作者：阿里云-移动云-大前端团队
前言
如果 app 连续 crash 两次无法启动，用户往往会选择卸载。
连续启动 crash 应该是 crash 类型中最严重的一类，该问题常常与数据库操作有关，比如：数据库损坏、服务端返回数据错误，存入数据库，app 读取时产生数组越界、找不到方法。
那么除了热修复，能否“自修复”该问题呢？
在微信读书团队发布的《iOS 启动连续闪退保护方案》 一文中，给出了连续启动crash的自修复技术的思路讲解，并在GitHub上给出了技术实现，并开源了 GYBootingProtection。方案思路很好，很轻量级。
实现原理
在微信读书团队给出的文章中已经有比较详细的阐述，在此不做赘述，实现的流程图如下所示：
但有个实现上可以优化下，可以降低50%以上误报机率，监听用户手动划掉 APP 这个事件，其中一些特定场景，是可以获取的。另外在这里也给出对其 API 设计的建议。最后给出优化后的实现。
优化：降低50%以上误报机率
用户主动 kill 掉 APP 分为两种情况：
- App在前台时用户手动划掉APP的时候
- APP在后台时划掉APP
第一种场景更为常见，可以通过监听 UIApplicationWillTerminateNotification 来捕获该动作，捕获后恢复计数。第二种情况，无法监听到。但也足以降低 50% 以上的误报机率。
对原有API设计的几点优化意见
1. 机制状态应当用枚举来做为API透出
该机制当前所处的状态，比如：NeedFix 、isFixing，建议用枚举来做为API透出。比如：
- APP 启动正常
- 正在检测是否会在特定时间内是否会 Crash，注意：检测状态下“连续启动崩溃计数”个数小于或等于上限值
- APP 出现连续启动 Crash，需要采取修复措施
- APP 出现连续启动 Crash，正在修复中
2. 关键数值应当做为初始化参数供用户设置
- 当前启动Crash的状态
- 达到需要执行上报操作的“连续启动崩溃计数”个数。
- 达到需要执行修复操作的“连续启动崩溃计数”个数。
- APP 启动后经过多少秒，可以将“连续启动崩溃计数”清零
3. 修复、上报逻辑应当支持用户异步操作
reportBlock 上报逻辑，repairtBlock 修复逻辑
比如：
typedef void (^BoolCompletionHandler)(BOOL succeeded, NSError *error);
typedef void (^RepairBlock)(ABSBoolCompletionHandler completionHandler);用户执行 BoolCompletionHandler 后即可知道是否执行完毕，并且支持异步操作。
异步操作带来的问题，可以通过前面提到的枚举API来实时监测状态，来决定各种其他操作。
什么时候会出现该异常？
连续启动 crash 自修复技术实现与原理解析
下面给出优化后的代码实现：
//
// CYLBootingProtection.h
//
//
// Created by ChenYilong on 18/01/10.
// Copyright © 2018年 ChenYilong. All rights reserved.
//
#import <Foundation/Foundation.h>
typedef void (^ABSBoolCompletionHandler)(BOOL succeeded, NSError *error);
typedef void (^ABSRepairBlock)(ABSBoolCompletionHandler completionHandler);
typedef void (^ABSReportBlock)(NSUInteger crashCounts);
typedef NS_ENUM(NSInteger, BootingProtectionStatus) {
BootingProtectionStatusNormal, /**< APP 启动正常 */
BootingProtectionStatusNormalChecking, /**< 正在检测是否会在特定时间内是否会 Crash，注意：检测状态下“连续启动崩溃计数”个数小于或等于上限值 */
BootingProtectionStatusNeedFix, /**< APP 出现连续启动 Crash，需要采取修复措施 */
BootingProtectionStatusFixing, /**< APP 出现连续启动 Crash，正在修复中... */
};
/**
* 启动连续 crash 保护。
* 启动后 `_crashOnLaunchTimeIntervalThreshold` 秒内 crash，反复超过 `_continuousCrashOnLaunchNeedToReport` 次则上报日志，超过 `_continuousCrashOnLaunchNeedToFix` 则启动修复操作。
*/
@interface CYLBootingProtection : NSObject
/**
* 启动连续 crash 保护方法。
* 前置条件：在 App 启动时注册 crash 处理函数，在 crash 时调用[CYLBootingProtection addCrashCountIfNeeded]。
* 启动后一定时间内（`crashOnLaunchTimeIntervalThreshold`秒内）crash，反复超过一定次数（`continuousCrashOnLaunchNeedToReport`次）则上报日志，超过一定次数（`continuousCrashOnLaunchNeedToFix`次）则启动修复程序；在一定时间内（`crashOnLaunchTimeIntervalThreshold`秒） 秒后若没有 crash 将“连续启动崩溃计数”计数置零。
`reportBlock` 上报逻辑，
`repairtBlock` 修复逻辑，完成后执行 `[self setCrashCount:0]`
*/
- (void)launchContinuousCrashProtect;
/*!
