasks, it can request executors via ExecutorAllocationManager. When the engine has executors that become idle, the executors are released, and the occupied resources are given

for computing resources is not the same for those queries. It is better for Spark to release some executors when either the query is lightweight, or the SQL engine is being idled. ## Tuning You can specify and sharing. On the other hand, we need to enable Spark's DRA feature for the engines' executors' elastic scaling. ## The Basics of Dynamic Resource Allocation Spark provides a mechanism to engine has a backlog of pending tasks, it can request executors via ExecutorAllocationManager. When the engine has executors that become idle, the executors are released, and the occupied resources are given

for computing resources is not the same for those queries. It is better for Spark to release some executors when either the query is lightweight, or the SQL engine is being idled. ## Tuning You can specify belong to the one and only SparkContext instance, including Classes/Classloaders, SparkConf, Driver/Executors, Hive Metastore Client, etc. But each session can still have its own SparkSession instance, which and sharing. On the other hand, we need to enable Spark's DRA feature for the engines' executors' elastic scaling. ## The Basics of Dynamic Resource Allocation Spark provides a mechanism