collector: fix potential forgetting unlock in Persister.Inject

[BurtonQin]

Description
This patch fixes a potential forgetting unlock bug in Persister.Inject, which may lead to a deadlock.
In src/mongoshake/collector/persister.go,
Function Inject calls p.diskQueueMutex.Lock() but forgets p.diskQueueMutex.Unlock() on a specific branch.

MongoShake/src/mongoshake/collector/persister.go

Lines 157 to 168 in 313fc4e

	p.diskQueueMutex.Lock()
	if p.DiskQueue != nil { // double check
	// should send to diskQueue
	atomic.AddUint64(&p.diskWriteCount, 1)
	if err := p.DiskQueue.Put(input); err != nil {
	LOG.Crashf("persister inject replset[%v] put oplog to disk queue failed[%v]",
	p.replset, err)
	}
	} else {
	// should send to pending queue
	p.PushToPendingQueue(input)
	}

Inject is called by next in syncer.go, and next is called multiple times by poll.
So it is possible that p.disQueueMutex.Lock() is acquired twice, leading to a deadlock.

MongoShake/src/mongoshake/collector/syncer.go

Line 486 in 313fc4e

sync.persister.Inject(log)

MongoShake/src/mongoshake/collector/syncer.go

Lines 443 to 451 in 313fc4e

	for quorum.IsMaster() {
	// limit the qps if enabled
	if sync.qos.Limit > 0 {
	sync.qos.FetchBucket()
	}
	// only get one
	sync.next()
	}

Fix
The fix is to add a defer p.diskQueueMutex.Unlock() right after p.diskQueueMutex.Lock(). This way Unlock is always called as long as Lock is called, even if there is a panic (LOG.Crashf).

Suggestions
One interesting thing is that there is a unit testing TestInject in src/mongoshake/collector/persister_test.go.
But it fails to detect this potential deadlock because of the limited configurations.
e.g.

MongoShake/src/mongoshake/collector/persister_test.go

Line 36 in 313fc4e

conf.Options.FullSyncReaderOplogStoreDisk = false

So p.enableDiskPersist will always be false and the Lock branch will never be taken.

MongoShake/src/mongoshake/collector/persister.go

Lines 52 to 53 in 313fc4e

	enableDiskPersist: conf.Options.SyncMode == utils.VarSyncModeAll &&
	conf.Options.FullSyncReaderOplogStoreDisk,

I suggest changing the configurations to cover the Lock branch.

Though unlikely, if leaving out Unlock is the intention of the developers (e.g., for misconfiguration detection), I suggest adding error messages and doc to better explain this.

[CLAassistant]

All committers have signed the CLA.

[vinllen]

Thanks for your PR. This PR will be merged later when I want to open the disk persistent feature.