The main thread anr exception is caused by too many binder development threads

Understand Ctrip ANR front , Let's get to know binder The previous life and present life of thread pool .

stay android In the system , adopt binder Conduct IPC when , The server always plays some Binder Threads To respond to client requests . This involves passing BInder Thread pool open up binder Threads .

Those Binder How to create threads , How to manage ？

stay APP Process creation or AIDL When the service process is created ,AMS Will notify Zygote process fork One APP process , stay Zygote Initialize the APP Process time , Will be called to Native Layer of app_main.cpp Medium onZygoteInit().

stay frameworks/base/cmds/app_process/app_main.cpp in

  virtual void onZygoteInit()
  {
      sp<ProcessState> proc = ProcessState::self();
      ALOGV("App process: starting thread pool.\n");
      proc->startThreadPool(); // Open thread pool 
  }

about Binder Number of thread pools , Creating for the current process ProcessState When , adopt mMaxThread Into Binder Drive the maximum that can be created Binder Number of threads , The default is 15 individual .

The first of this variable   stay The first definition is dead .

native/libs/binder/ProcessState.cpp

#define DEFAULT_MAX_BINDER_THREADS 15 // Thread pool   The largest number 

binder The maximum number of threads is 15 individual .

Ctrip actual combat case analysis .

Let's take a look at this log ：

 

Obviously, I was doing Binder signal communication ,ANR It happened in transcatNative Function , transcatNative The function of the initiator of the functional client .

Illustrate this anr Medium   The client may be waiting Binder End to end response , We know Binder Communication is blocking and waiting for response by default for the caller , Only when there is a return result can the execution continue .

There are two situations ：

1. When the server calls a method   There's a blockage , Cause the client to hang .

2. Binder Thread pool exceeded 15 individual , As a result, the allocation cannot continue , Client pending .

Through analysis happen anr When  cpu The state of , Sure, but binder The number of threads is more than 15 individual , That is, it can be determined to be binder Thread pools cause .

Why? Binder Thread pools can cause ANR Well ？

Let's imagine this situation ,A The process initiates the... At the same time 16 individual binder After the request , Because there is no C The process can handle , Because the thread pool is full at this time .

The client initiated the 16 individual binder Request to be in the blocking state at this time .

 

At a deeper level It is other processes that give systemserver send out Binder Messages are blocked , And there is no opposite end binder Thread . As a result, the caller needs to wait for the result , It's caused ANR.

So the focus of the problem , stay binder Thread allocation completed , No new thread assignments , Cause the caller to hang on the main thread .

There will be Binder cause ANR Or other abnormal conditions ？

The answer is yes .

• Activity To transfer data Pass on 1M when , Caused an anomaly

• Activity When asynchronous   Pass on 512k when Caused an anomaly

• Binder Server Server side Virtualization of complex objects results in ANR

 

from ： Ctrip ANR Optimization Practice - Binder Too many development threads lead to the main thread ANR abnormal