Optimize shard index loading

On data sets with many series and potentially large series keys,
the cost of parsing the key and re-indexing can be high.

Loading the TSM keys into the index was being done repeatedly for
series that were already index by an earlier TSM file.  This was
wasted worked and slows down shard loading.

Parsing the key was also innefficient and allocated a new string
slice.  This was simplified to remove that allocation.

CHANGELOG.md

@@ -12,6 +12,7 @@
 ### Bugfixes
 
 - [#6604](https://github.com/influxdata/influxdb/pull/6604): Remove old cluster code
+- [#6618](https://github.com/influxdata/influxdb/pull/6618): Optimize shard loading
 
 ## v0.13.0 [2016-05-12]
 

tsdb/engine/tsm1/engine.go

@@ -379,6 +379,12 @@ func (e *Engine) readFileFromBackup(tr *tar.Reader, shardRelativePath string) er
 // database index and measurement fields
 func (e *Engine) addToIndexFromKey(shardID uint64, key string, fieldType influxql.DataType, index *tsdb.DatabaseIndex) error {
 	seriesKey, field := seriesAndFieldFromCompositeKey(key)
+	// Have we already indexed this series?
+	ss := index.Series(seriesKey)
+	if ss != nil {
+		return nil
+	}
+
 	measurement := tsdb.MeasurementFromSeriesKey(seriesKey)
 
 	m := index.CreateMeasurementIndexIfNotExists(measurement)
@@ -1230,9 +1236,10 @@ func tsmFieldTypeToInfluxQLDataType(typ byte) (influxql.DataType, error) {
 }
 
 func seriesAndFieldFromCompositeKey(key string) (string, string) {
-	parts := strings.Split(key, keyFieldSeparator)
-	if len(parts) != 0 {
-		return parts[0], strings.Join(parts[1:], keyFieldSeparator)
+	sep := strings.Index(key, keyFieldSeparator)
+	if sep == -1 {
+		// No field???
+		return key, ""
 	}
-	return parts[0], parts[1]
+	return key[:sep], key[sep+len(keyFieldSeparator):]
 }