feat: change stalled disk propagation from an error to a field in ful…

…l status output

Signed-off-by: Manan Gupta <[email protected]>

go/vt/vtorc/inst/instance_dao.go

@@ -206,9 +206,10 @@ func ReadTopologyInstanceBufferable(tabletAlias string, latency *stopwatch.Named
 
 	fs, err = fullStatus(tabletAlias)
 	if err != nil {
-		if config.GetStalledDiskPrimaryRecovery() && strings.Contains(err.Error(), "stalled disk") {
-			stalledDisk = true
-		}
+		goto Cleanup
+	}
+	if config.GetStalledDiskPrimaryRecovery() && fs.DiskStalled {
+		stalledDisk = true
 		goto Cleanup
 	}
 	partialSuccess = true // We at least managed to read something from the server.

go/vt/vttablet/tabletmanager/rpc_replication.go

@@ -18,7 +18,6 @@ package tabletmanager
 
 import (
 	"context"
-	"errors"
 	"fmt"
 	"runtime"
 	"strings"
@@ -62,10 +61,13 @@ func (tm *TabletManager) FullStatus(ctx context.Context) (*replicationdatapb.Ful
 		return nil, err
 	}
 
-	// Return error if the disk is stalled or rejecting writes.
-	// Noop by default, must be enabled with the flag "disk-write-dir".
+	// Return if the disk is stalled or rejecting writes.
+	// If the disk is stalled, we can't be sure if reads will go through
+	// or not, so we should not run any reads either.
 	if tm.dhMonitor.IsDiskStalled() {
-		return nil, errors.New("stalled disk")
+		return &replicationdatapb.FullStatus{
+			DiskStalled: true,
+		}, nil
 	}
 
 	// Server ID - "select @@global.server_id"

proto/replicationdata.proto

@@ -105,4 +105,5 @@ message FullStatus {
   uint32 semi_sync_wait_for_replica_count = 20;
   bool super_read_only = 21;
   replicationdata.Configuration replication_configuration = 22;
+  bool disk_stalled = 23;
 }