* 当前启动Crash的状态
*/
@property (nonatomic, assign, readonly) BootingProtectionStatus bootingProtectionStatus;
/*!
* 达到需要执行上报操作的“连续启动崩溃计数”个数。
*/
@property (nonatomic, assign, readonly) NSUInteger continuousCrashOnLaunchNeedToReport;
/*!
* 达到需要执行修复操作的“连续启动崩溃计数”个数。
*/
@property (nonatomic, assign, readonly) NSUInteger continuousCrashOnLaunchNeedToFix;
/*!
* APP 启动后经过多少秒，可以将“连续启动崩溃计数”清零
*/
@property (nonatomic, assign, readonly) NSTimeInterval crashOnLaunchTimeIntervalThreshold;
/*!
* 借助 context 可以让多个模块注册事件，并且事件 block 能独立执行，互不干扰。
*/
@property (nonatomic, copy, readonly) NSString *context;
/*!
* @details 启动后kCrashOnLaunchTimeIntervalThreshold秒内crash，反复超过continuousCrashOnLaunchNeedToReport次则上报日志，超过continuousCrashOnLaunchNeedToFix则启动修复程序；当所有操作完成后，执行 completion。在 crashOnLaunchTimeIntervalThreshold 秒后若没有 crash 将 kContinuousCrashOnLaunchCounterKey 计数置零。
* @param context 借助 context 可以让多个模块注册事件，并且事件 block 能独立执行，互不干扰。
*/
- (instancetype)initWithContinuousCrashOnLaunchNeedToReport:(NSUInteger)continuousCrashOnLaunchNeedToReport
continuousCrashOnLaunchNeedToFix:(NSUInteger)continuousCrashOnLaunchNeedToFix
crashOnLaunchTimeIntervalThreshold:(NSTimeInterval)crashOnLaunchTimeIntervalThreshold
context:(NSString *)context;
/*!
* 当前“连续启动崩溃“的状态
*/
+ (BootingProtectionStatus)bootingProtectionStatusWithContext:(NSString *)context continuousCrashOnLaunchNeedToFix:(NSUInteger)continuousCrashOnLaunchNeedToFix;
/*!
* 设置上报逻辑，参数 crashCounts 为启动连续 crash 次数
*/
- (void)setReportBlock:(ABSReportBlock)reportBlock;
/*!
* 设置修复逻辑
*/
- (void)setRepairBlock:(ABSRepairBlock)repairtBlock;
+ (void)setLogger:(void (^)(NSString *))logger;
@end
//
// CYLBootingProtection.m
//
//
// Created by ChenYilong on 18/01/10.
// Copyright © 2018年 ChenYilong. All rights reserved.
//
#import "CYLBootingProtection.h"
#import <UIKit/UIKit.h>
static dispatch_queue_t _exceptionOperationQueue = 0;
void (^Logger)(NSString *log);
@interface CYLBootingProtection ()
@property (nonatomic, assign) NSUInteger continuousCrashOnLaunchNeedToReport;
@property (nonatomic, assign) NSUInteger continuousCrashOnLaunchNeedToFix;
@property (nonatomic, assign) NSTimeInterval crashOnLaunchTimeIntervalThreshold;
@property (nonatomic, copy) NSString *context;
@property (nonatomic, copy) ABSReportBlock reportBlock;
@property (nonatomic, copy) ABSRepairBlock repairBlock;
/*!
