compaction: elision-only compactions for tables with only range keys

Currently, a table is marked as eligible for elision-only compaction by
the `elisionOnlyAnnotator` under the following circumstances:

- the table's range deletion estimate is greater than or equal to 10% of
  the total table size, OR
- the number of deletions is greater than or equal to 10% of the table's
  total point key entries.

If a table contains only range keys, the second predicate is true (given
that `0 >= 0`). If the table contains a range key delete or a range key
unset, an elision-only compaction is possible (i.e. if there are no
spans underneath, or snapshots preventing the elision, etc.).

However, if the table contains just range key sets, the table is always
marked for elision, but an elision-only compaction has no effect. The
same elision-only compaction will continue to be scheduled, which can
tie up a compaction slot.

Update the elision-only compaction heuristics to only schedule such
compactions when there is at least one range deletion, range key
deletion, or range key unset. This requires storing an additional uint
on the table stats, which allows determining whether a table contains
only range key sets.

compaction_picker.go

@@ -1230,6 +1230,12 @@ func (a elisionOnlyAnnotator) Accumulate(f *fileMetadata, dst interface{}) (inte
 	if !f.Stats.Valid {
 		return dst, false
 	}
+	// Tables without range dels, range key dels or range key unsets are not
+	// eligible for consideration for elision. Note the number of range key DELs +
+	// UNSETs can be computed using the number of range key SETs.
+	if f.Stats.NumDeletions+(f.Stats.NumRangeKeys-f.Stats.NumRangeKeySets) == 0 {
+		return dst, true
+	}
 	// Bottommost files are large and not worthwhile to compact just
 	// to remove a few tombstones. Consider a file ineligible if its
 	// own range deletions delete less than 10% of its data and its

compaction_picker_test.go

@@ -1176,6 +1176,7 @@ func TestCompactionOutputFileSize(t *testing.T) {
 				}
 				m.Stats.Valid = true
 				m.Stats.RangeDeletionsBytesEstimate = uint64(v)
+				m.Stats.NumDeletions = 1 // At least one range del responsible for the deletion bytes.
 			}
 		}
 		m.SmallestSeqNum = m.Smallest.SeqNum()

compaction_test.go

@@ -2097,6 +2097,7 @@ func TestCompactionTombstones(t *testing.T) {
 							compactInfo = &info
 						},
 					},
+					FormatMajorVersion: FormatNewest,
 				}
 				var err error
 				d, err = runDBDefineCmd(td, opts)

data_test.go

@@ -881,6 +881,7 @@ func runTableStatsCmd(td *datadriven.TestData, d *DB) string {
 			fmt.Fprintf(&b, "num-entries: %d\n", f.Stats.NumEntries)
 			fmt.Fprintf(&b, "num-deletions: %d\n", f.Stats.NumDeletions)
 			fmt.Fprintf(&b, "num-range-keys: %d\n", f.Stats.NumRangeKeys)
+			fmt.Fprintf(&b, "num-range-key-sets: %d\n", f.Stats.NumRangeKeySets)
 			fmt.Fprintf(&b, "point-deletions-bytes-estimate: %d\n", f.Stats.PointDeletionsBytesEstimate)
 			fmt.Fprintf(&b, "range-deletions-bytes-estimate: %d\n", f.Stats.RangeDeletionsBytesEstimate)
 			return b.String()

internal/manifest/version.go

@@ -52,6 +52,11 @@ type TableStats struct {
 	NumDeletions uint64
 	// NumRangeKeys is the total number of range keys in the table.
 	NumRangeKeys uint64
+	// NumRangeKeySets is the total number of range key sets in the table.
+	// NOTE: this field is currently used only for calculating the sum of range
+	// key DELs + UNSETs in the table, without the need for separate fields (i.e.
+	// it can be computed via subtraction).
+	NumRangeKeySets uint64
 	// Estimate of the total disk space that may be dropped by this table's
 	// point deletions by compacting them.
 	PointDeletionsBytesEstimate uint64

table_stats.go

@@ -270,6 +270,7 @@ func (d *DB) loadTableStats(
 		// additional stats that may provide improved heuristics for compaction
 		// picking.
 		stats.NumRangeKeys = r.Properties.NumRangeKeys()
+		stats.NumRangeKeySets = r.Properties.NumRangeKeySets
 		return
 	})
 	if err != nil {
@@ -535,6 +536,7 @@ func maybeSetStatsFromProperties(meta *fileMetadata, props *sstable.Properties)
 		NumEntries:                  props.NumEntries,
 		NumDeletions:                props.NumDeletions,
 		NumRangeKeys:                props.NumRangeKeys(),
+		NumRangeKeySets:             props.NumRangeKeySets,
 		PointDeletionsBytesEstimate: pointEstimate,
 		RangeDeletionsBytesEstimate: 0,
 	}

testdata/compaction_delete_only_hints

@@ -250,6 +250,7 @@ wait-pending-table-stats
 num-entries: 2
 num-deletions: 1
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 26
 

testdata/compaction_tombstones

@@ -14,6 +14,7 @@ wait-pending-table-stats
 num-entries: 2
 num-deletions: 2
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 0
 
@@ -36,6 +37,7 @@ wait-pending-table-stats
 num-entries: 2
 num-deletions: 2
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 0
 
@@ -57,6 +59,7 @@ wait-pending-table-stats
 num-entries: 2
 num-deletions: 1
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 26
 
@@ -80,6 +83,7 @@ wait-pending-table-stats
 num-entries: 2
 num-deletions: 1
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 0
 
@@ -119,6 +123,7 @@ wait-pending-table-stats
 num-entries: 6
 num-deletions: 2
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 76
 
