Deadlock occurred during update

[tttttttts]

Describe the bug

I encountered a deadlock situation during the update process

To Reproduce

We can only confirm that there were a large number of concurrent reads and writes at that time, calling update and view

Expected behavior

My program holds its own lock, but when using the badger update function, it encounters a deadlock that cannot be executed. The version used is v3.2103.5, and this is the stack information at the time of the deadlock. I want to confirm if this issue has been fixed by a commit or if it is an unknown problem.

Screenshots

goroutine 16782802941 [semacquire, 27 minutes]:
sync.runtime_SemacquireMutex(0x2030f3?, 0xf3?, 0xc3cd2ed040?)
runtime/sema.go:71 +0x25
sync.(*Mutex).lockSlow(0xc00022c36c)
sync/mutex.go:162 +0x165
sync.(*Mutex).Lock(...)
sync/mutex.go:81
github.com/dgraph-io/badger/v3.(*Txn).commitAndSend(0xc128d46200)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:537 +0x96
github.com/dgraph-io/badger/v3.(*Txn).Commit(0xc128d46200)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:671 +0x8c
github.com/dgraph-io/badger/v3.(*DB).Update(0xf622e0?, 0xc0214e1df8)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:826 +0xce

goroutine 16843731325 [semacquire, 1 minutes]:
sync.runtime_SemacquireMutex(0xd92f2b?, 0xd8?, 0xc543cd9a40?)
runtime/sema.go:71 +0x25
sync.(*Mutex).lockSlow(0xc00022c364)
sync/mutex.go:162 +0x165
sync.(*Mutex).Lock(...)
sync/mutex.go:81
github.com/dgraph-io/badger/v3.(*oracle).readTs(0xc00022c360)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:94 +0x4f
github.com/dgraph-io/badger/v3.(*DB).newTransaction(0xc000667200, 0xe0?, 0x0)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:786 +0x110
github.com/dgraph-io/badger/v3.(*DB).NewTransaction(...)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:764
github.com/dgraph-io/badger/v3.(*DB).View(0x0?, 0xc543cd9b40)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:802 +0x6c

Environment

• OS: debian 10
• Language GO
• Version v3.2103.5

Additional context

Add any other context about the problem here.

[matthewmcneely]

Badger's current release is at 4.8. Have you tried with that version? Could you share code that reproduces the lock?

[tttttttts]

When the stack information is roughly locked on the commitAndSend function, calling the commit and view functions can help analyze what situations may cause the inability to obtain the internal lock of commitAndSend to be released, while other goroutines are waiting. At that time, there were about 1000+goroutines that could not continue to write data

1000+goroutines:
goroutine 16813727231 [semacquire, 1 minutes]:
sync.runtime_SemacquireMutex(0xd92f2b?, 0xd8?, 0xc2adbe5a40?)
runtime/sema.go:71 +0x25
sync.(*Mutex).lockSlow(0xc00022c364)
sync/mutex.go:162 +0x165
sync.(*Mutex).Lock(...)
sync/mutex.go:81
github.com/dgraph-io/badger/v3.(*oracle).readTs(0xc00022c360)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:94 +0x4f
github.com/dgraph-io/badger/v3.(*DB).newTransaction(0xc000667200, 0xc0?, 0x0)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:786 +0x110
github.com/dgraph-io/badger/v3.(*DB).NewTransaction(...)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:764
github.com/dgraph-io/badger/v3.(*DB).View(0x0?, 0xc2adbe5b40)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:802 +0x6c

1000+goroutines:
goroutine 16782802941 [semacquire, 27 minutes]:
sync.runtime_SemacquireMutex(0x2030f3?, 0xf3?, 0xc3cd2ed040?)
runtime/sema.go:71 +0x25
sync.(*Mutex).lockSlow(0xc00022c36c)
sync/mutex.go:162 +0x165
sync.(*Mutex).Lock(...)
sync/mutex.go:81
github.com/dgraph-io/badger/v3.(*Txn).commitAndSend(0xc128d46200)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:537 +0x96
github.com/dgraph-io/badger/v3.(*Txn).Commit(0xc128d46200)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:671 +0x8c
github.com/dgraph-io/badger/v3.(*DB).Update(0xf622e0?, 0xc0214e1df8)
github.com/dgraph-io/badger/v3@v3.2103.5/txn.go:826 +0xce

I also tried to reproduce it, but it was unsuccessful， Have similar issues been fixed after version v3.2103.5