* 设置“连续启动崩溃计数”个数
*/
- (void)setCrashCount:(NSInteger)count;
/*!
* 设置“连续启动崩溃计数”个数
*/
+ (void)setCrashCount:(NSUInteger)count context:(NSString *)context;
/*!
* “连续启动崩溃计数”个数
*/
- (NSUInteger)crashCount;
/*!
* “连续启动崩溃计数”个数
*/
+ (NSUInteger)crashCountWithContext:(NSString *)context;
@end
@implementation CYLBootingProtection
+ (void)initialize {
static dispatch_once_t onceToken;
dispatch_once(&onceToken, ^{
_exceptionOperationQueue = dispatch_queue_create("com.ChenYilong.CYLBootingProtection.fileCacheQueue", DISPATCH_QUEUE_SERIAL);
});
}
- (instancetype)initWithContinuousCrashOnLaunchNeedToReport:(NSUInteger)continuousCrashOnLaunchNeedToReport
continuousCrashOnLaunchNeedToFix:(NSUInteger)continuousCrashOnLaunchNeedToFix
crashOnLaunchTimeIntervalThreshold:(NSTimeInterval)crashOnLaunchTimeIntervalThreshold
context:(NSString *)context {
if (!(self = [super init])) {
return nil;
}
_continuousCrashOnLaunchNeedToReport = continuousCrashOnLaunchNeedToReport;
_continuousCrashOnLaunchNeedToFix = continuousCrashOnLaunchNeedToFix;
_crashOnLaunchTimeIntervalThreshold = crashOnLaunchTimeIntervalThreshold;
_context = [context copy];
[[NSNotificationCenter defaultCenter] addObserver:self
selector:@selector(applicationWillTerminate:)
name:UIApplicationWillTerminateNotification
object:[UIApplication sharedApplication]];
return self;
}
/*!
* App在前台时用户手动划掉APP的时候，不计入检测。
* 但是APP在后台时划掉APP，无法检测出来。
* 见：https://*.com/a/35041565/3395008
*/
- (void)applicationWillTerminate:(NSNotification *)note {
BOOL isNormalChecking = [self isNormalChecking];
if (isNormalChecking) {
[self decreaseCrashCount];
}
}
- (void)dealloc {
[[NSNotificationCenter defaultCenter] removeObserver:self];
}
/*
支持同步修复、异步修复，两种修复方式
- 异步修复，不卡顿主UI，但有修复未完成就被再次触发crash、或者用户kill掉的可能。需要用户手动根据修复状态，来选择性地进行操作，应该有回掉。
- 同步修复，最简单直观，在主线程删除或者下载修复包。
*/
- (void)launchContinuousCrashProtect {
NSAssert(_repairBlock, @"_repairBlock is nil!");
[[self class] Logger:@"CYLBootingProtection: Launch continuous crash report"];
[self resetBootingProtectionStatus];
NSUInteger launchCrashes = [self crashCount];
// 上报
if (launchCrashes >= self.continuousCrashOnLaunchNeedToReport) {
NSString *logString = [NSString stringWithFormat:@"CYLBootingProtection: App has continuously crashed for %@ times. Now synchronize uploading crash report and begin fixing procedure.", @(launchCrashes)];
[[self class] Logger:logString];
if (_reportBlock) {
dispatch_async(dispatch_get_main_queue(),^{
_reportBlock(launchCrashes);
});
}
}
// 修复
if ([self isUpToBootingProtectionCount]) {
[[self class] Logger:@"need to repair"];
[self setIsFixing:YES];
if (_repairBlock) {
ABSBoolCompletionHandler completionHandler = ^(BOOL succeeded, NSError *__nullable error){
if (succeeded) {
[self resetCrashCount];
} else {
[[self class] Logger:error.description];
}
};
dispatch_async(dispatch_get_main_queue(),^{
_repairBlock(completionHandler);
});
}
} else {
[self increaseCrashCount:launchCrashes];
// 正常流程，无需修复
