advanceClock和Flush间存在并发问题

[M6ZeroG]

timingwheel/timingwheel.go

Lines 147 to 149 in 54845bd

	b := elem.(*bucket)
	tw.advanceClock(b.Expiration())
	b.Flush(tw.addOrRun)

考虑tw.advanceClock(b.Expiration())和b.Flush(tw.addOrRun)之间的窗口时间内，新的Timer被哈希到b且b.expiration发生变化的情况：
1、新的Timer会在Flush的for循环中被reinsert到和之前同样的、正在Flush的bucket，新的Timer会被处理链表的逻辑处理掉
2、即使新的Time没有被处理掉依然在bucket中，Flush完成后bucket的expiration会被置为-1，影响后续时间轮的正常运转

[wtfstring]

您好！邮件已收到！我会尽快回复！！

[M6ZeroG]

仔细看了一下，1和2的情况不会发生，但是情况会更糟糕，实际上新的Timer在reinsert回bucket时就已将因为锁重入的问题死锁了

[ml-liu]

goroutine 539 [sync.Mutex.Lock, 2644 minutes]:
sync.runtime_SemacquireMutex(0x451d2b?, 0x60?, 0x2?)
/usr/local/go/src/runtime/sema.go:77 +0x25
sync.(*Mutex).lockSlow(0xc000c7df70)
/usr/local/go/src/sync/mutex.go:171 +0x15d
sync.(*Mutex).Lock(...)
/usr/local/go/src/sync/mutex.go:90
github.com/RussellLuo/timingwheel.(*bucket).Add(0xc000c7df68, 0xc00c5c74c0)
/go/pkg/mod/github.com/!russell!luo/timingwheel@v0.0.0-20220218152713-54845bda3108/bucket.go:84 +0x45
github.com/RussellLuo/timingwheel.(*TimingWheel).add(0xc000b374a0, 0x0?)
/go/pkg/mod/github.com/!russell!luo/timingwheel@v0.0.0-20220218152713-54845bda3108/timingwheel.go:75 +0xfc
github.com/RussellLuo/timingwheel.(*TimingWheel).addOrRun(0xc000d37754?, 0xc00c5c74c0)
/go/pkg/mod/github.com/!russell!luo/timingwheel@v0.0.0-20220218152713-54845bda3108/timingwheel.go:112 +0x18
github.com/RussellLuo/timingwheel.(*bucket).Flush(0xc000c7df68, 0xc000d37780)
/go/pkg/mod/github.com/!russell!luo/timingwheel@v0.0.0-20220218152713-54845bda3108/bucket.go:127 +0xc2
github.com/RussellLuo/timingwheel.(*TimingWheel).Start.func2()
/go/pkg/mod/github.com/!russell!luo/timingwheel@v0.0.0-20220218152713-54845bda3108/timingwheel.go:149 +0x6d
github.com/RussellLuo/timingwheel.(*waitGroupWrapper).Wrap.func1()
/go/pkg/mod/github.com/!russell!luo/timingwheel@v0.0.0-20220218152713-54845bda3108/utils.go:35 +0x22
created by github.com/RussellLuo/timingwheel.(*waitGroupWrapper).Wrap in goroutine 548
/go/pkg/mod/github.com/!russell!luo/timingwheel@v0.0.0-20220218152713-54845bda3108/utils.go:34 +0x73

确实生产环境出现了这个问题，Flush 一个Bucket的时候又将其中的timer添加导致了死锁

[wtfstring]

您好！邮件已收到！我会尽快回复！！