Multithreading 通过Akka应用程序中的上下文切换实现高CPU使用率
我正在维护和开发两个Akka Scala应用程序,它们与串行设备接口以收集传感器信息。两者之间的主要区别在于,一个(我的二氧化碳传感器应用程序)使用1%的CPU,而另一个(我的功率传感器应用程序)使用250%的CPU。Linux机器(Raspberry Pi 3)和我的Windows桌面PC上都是如此。代码方面的主要区别是CO2直接使用串行库(),而功率传感器应用程序通过一层中间件将串行库的输入/输出流转换为Akka源/接收器:Multithreading 通过Akka应用程序中的上下文切换实现高CPU使用率,multithreading,scala,akka,cpu-usage,context-switch,Multithreading,Scala,Akka,Cpu Usage,Context Switch,我正在维护和开发两个Akka Scala应用程序,它们与串行设备接口以收集传感器信息。两者之间的主要区别在于,一个(我的二氧化碳传感器应用程序)使用1%的CPU,而另一个(我的功率传感器应用程序)使用250%的CPU。Linux机器(Raspberry Pi 3)和我的Windows桌面PC上都是如此。代码方面的主要区别是CO2直接使用串行库(),而功率传感器应用程序通过一层中间件将串行库的输入/输出流转换为Akka源/接收器: val port = SerialPort.getCommPo
val port = SerialPort.getCommPort(comPort)
port.setBaudRate(baudRate)
port.setFlowControl(flowControl)
port.setComPortParameters(baudRate, dataBits, stopBits, parity)
port.setComPortTimeouts(timeoutMode, timeout, timeout)
val isOpen = port.openPort()
if(!isOpen) {
error(s"Port $comPort could not opened. Use the following documentation for troubleshooting: https://github.com/Fazecast/jSerialComm/wiki/Troubleshooting")
throw new Exception("Port could not be opened")
}
(reactive.streamSource(port.getInputStream), reactive.streamSink(port.getOutputStream))
当我看到如此高的CPU使用率时,我立即用分析器(VisualVM)对其进行了攻击,它告诉我以下内容:
在谷歌搜索Unsafe.park后,我找到了以下答案:-使用此信息,我检查了有无功率传感器应用程序的上下文切换量,结果非常清楚问题的根本原因:
pi@dex:~ $ vmstat 1
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
r b swpd free buff cache si so bi bo in cs us sy id wa st
10 0 32692 80144 71228 264356 0 0 0 5 7 8 38 5 55 2 0
1 0 32692 80176 71228 264356 0 0 0 76 12932 18856 59 6 35 0 0
1 0 32692 80208 71228 264356 0 0 0 0 14111 20570 60 8 32 0 0
1 0 32692 80208 71228 264356 0 0 0 0 13186 16095 65 6 29 0 0
1 0 32692 80176 71228 264356 0 0 0 0 14008 23449 56 6 38 0 0
3 0 32692 80208 71228 264356 0 0 0 0 13528 17783 65 6 29 0 0
1 0 32692 80208 71228 264356 0 0 0 28 12960 16588 63 6 31 0 0
pi@dex:~ $ vmstat 1
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
r b swpd free buff cache si so bi bo in cs us sy id wa st
1 0 32692 147320 71228 264332 0 0 0 5 7 8 38 5 55 2 0
0 0 32692 147296 71228 264332 0 0 0 84 963 1366 0 0 98 2 0
0 0 32692 147296 71228 264332 0 0 0 0 962 1347 1 0 99 0 0
0 0 32692 147296 71228 264332 0 0 0 0 947 1318 1 0 99 0 0
正如您所看到的,仅通过关闭我的应用程序,上下文切换的数量就减少了约12000次/秒。我继续检查哪些线程正在执行此操作,Akka似乎真的很想执行以下操作:
这里的评论和另一个问题都指向调整Akka的并行度设置。我在application.conf中添加了以下内容,但没有结果
akka {
log-config-on-start = "on"
actor{
default-dispatcher {
# Dispatcher is the name of the event-based dispatcher
type = Dispatcher
# What kind of ExecutionService to use
executor = "fork-join-executor"
# Configuration for the fork join pool
default-executor {
fallback = "fork-join-executor"
