config: validate TSDB retention settings during config parsing

Move retention validation from tsdb/db.go into a TSDBRetentionConfig
UnmarshalYAML method so that invalid values are rejected at config
load/reload time rather than at apply time.

- Reject negative retention size values.
- Reject retention percentage values above 100.
- Simplify ApplyConfig to assign retention values unconditionally,
  enabling setting a value back to 0 to disable it.

Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>

config/config.go

@@ -1097,6 +1097,22 @@ type TSDBRetentionConfig struct {
 	Percentage uint `yaml:"percentage,omitempty"`
 }
 
+// UnmarshalYAML implements the yaml.Unmarshaler interface.
+func (t *TSDBRetentionConfig) UnmarshalYAML(unmarshal func(any) error) error {
+	*t = TSDBRetentionConfig{}
+	type plain TSDBRetentionConfig
+	if err := unmarshal((*plain)(t)); err != nil {
+		return err
+	}
+	if t.Size < 0 {
+		return fmt.Errorf("'storage.tsdb.retention.size' must be greater than or equal to 0, got %v", t.Size)
+	}
+	if t.Percentage > 100 {
+		return fmt.Errorf("'storage.tsdb.retention.percentage' must be in the range [0, 100], got %v", t.Percentage)
+	}
+	return nil
+}
+
 // TSDBConfig configures runtime reloadable configuration options.
 type TSDBConfig struct {
 	// OutOfOrderTimeWindow sets how long back in time an out-of-order sample can be inserted

config/config_test.go

@@ -2626,6 +2626,22 @@ var expectedErrors = []struct {
 		filename: "stackit_endpoint.bad.yml",
 		errMsg:   "invalid endpoint",
 	},
+	{
+		filename: "tsdb_retention_time.bad.yml",
+		errMsg:   `not a valid duration string: "-1h"`,
+	},
+	{
+		filename: "tsdb_retention_size.bad.yml",
+		errMsg:   `'storage.tsdb.retention.size' must be greater than or equal to 0`,
+	},
+	{
+		filename: "tsdb_retention_percentage.bad.yml",
+		errMsg:   `'storage.tsdb.retention.percentage' must be in the range [0, 100]`,
+	},
+	{
+		filename: "tsdb_retention_percentage_negative.bad.yml",
+		errMsg:   "cannot unmarshal !!int `-1` into uint",
+	},
 }
 
 func TestBadConfigs(t *testing.T) {

config/testdata/tsdb_retention_percentage.bad.yml

@@ -0,0 +1,4 @@
+storage:
+  tsdb:
+    retention:
+      percentage: 101

config/testdata/tsdb_retention_percentage_negative.bad.yml

@@ -0,0 +1,4 @@
+storage:
+  tsdb:
+    retention:
+      percentage: -1

config/testdata/tsdb_retention_size.bad.yml

@@ -0,0 +1,4 @@
+storage:
+  tsdb:
+    retention:
+      size: -1GB

config/testdata/tsdb_retention_time.bad.yml

@@ -0,0 +1,4 @@
+storage:
+  tsdb:
+    retention:
+      time: -1h

tsdb/db.go

@@ -1277,18 +1277,12 @@ func (db *DB) ApplyConfig(conf *config.Config) error {
 		// Update retention configuration if provided.
 		if conf.StorageConfig.TSDBConfig.Retention != nil {
 			db.retentionMtx.Lock()
-			if conf.StorageConfig.TSDBConfig.Retention.Time > 0 {
-				db.opts.RetentionDuration = int64(conf.StorageConfig.TSDBConfig.Retention.Time)
-				db.metrics.retentionDuration.Set((time.Duration(db.opts.RetentionDuration) * time.Millisecond).Seconds())
-			}
-			if conf.StorageConfig.TSDBConfig.Retention.Size > 0 {
-				db.opts.MaxBytes = int64(conf.StorageConfig.TSDBConfig.Retention.Size)
-				db.metrics.maxBytes.Set(float64(db.opts.MaxBytes))
-			}
-			if conf.StorageConfig.TSDBConfig.Retention.Percentage > 0 {
-				db.opts.MaxPercentage = conf.StorageConfig.TSDBConfig.Retention.Percentage
-				db.metrics.maxPercentage.Set(float64(db.opts.MaxPercentage))
-			}
+			db.opts.RetentionDuration = int64(conf.StorageConfig.TSDBConfig.Retention.Time)
+			db.metrics.retentionDuration.Set((time.Duration(db.opts.RetentionDuration) * time.Millisecond).Seconds())
+			db.opts.MaxBytes = int64(conf.StorageConfig.TSDBConfig.Retention.Size)
+			db.metrics.maxBytes.Set(float64(db.opts.MaxBytes))
+			db.opts.MaxPercentage = conf.StorageConfig.TSDBConfig.Retention.Percentage
+			db.metrics.maxPercentage.Set(float64(db.opts.MaxPercentage))
 			db.retentionMtx.Unlock()
 		}
 	} else {