[[self class] Logger:@"need no repair"];
// 记录启动时刻，用于计算启动连续 crash
// 重置启动 crash 计数
dispatch_after(dispatch_time(DISPATCH_TIME_NOW, (int64_t)(self.crashOnLaunchTimeIntervalThreshold * NSEC_PER_SEC)), dispatch_get_main_queue(), ^(void){
// APP活过了阈值时间，重置崩溃计数
NSString *logString = [NSString stringWithFormat:@"CYLBootingProtection: long live the app ( more than %@ seconds ), now reset crash counts", @(self.crashOnLaunchTimeIntervalThreshold)];
[[self class] Logger:logString];
[self resetCrashCount];
});
}
}
//减少计数的时机：用户手动划掉APP
- (void)decreaseCrashCount {
NSUInteger oldCrashCount = [self crashCount];
[self decreaseCrashCountWithOldCrashCount:oldCrashCount];
}
- (void)decreaseCrashCountWithOldCrashCount:(NSUInteger)oldCrashCount {
dispatch_sync(_exceptionOperationQueue, ^{
if (oldCrashCount > 0) {
[self setCrashCount:oldCrashCount-1];
}
[self resetBootingProtectionStatus];
});
}
//重制计数的时机：修复完成、或者用户手动划掉APP
- (void)resetCrashCount {
[self setCrashCount:0];
[self resetBootingProtectionStatus];
}
//只在未达到计数上限时才会增加计数
- (void)increaseCrashCount:(NSUInteger)oldCrashCount {
dispatch_sync(_exceptionOperationQueue, ^{
[self setIsNormalChecking:YES];
[self setCrashCount:oldCrashCount+1];
});
}
- (void)resetBootingProtectionStatus {
[self setIsNormalChecking:NO];
[self setIsFixing:NO];
}
- (BootingProtectionStatus)bootingProtectionStatus {
return [[self class] bootingProtectionStatusWithContext:_context continuousCrashOnLaunchNeedToFix:_continuousCrashOnLaunchNeedToFix];
}
/*!
*
@attention 注意之所以要检查 `BootingProtectionStatusNormalChecking` 原因如下：
`-launchContinuousCrashProtect` 方法与 `-bootingProtectionStatus` 方法，如果 `-launchContinuousCrashProtect` 先执行，那么会造成如下问题：
假设n为上限，但crash（n-1）次，但是用 `-bootingProtectionStatus` 判断出来，当前已经处于n次了。原因如下：
crash（n-1）次，正常流程，计数+1，变成n次，
随后在检查 `-bootingProtectionStatus` 时，发现已经处于异常状态了，实际是正常状态。所以需要使用`BootingProtectionStatusNormalChecking` 来进行区分。
*/
+ (BootingProtectionStatus)bootingProtectionStatusWithContext:(NSString *)context continuousCrashOnLaunchNeedToFix:(NSUInteger)continuousCrashOnLaunchNeedToFix {
BOOL isNormalChecking = [self isNormalCheckingWithContext:context];
if (isNormalChecking) {
return BootingProtectionStatusNormalChecking;
}
BOOL isUpToBootingProtectionCount = [self isUpToBootingProtectionCountWithContext:context
continuousCrashOnLaunchNeedToFix:continuousCrashOnLaunchNeedToFix];
if (!isUpToBootingProtectionCount) {
return BootingProtectionStatusNormal;
}
BootingProtectionStatus type;
BOOL isFixingCrash = [self isFixingCrashWithContext:context];
if (isFixingCrash) {
type = BootingProtectionStatusFixing;
} else {
type = BootingProtectionStatusNeedFix;
}
return type;
}
- (NSUInteger)crashCount {
return [[self class] crashCountWithContext:_context];
}
- (void)setCrashCount:(NSInteger)count {
if (count >=0) {
[[self class] setCrashCount:count context:_context];
}
}
- (void)setIsFixing:(BOOL)isFixingCrash {
[[self class] setIsFixing:isFixingCrash context:_context];
}
/*!
