Merge pull request #256 from SiaFoundation/nate/add-volume-read-write…

…-alerts

Add volume read and write alerts

README.md

@@ -101,14 +101,14 @@ rhp3:
 log:
   level: info # global log level
   stdout:
-	enabled: true # enable logging to stdout
+    enabled: true # enable logging to stdout
     level: info # log level for console logger
-	format: human # log format (human, json)
-	enableANSI: true # enable ANSI color codes (disabled on Windows)
+    format: human # log format (human, json)
+    enableANSI: true # enable ANSI color codes (disabled on Windows)
   file:
     enabled: true # enable logging to file
-	level: info # log level for file logger
-	path: /var/log/hostd/hostd.log # the path of the log file
+    level: info # log level for file logger
+    path: /var/log/hostd/hostd.log # the path of the log file
     format: json # log format (human, json)
 ```
 

host/storage/storage.go

@@ -844,6 +844,20 @@ func (vm *VolumeManager) Read(root types.Hash256) (*[rhp2.SectorSize]byte, error
 	vm.mu.Unlock()
 	sector, err := v.ReadSector(loc.Index)
 	if err != nil {
+		stats := v.Stats()
+		vm.a.Register(alerts.Alert{
+			ID:       v.alertID("read"),
+			Severity: alerts.SeverityError,
+			Message:  "Failed to read sector",
+			Data: map[string]interface{}{
+				"volume":       v.Location(),
+				"failedReads":  stats.FailedReads,
+				"failedWrites": stats.FailedWrites,
+				"sector":       root,
+				"error":        err.Error(),
+			},
+			Timestamp: time.Now(),
+		})
 		return nil, fmt.Errorf("failed to read sector data: %w", err)
 	}
 
@@ -910,6 +924,20 @@ func (vm *VolumeManager) Write(root types.Hash256, data *[rhp2.SectorSize]byte)
 
 		// write the sector to the volume
 		if err := vol.WriteSector(data, loc.Index); err != nil {
+			stats := vol.Stats()
+			vm.a.Register(alerts.Alert{
+				ID:       vol.alertID("write"),
+				Severity: alerts.SeverityError,
+				Message:  "Failed to write sector",
+				Data: map[string]interface{}{
+					"volume":       vol.Location(),
+					"failedReads":  stats.FailedReads,
+					"failedWrites": stats.FailedWrites,
+					"sector":       root,
+					"error":        err.Error(),
+				},
+				Timestamp: time.Now(),
+			})
 			return fmt.Errorf("failed to write sector data: %w", err)
 		}
 		vm.log.Debug("wrote sector", zap.String("root", root.String()), zap.Int64("volume", loc.Volume), zap.Uint64("index", loc.Index), zap.Duration("elapsed", time.Since(start)))

host/storage/volume.go

@@ -8,6 +8,7 @@ import (
 	"sync"
 
 	rhp2 "go.sia.tech/core/rhp/v2"
+	"go.sia.tech/core/types"
 	"lukechampine.com/frand"
 )
 
@@ -30,8 +31,9 @@ type (
 		// held.
 		mu sync.RWMutex
 
-		data  volumeData // data is a flatfile that stores the volume's sector data
-		stats VolumeStats
+		location string     // location is the path to the volume's file
+		data     volumeData // data is a flatfile that stores the volume's sector data
+		stats    VolumeStats
 	}
 
 	// VolumeStats contains statistics about a volume
@@ -93,6 +95,20 @@ func (v *volume) appendError(err error) {
 	}
 }
 
+// Location returns the location of the volume
+func (v *volume) Location() string {
+	v.mu.RLock()
+	defer v.mu.RUnlock()
+	return v.location
+}
+
+// alertID returns a deterministic alert ID for a volume and context
+func (v *volume) alertID(context string) types.Hash256 {
+	v.mu.RLock()
+	defer v.mu.RUnlock()
+	return types.HashBytes([]byte(v.location + context))
+}
+
 // OpenVolume opens the volume at localPath
 func (v *volume) OpenVolume(localPath string, reload bool) error {
 	v.mu.Lock()
@@ -104,6 +120,7 @@ func (v *volume) OpenVolume(localPath string, reload bool) error {
 	if err != nil {
 		return err
 	}
+	v.location = localPath
 	v.data = f
 	return nil
 }