storage: don't sync with reads on lease extension

[andreimatei]

tschottdorf has a note in the code about why new leases need to
sometimes block reads while the lease is applied - you don't want reads
in between setting the new lease and updating the timestamp cache's
low-water mark. It seems safe to me to not do this in case of a lease
extension, since in that case we're not updating the aforementioned
low-water mark.
Besides hopefully being a perf optimization, this allows tests to do
lease extensions while a read is blocked, which I need for #8310.

---

This change is

[andreimatei]

cc @tschottdorf

[bdarnell]

---

Review status: 0 of 2 files reviewed at latest revision, 1 unresolved discussion, some commit checks failed.

---

storage/replica_command.go, line 1664 [r1] (raw file):

  ms *enginepb.MVCCStats,
  lease roachpb.Lease,
  isTransfer bool,

The name isTransfer suggests that this is only true for TransferLease, but that's not how it's used. Reverse this and call it isExtension?

---

Comments from Reviewable

[andreimatei]

Review status: 0 of 2 files reviewed at latest revision, 1 unresolved discussion, some commit checks pending.

---

storage/replica_command.go, line 1664 [r1] (raw file):

Previously, bdarnell (Ben Darnell) wrote…

The name isTransfer suggests that this is only true for TransferLease, but that's not how it's used. Reverse this and call it isExtension?

I didn't use `isExtension` because, at least to me, an "extension" sounds like it applies just when you have an active lease (as opposed to when the most recent lease is an expired one). But that's not how the word is used in the source, at lease in `RequestLease`, so ok, switching.

---

Comments from Reviewable