Set write batch readTs to the readMark.DoneUntil() value

[jarifibrahim]

Every Transaction stores the latest value of readTs it is aware of. When the transaction is discarded (which happens even when we commit), the global value of readMark is updated.

badger/txn.go

Lines 501 to 503 in 1fcc96e

	if !txn.db.orc.isManaged {
	txn.db.orc.readMark.Done(txn.readTs)
	}

Previously, the readTs of transaction inside the write batch struct was set to 0. So the global value of readMark would also be set to 0 (unless someone ran a transaction after using write batch).
Due to the 0 value of the global readMark, the compaction algorithm would skip all the values which were inserted in the write batch call.

badger/levels.go

Lines 480 to 484 in 1fcc96e

	// Pick a discard ts, so we can discard versions below this ts. We should
	// never discard any versions starting from above this timestamp, because
	// that would affect the snapshot view guarantee provided by transactions.
	discardTs := s.kv.orc.discardAtOrBelow()

and

badger/txn.go

Lines 138 to 145 in 1fcc96e

	func (o *oracle) discardAtOrBelow() uint64 {
	if o.isManaged {
	o.Lock()
	defer o.Unlock()
	return o.discardTs
	}
	return o.readMark.DoneUntil()
	}

The o.readMark.DoneUntil() call would always return 0 and so the compaction wouldn't compact the newer values.

With this commit, the compaction algorithm works as expected with key-values inserted via Transaction API or via the Write Batch API.

Fixes #767

---

This change is 

[manishrjain]

Reviewable status: 0 of 1 files reviewed, 1 unresolved discussion (waiting on @ashish-goswami, @jarifibrahim, @mangalaman93, and @manishrjain)

---

batch.go, line 138 at r1 (raw file):

	wb.txn.CommitWith(wb.callback)
	wb.txn = wb.db.newTransaction(true, true)
	// See comment about readTs in NewWriteBatch. This readTs is used only to detect conflicts in a transaction.

100 chars.

Remove that comment about "this readts ..."

[mangalaman93]

@jarifibrahim will it now affect the performance as @manishrjain mentioned? How are we solving that? We should probably add comments around it in the PR.

[jarifibrahim]

@jarifibrahim will it now affect the performance as @manishrjain mentioned? How are we solving that? We should probably add comments around it in the PR.

@mangalaman93 No, this change shouldn't affect performance. My previous changes would've affected the performance but this one shouldn't.

[jarifibrahim]

Reviewable status: 0 of 1 files reviewed, 1 unresolved discussion (waiting on @ashish-goswami, @mangalaman93, and @manishrjain)

---

batch.go, line 138 at r1 (raw file):

Previously, manishrjain (Manish R Jain) wrote…

100 chars.

Remove that comment about "this readts ..."

Done.

[ashish-goswami]

@jarifibrahim can you check why unit tests are failing on Teamcity?