storage: add cluster setting for ingest splits

Previously, the ingest-time splitting feature was only
tunable through a code change. This change adds a cluster
setting to allow this feature to be turned on dynamically.

Fixes #115430.

Epic: none

Release note: None

docs/generated/settings/settings.html

@@ -261,6 +261,7 @@
 <tr><td><div id="setting-sql-txn-read-committed-isolation-enabled" class="anchored"><code>sql.txn.read_committed_isolation.enabled</code></div></td><td>boolean</td><td><code>false</code></td><td>set to true to allow transactions to use the READ COMMITTED isolation level if specified by BEGIN/SET commands</td><td>Serverless/Dedicated/Self-Hosted</td></tr>
 <tr><td><div id="setting-sql-txn-fingerprint-id-cache-capacity" class="anchored"><code>sql.txn_fingerprint_id_cache.capacity</code></div></td><td>integer</td><td><code>100</code></td><td>the maximum number of txn fingerprint IDs stored</td><td>Serverless/Dedicated/Self-Hosted</td></tr>
 <tr><td><div id="setting-storage-experimental-eventually-file-only-snapshots-enabled" class="anchored"><code>storage.experimental.eventually_file_only_snapshots.enabled</code></div></td><td>boolean</td><td><code>false</code></td><td>set to true to use eventually-file-only-snapshots even when kv.snapshot_receiver.excise.enabled is false</td><td>Dedicated/Self-Hosted</td></tr>
+<tr><td><div id="setting-storage-ingest-split-enabled" class="anchored"><code>storage.ingest_split.enabled</code></div></td><td>boolean</td><td><code>false</code></td><td>set to true to use ingest-time splitting to lower write-amplification (experimental)</td><td>Dedicated/Self-Hosted</td></tr>
 <tr><td><div id="setting-storage-max-sync-duration" class="anchored"><code>storage.max_sync_duration</code></div></td><td>duration</td><td><code>20s</code></td><td>maximum duration for disk operations; any operations that take longer than this setting trigger a warning log entry or process crash</td><td>Serverless/Dedicated/Self-Hosted (read-only)</td></tr>
 <tr><td><div id="setting-storage-max-sync-duration-fatal-enabled" class="anchored"><code>storage.max_sync_duration.fatal.enabled</code></div></td><td>boolean</td><td><code>true</code></td><td>if true, fatal the process when a disk operation exceeds storage.max_sync_duration</td><td>Serverless/Dedicated/Self-Hosted</td></tr>
 <tr><td><div id="setting-storage-value-blocks-enabled" class="anchored"><code>storage.value_blocks.enabled</code></div></td><td>boolean</td><td><code>true</code></td><td>set to true to enable writing of value blocks in sstables</td><td>Serverless/Dedicated/Self-Hosted</td></tr>

pkg/storage/pebble.go

@@ -126,6 +126,21 @@ var UseExciseForSnapshots = settings.RegisterBoolSetting(
 	settings.WithPublic,
 )
 
+// IngestSplitEnabled controls whether ingest-time splitting is enabled in
+// Pebble. This feature allows for existing sstables to be split into multiple
+// virtual sstables at ingest time if that allows for an ingestion sstable to go
+// into a lower level than it would otherwise be in. No keys are masked with
+// this split; it only happens if there are no keys in that existing sstable
+// in the span of the incoming sstable.
+var IngestSplitEnabled = settings.RegisterBoolSetting(
+	settings.SystemOnly,
+	"storage.ingest_split.enabled",
+	"set to true to use ingest-time splitting to lower write-amplification (experimental)",
+	util.ConstantWithMetamorphicTestBool(
+		"storage.ingest_split.enabled", false), /* defaultValue */
+	settings.WithPublic,
+)
+
 // IngestAsFlushable controls whether ingested sstables that overlap the
 // memtable may be lazily ingested: written to the WAL and enqueued in the list
 // of flushables (eg, memtables, large batches and now lazily-ingested
@@ -1075,6 +1090,9 @@ func NewPebble(ctx context.Context, cfg PebbleConfig) (p *Pebble, err error) {
 	// See https://github.com/cockroachdb/pebble/issues/3120
 	// TODO(travers): Re-enable, once the issues are resolved.
 	opts.Experimental.MultiLevelCompactionHeuristic = pebble.NoMultiLevel{}
+	opts.Experimental.IngestSplit = func() bool {
+		return IngestSplitEnabled.Get(&cfg.Settings.SV)
+	}
 
 	auxDir := opts.FS.PathJoin(cfg.Dir, base.AuxiliaryDir)
 	if err := opts.FS.MkdirAll(auxDir, 0755); err != nil {