@@ -152,6 +157,7 @@ wait-pending-table-stats
 num-entries: 11
 num-deletions: 1
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 149
 range-deletions-bytes-estimate: 0
 
@@ -196,6 +202,7 @@ wait-pending-table-stats
 num-entries: 5
 num-deletions: 1
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 16488
 
@@ -233,6 +240,7 @@ wait-pending-table-stats
 num-entries: 3
 num-deletions: 3
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 13167
 range-deletions-bytes-estimate: 0
 
@@ -244,3 +252,130 @@ range-deletions-bytes-estimate: 0
 maybe-compact
 ----
 [JOB 100] compacted(default) L5 [000004] (794 B) + L6 [000006] (13 K) -> L6 [] (0 B), in 1.0s (2.0s total), output rate 0 B/s
+
+# A table with range keys is only eligible for elision if it contains a
+# RANGEKEYDEL or RANGEKEYUNSET.
+
+define
+L6
+  rangekey:a-b:{(#1,RANGEKEYDEL)}
+----
+6:
+  000004:[a#1,RANGEKEYDEL-b#72057594037927935,RANGEKEYDEL]
+
+wait-pending-table-stats
+000004
+----
+num-entries: 0
+num-deletions: 0
+num-range-keys: 1
+num-range-key-sets: 0
+point-deletions-bytes-estimate: 0
+range-deletions-bytes-estimate: 0
+
+maybe-compact
+----
+[JOB 100] compacted(elision-only) L6 [000004] (895 B) + L6 [] (0 B) -> L6 [] (0 B), in 1.0s (2.0s total), output rate 0 B/s
+
+version
+----
+
+define
+L6
+  rangekey:a-b:{(#1,RANGEKEYUNSET,@1)}
+----
+6:
+  000004:[a#1,RANGEKEYUNSET-b#72057594037927935,RANGEKEYUNSET]
+
+wait-pending-table-stats
+000004
+----
+num-entries: 0
+num-deletions: 0
+num-range-keys: 1
+num-range-key-sets: 0
+point-deletions-bytes-estimate: 0
+range-deletions-bytes-estimate: 0
+
+maybe-compact
+----
+[JOB 100] compacted(elision-only) L6 [000004] (899 B) + L6 [] (0 B) -> L6 [] (0 B), in 1.0s (2.0s total), output rate 0 B/s
+
+version
+----
+
+define
+L6
+  rangekey:a-b:{(#1,RANGEKEYSET,@1)}
+----
+6:
+  000004:[a#1,RANGEKEYSET-b#72057594037927935,RANGEKEYSET]
+
+wait-pending-table-stats
+000004
+----
+num-entries: 0
+num-deletions: 0
+num-range-keys: 1
+num-range-key-sets: 1
+point-deletions-bytes-estimate: 0
+range-deletions-bytes-estimate: 0
+
+maybe-compact
+----
+(none)
+
+version
+----
+6:
+  000004:[a#1,RANGEKEYSET-b#72057594037927935,RANGEKEYSET]
+
+# A slightly more complicated case than the one above. A table with three
+# separate range key spans, each corresponding to one of the range key types.
+# Snapshots initially prevent any elision.
+
+define snapshots=(25,75)
+L6
+  rangekey:a-b:{(#100,RANGEKEYDEL)}
+  rangekey:b-c:{(#50,RANGEKEYUNSET,@1)}
+  rangekey:c-d:{(#10,RANGEKEYSET,@1,foo)}
+----
+6:
+  000004:[a#100,RANGEKEYDEL-d#72057594037927935,RANGEKEYSET]
+
+wait-pending-table-stats
+000004
+----
+num-entries: 0
+num-deletions: 0
+num-range-keys: 3
+num-range-key-sets: 1
+point-deletions-bytes-estimate: 0
+range-deletions-bytes-estimate: 0
+
+# Snapshots prevent any elision-only compaction.
+maybe-compact
+----
+(none)
+
+# Closing the snapshot at 25, despite it being the earliest snapshot, does not
+# schedule an elision-only compaction, as there is still a snapshot open below
+# the highest seq num of the file (100).
+close-snapshot
+25
+----
+(none)
+
+# Closing the last remaining snapshot triggers an elision-only compaction, as
+# there is is nothing left keeping the RANGEKEYUNSET and RANGEKEYDEL around.
+close-snapshot
+75
+----
+[JOB 100] compacted(elision-only) L6 [000004] (941 B) + L6 [] (0 B) -> L6 [000005] (903 B), in 1.0s (2.0s total), output rate 903 B/s
+
+# The only span remaining in the table is the RANGEKEYSET. The RANGEKEYDEL and
+# RANGEKEYUNSET were elided.
+version
+----
+6:
+  000005:[c#10,RANGEKEYSET-d#72057594037927935,RANGEKEYSET]

testdata/ingest

@@ -75,6 +75,7 @@ wait-pending-table-stats
 num-entries: 2
 num-deletions: 0
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 0
 
@@ -348,6 +349,7 @@ wait-pending-table-stats
 num-entries: 2
 num-deletions: 2
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 1666
 

testdata/manual_compaction

@@ -87,6 +87,7 @@ wait-pending-table-stats
 num-entries: 1
 num-deletions: 1
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 1552
 
@@ -105,6 +106,7 @@ wait-pending-table-stats
 num-entries: 2
 num-deletions: 1
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 776
 

testdata/manual_compaction_set_with_del

@@ -87,6 +87,7 @@ wait-pending-table-stats
 num-entries: 1
 num-deletions: 1
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 1552
 
@@ -105,6 +106,7 @@ wait-pending-table-stats
 num-entries: 2
 num-deletions: 1
 num-range-keys: 0
+num-range-key-sets: 0
 point-deletions-bytes-estimate: 0
 range-deletions-bytes-estimate: 776