db: verify ingest sst bounds are suffixless or adjacent

[itsbilal]

We now allow sstables to have range key bounds that have suffixes in them, during ingestion. This is to allow for Cockroach-side defragmentation of range keys. This change ensures that such suffixed range keys are adjacent and defrag cleanly with rangekeys in other ingested sstables, to not cause panics later on in the read path.

Fixes #3829.

[cockroach-teamcity]

This change is 

[itsbilal]

We might need to figure out a way to exclude disaggregated storage sstables from these assertions. Rangekeydels don't play well with the current logic, and are tripping up the metamorphic tests.

[RaduBerinde]

Reviewable status: 0 of 5 files reviewed, 2 unresolved discussions (waiting on @itsbilal and @sumeerbhola)

---

ingest.go line 245 at r1 (raw file):

// ingestLoad1 creates the FileMetadata for one file. This file will be owned
// by this store.
//

We should move the checking logic to a separate struct

type rangeKeyCheck struct {
  lastRangeKeySpan keyspan.Span
}

func (c *rangeKeyCheck) Check(span *keyspan.Span) error { .. }

We would have a checkNoSuffixGap constructor, or use rangeKeyCheck{} when we don't want any checks.

But I wonder if there is a better way than doing this here. Couldn't we run this check on the metas after we do ingestLoad1? I'm assuming the SmallestRangeKey and LargestRangeKey in each meta would be enough to check for gaps?

---

ingest.go line 258 at r1 (raw file):

	lastRangeKey keyspan.Span,
	disableRangeKeyChecks bool,
) (*fileMetadata, keyspan.Span, error) {

The Span return looks very random, we should at least name it lastRangeKeySpan

[itsbilal]

TFTR!

Reviewable status: 0 of 5 files reviewed, all discussions resolved (waiting on @sumeerbhola)

---

ingest.go line 245 at r1 (raw file):

Previously, RaduBerinde wrote…

We should move the checking logic to a separate struct

type rangeKeyCheck struct {
  lastRangeKeySpan keyspan.Span
}

func (c *rangeKeyCheck) Check(span *keyspan.Span) error { .. }

We would have a checkNoSuffixGap constructor, or use rangeKeyCheck{} when we don't want any checks.

But I wonder if there is a better way than doing this here. Couldn't we run this check on the metas after we do ingestLoad1? I'm assuming the SmallestRangeKey and LargestRangeKey in each meta would be enough to check for gaps?

We can't look at just the metas, unfortunately. We need to ensure the RangeKeySets and Unsets in the []Key inside the bound spans are equivalent, otherwise we'd have to surface two different fragmented spans and one of those could have a suffix.

Did the struct change though

---

ingest.go line 258 at r1 (raw file):

Previously, RaduBerinde wrote…

The Span return looks very random, we should at least name it lastRangeKeySpan

Done.