}
fork-join-executor {
# Min number of threads to cap factor-based parallelism number to
parallelism-min = 1
# Parallelism (threads) ... ceil(available processors * factor)
parallelism-factor = 1.0
# Max number of threads to cap factor-based parallelism number to
parallelism-max = 1
}
# Throughput defines the maximum number of messages to be
# processed per actor before the thread jumps to the next actor.
# Set to 1 for as fair as possible.
throughput = 1
}
}
stream{
default-blocking-io-dispatcher {
type = PinnedDispatcher
executor = "fork-join-executor"
throughput = 1
thread-pool-executor {
core-pool-size-min = 1
core-pool-size-factor = 1.0
core-pool-size-max = 1
}
fork-join-executor {
parallelism-min = 1
parallelism-factor = 1.0
parallelism-max = 1
}
}
}
}
这似乎可以提高CPU使用率(100%->65%),但CPU使用率仍然过高
更新21-11-'16
问题似乎在我的图表中。不运行图形时,CPU使用率会立即降至正常水平。图表如下所示:
val streamGraph = RunnableGraph.fromGraph(GraphDSL.create() { implicit builder =>
import GraphDSL.Implicits._
val responsePacketSource = serialSource
.via(Framing.delimiter(ByteString(frameDelimiter), maxFrameLength, allowTruncation = true))
.via(cleanPacket)
.via(printOutput("Received: ",debug(_)))
.via(byteStringToResponse)
val packetSink = pushSource
.via(throttle(throttle))
val zipRequestStickResponse = builder.add(Zip[RequestPacket, ResponsePacket])
val broadcastRequest = builder.add(Broadcast[RequestPacket](2))
val broadcastResponse = builder.add(Broadcast[ResponsePacket](2))
packetSink ~> broadcastRequest.in
broadcastRequest.out(0) ~> makePacket ~> printOutput("Sent: ",debug(_)) ~> serialSink
broadcastRequest.out(1) ~> zipRequestStickResponse.in0
responsePacketSource ~> broadcastResponse.in
broadcastResponse.out(0).filter(isStickAck) ~> zipRequestStickResponse.in1
broadcastResponse.out(1).filter(!isStickAck(_)).map (al => {
val e = completeRequest(al)
debug(s"Sinking: $e")
e
}) ~> Sink.ignore
zipRequestStickResponse.out.map { case(request, stickResponse) =>
debug(s"Mapping: request=$request, stickResponse=$stickResponse")
pendingPackets += stickResponse.sequenceNumber -> request
request.stickResponse trySuccess stickResponse
} ~> Sink.ignore
ClosedShape
})
streamGraph.run()
当从broadcastResponse中删除过滤器时,CPU使用率将降至正常水平。这让我相信zip永远不会发生,因此,图形进入了错误的状态 问题在于Fazecast的JSerialCom库有许多不同的超时模式
static final public int TIMEOUT\u NONBLOCKING=0x00000000;
静态最终公共整数超时\读取\半阻塞=0x00000001;
静态最终公共整数超时\写入\半阻塞=0x00000010;
静态最终公共整数超时\读取\阻塞=0x00000100;
静态最终公共整数超时\写入\阻塞=0x00001000;
静态最终公共整数超时\u扫描器=0x00010000;
使用非阻塞
read()
方法(TIMEOUT\u NONBLOCKING
)与Akka流的InputStreamPublisher
结合使用时,会导致非常高的CPU使用率。要防止这种情况,只需使用TIMEOUT\u READ\u SEMI\u BLOCKING
或TIMEOUT\u READ\u BLOCKING
减少默认调度程序和阻塞io调度程序的并行数。如果您的工作负载通常不受CPU约束,请考虑切换到线程池执行器。@ ViktorKlang编辑了我的问题,以反映我在调试调度器时没有重新配置默认阻塞IO的尝试。dispatcher@ViktorKlang更新-似乎有一些改进,但CPU使用率仍然非常高(65%)