vfs: make Unwrap part of FS, fix metamorphic restart op

[RaduBerinde]

The metamorphic restart op never resolved the "base" MemFS because
diskHealthCheckingFS does not implement Unwrap. The code path
where we discard writes done by Close was inactive.

This change makes Unwrap part of FS and implements it correctly in
diskHealthCheckingFS.

In the metamorphic test, we now support the restartDB operation even
when the FS is not strict. When it is strict, we discard unsynced data
prior to Close. We no longer silently tolerate Root returning
something other than *MemFS in strictFS mode.

[cockroach-teamcity]

This change is 

[sumeerbhola]

Reviewed 6 of 6 files at r1, all commit messages.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @jbowens and @RaduBerinde)

---

metamorphic/test.go line 276 at r1 (raw file):

	var memFS *vfs.MemFS
	if !t.testOpts.strictFS {

Shouldn't this say if t.testOpts.strictFS?
I am surprised this did not panic when calling SetIgnoreSyncs because of it not being a strict MemFS.

Oh, this "worked" but is actually broken because we return earlier in this function due to

if !t.testOpts.strictFS {
		return nil
	}

And because of this new if-block we will never do ignore syncs.

[RaduBerinde]

Reviewable status: 5 of 6 files reviewed, 1 unresolved discussion (waiting on @jbowens and @sumeerbhola)

---

metamorphic/test.go line 276 at r1 (raw file):

Previously, sumeerbhola wrote…

Shouldn't this say if t.testOpts.strictFS?
I am surprised this did not panic when calling SetIgnoreSyncs because of it not being a strict MemFS.

Oh, this "worked" but is actually broken because we return earlier in this function due to

if !t.testOpts.strictFS {
		return nil
	}

And because of this new if-block we will never do ignore syncs.

faceslap Fixed it.. but now I get meta failures.. I'll have to investigate.

[RaduBerinde]

Reviewable status: 5 of 7 files reviewed, 1 unresolved discussion (waiting on @jbowens and @sumeerbhola)

---

metamorphic/test.go line 276 at r1 (raw file):

Previously, RaduBerinde wrote…

faceslap Fixed it.. but now I get meta failures.. I'll have to investigate.

Oh, I see. Without strictFS, we are using NoSync for write options, so we can't really restart the database and expect a deterministic outcome.

Changing it back to only run this op in strictFS mode. I'm still running into some failures that I'm investigating.

[sumeerbhola]

Reviewed 1 of 2 files at r2, all commit messages.
Reviewable status: 6 of 7 files reviewed, 1 unresolved discussion (waiting on @jbowens)

[RaduBerinde]

Reviewable status: 4 of 7 files reviewed, 1 unresolved discussion (waiting on @jbowens and @sumeerbhola)

---

metamorphic/test.go line 276 at r1 (raw file):

Previously, RaduBerinde wrote…

Oh, I see. Without strictFS, we are using NoSync for write options, so we can't really restart the database and expect a deterministic outcome.

Changing it back to only run this op in strictFS mode. I'm still running into some failures that I'm investigating.

Added a couple more exceptions and a fix for write-after-EOF in MemFS, seems to work now.

[jbowens]

Reviewed 4 of 6 files at r1, 3 of 3 files at r3, all commit messages.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @sumeerbhola)

[RaduBerinde]

TFTR!