* 是否正在修复
*/
- (BOOL)isFixingCrash {
return [[self class] isFixingCrashWithContext:_context];
}
- (void)setIsNormalChecking:(BOOL)isNormalChecking {
[[self class] setIsNormalChecking:isNormalChecking context:_context];
}
/*!
* 是否正在检查
*/
- (BOOL)isNormalChecking {
return [[self class] isNormalCheckingWithContext:_context];
}
+ (NSUInteger)crashCountWithContext:(NSString *)context {
NSString *continuousCrashOnLaunchCounterKey = [self continuousCrashOnLaunchCounterKeyWithContext:context];
NSUInteger crashCount = [[NSUserDefaults standardUserDefaults] integerForKey:continuousCrashOnLaunchCounterKey];
NSString *logString = [NSString stringWithFormat:@"crashCount:%@", @(crashCount)];
[[self class] Logger:logString];
return crashCount;
}
+ (void)setCrashCount:(NSUInteger)count context:(NSString *)context {
NSString *continuousCrashOnLaunchCounterKey = [self continuousCrashOnLaunchCounterKeyWithContext:context];
NSString *logString = [NSString stringWithFormat:@"setCrashCount:%@", @(count)];
[[self class] Logger:logString];
NSUserDefaults *defaults = [NSUserDefaults standardUserDefaults];
[defaults setInteger:count forKey:continuousCrashOnLaunchCounterKey];
[defaults synchronize];
}
+ (void)setIsFixing:(BOOL)isFixingCrash context:(NSString *)context {
NSString *continuousCrashFixingKey = [[self class] continuousCrashFixingKeyWithContext:context];
NSString *logString = [NSString stringWithFormat:@"setisFixingCrash:{%@}", @(isFixingCrash)];
[[self class] Logger:logString];
NSUserDefaults *defaults = [NSUserDefaults standardUserDefaults];
[defaults setBool:isFixingCrash forKey:continuousCrashFixingKey];
[defaults synchronize];
}
+ (BOOL)isFixingCrashWithContext:(NSString *)context {
NSString *continuousCrashFixingKey = [[self class] continuousCrashFixingKeyWithContext:context];
BOOL isFixingCrash = [[NSUserDefaults standardUserDefaults] boolForKey:continuousCrashFixingKey];
NSString *logString = [NSString stringWithFormat:@"isFixingCrash:%@", @(isFixingCrash)];
[[self class] Logger:logString];
return isFixingCrash;
}
+ (void)setIsNormalChecking:(BOOL)isNormalChecking context:(NSString *)context {
NSString *continuousCrashNormalCheckingKey = [[self class] continuousCrashNormalCheckingKeyWithContext:context];
NSString *logString = [NSString stringWithFormat:@"setIsNormalChecking:{%@}", @(isNormalChecking)];
[[self class] Logger:logString];
NSUserDefaults *defaults = [NSUserDefaults standardUserDefaults];
[defaults setBool:isNormalChecking forKey:continuousCrashNormalCheckingKey];
[defaults synchronize];
}
+ (BOOL)isNormalCheckingWithContext:(NSString *)context {
NSString *continuousCrashFixingKey = [[self class] continuousCrashNormalCheckingKeyWithContext:context];
BOOL isFixingCrash = [[NSUserDefaults standardUserDefaults] boolForKey:continuousCrashFixingKey];
NSString *logString = [NSString stringWithFormat:@"isIsNormalChecking:%@", @(isFixingCrash)];
[[self class] Logger:logString];
return isFixingCrash;
}
- (BOOL)isUpToBootingProtectionCount {
return [[self class] isUpToBootingProtectionCountWithContext:_context continuousCrashOnLaunchNeedToFix:_continuousCrashOnLaunchNeedToFix];
}
+ (BOOL)isUpToBootingProtectionCountWithContext:(NSString *)context continuousCrashOnLaunchNeedToFix:(NSUInteger)continuousCrashOnLaunchNeedToFix {
BOOL isUpToCount = [self crashCountWithContext:context] >= continuousCrashOnLaunchNeedToFix;
if (isUpToCount) {
return YES;
}
return NO;
}
- (void)setReportBlock:(ABSReportBlock)block {
_reportBlock = block;
}
- (void)setRepairBlock:(ABSRepairBlock)block {
_repairBlock = block;
}
/*!
* “连续启动崩溃计数”个数，对应的Key
* 默认为 "_CONTINUOUS_CRASH_COUNTER_KEY"
*/
+ (NSString *)continuousCrashOnLaunchCounterKeyWithContext:(NSString *)context {
BOOL isValid = [[self class] isValidString:context];
NSString *validContext = isValid ? context : @"";
NSString *continuousCrashOnLaunchCounterKey = [NSString stringWithFormat:@"%@_CONTINUOUS_CRASH_COUNTER_KEY", validContext];
return continuousCrashOnLaunchCounterKey;
}
/*!
* 是否正在修复记录，对应的Key
* 默认为 "_CONTINUOUS_CRASH_FIXING_KEY"
*/
+ (NSString *)continuousCrashFixingKeyWithContext:(NSString *)context {
BOOL isValid = [[self class] isValidString:context];
NSString *validContext = isValid ? context : @"";
NSString *continuousCrashFixingKey = [NSString stringWithFormat:@"%@_CONTINUOUS_CRASH_FIXING_KEY", validContext];
return continuousCrashFixingKey;
}
/*!
* 是否正在检查是否在特定时间内会Crash，对应的Key
* 默认为 "_CONTINUOUS_CRASH_CHECKING_KEY"
*/
+ (NSString *)continuousCrashNormalCheckingKeyWithContext:(NSString *)context {
BOOL isValid = [[self class] isValidString:context];
NSString *validContext = isValid ? context : @"";
NSString *continuousCrashFixingKey = [NSString stringWithFormat:@"%@_CONTINUOUS_CRASH_CHECKING_KEY", validContext];
return continuousCrashFixingKey;
}
#pragma mark -
#pragma mark - log and util Methods
+ (void)setLogger:(void (^)(NSString *))logger {
Logger = [logger copy];
}
+ (void)Logger:(NSString *)log {
if (Logger) Logger(log);
}
+ (BOOL)isValidString:(id)notValidString {
if (!notValidString) {
return NO;
}
if (![notValidString isKindOfClass:[NSString class]]) {
return NO;
}
NSInteger stringLength = 0;
@try {
stringLength = [notValidString length];
} @catch (NSException *exception) {}
if (stringLength == 0) {
return NO;
}
return YES;
}
@end
下面是相应的验证步骤：
等待15秒会有对应计数清零的操作日志输出：
2018-01-18 16:25:37.162980+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：CYLBootingProtection: Launch continuous crash report
2018-01-18 16:25:37.163140+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：setIsNormalChecking:{0}
2018-01-18 16:25:37.165738+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：setisFixingCrash:{0}
2018-01-18 16:25:37.166883+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：crashCount:0
2018-01-18 16:25:37.167102+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：crashCount:0
2018-01-18 16:25:37.167253+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：setIsNormalChecking:{1}
2018-01-18 16:25:37.167938+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：setCrashCount:1
2018-01-18 16:25:37.168806+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：need no repair
2018-01-18 16:25:52.225197+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：CYLBootingProtection: long live the app ( more than 15 seconds ), now reset crash counts
2018-01-18 16:25:52.225378+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：setCrashCount:0
2018-01-18 16:25:52.226234+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：setIsNormalChecking:{0}
2018-01-18 16:25:52.226595+0800 BootingProtection[89773:15553277] 类名与方法名：-[AppDelegate onBeforeBootingProtection]_block_invoke（在第45行），描述：setisFixingCrash:{0}