cmd/swarm/swarm-smoke: Trigger chunk debug on timeout

cmd/swarm/swarm-smoke/main.go

@@ -37,15 +37,16 @@ var (
 )
 
 var (
-	allhosts  string
-	hosts     []string
-	filesize  int
-	syncDelay int
-	httpPort  int
-	wsPort    int
-	verbosity int
-	timeout   int
-	single    bool
+	allhosts          string
+	hosts             []string
+	filesize          int
+	syncDelay         int
+	httpPort          int
+	wsPort            int
+	verbosity         int
+	timeout           int
+	single            bool
+	getAllRefsTimeout int
 )
 
 func main() {
@@ -102,6 +103,12 @@ func main() {
 			Usage:       "whether to fetch content from a single node or from all nodes",
 			Destination: &single,
 		},
+		cli.IntFlag{
+			Name:        "refs-timeout",
+			Value:       5,
+			Usage:       "timeout in seconds to wait for GetAllReferences to return",
+			Destination: &getAllRefsTimeout,
+		},
 	}
 
 	app.Flags = append(app.Flags, []cli.Flag{

cmd/swarm/swarm-smoke/upload_and_sync.go

@@ -18,13 +18,19 @@ package main
 
 import (
 	"bytes"
+	"context"
 	"fmt"
+	"io/ioutil"
 	"math/rand"
+	"os"
 	"sync"
 	"time"
 
 	"github.com/ethereum/go-ethereum/log"
 	"github.com/ethereum/go-ethereum/metrics"
+	"github.com/ethereum/go-ethereum/rpc"
+	"github.com/ethereum/go-ethereum/swarm/api"
+	"github.com/ethereum/go-ethereum/swarm/storage"
 	"github.com/ethereum/go-ethereum/swarm/testutil"
 	"github.com/pborman/uuid"
 
@@ -49,12 +55,75 @@ func uploadAndSyncCmd(ctx *cli.Context, tuid string) error {
 	case <-time.After(time.Duration(timeout) * time.Second):
 		metrics.GetOrRegisterCounter(fmt.Sprintf("%s.timeout", commandName), nil).Inc(1)
 
+		e := fmt.Errorf("timeout after %v sec", timeout)
 		// trigger debug functionality on randomBytes
+		err := triggerChunkDebug(randomBytes[:])
+		if err != nil {
+			e = fmt.Errorf("%v; triggerChunkDebug failed: %v", e, err)
+		}
 
-		return fmt.Errorf("timeout after %v sec", timeout)
+		return e
 	}
 }
 
+func triggerChunkDebug(testData []byte) error {
+	log.Warn("Test timed out; running chunk debug sequence")
+
+	addrs, err := getAllRefs(testData)
+	if err != nil {
+		return err
+	}
+	log.Trace("All references retrieved")
+
+	// has-chunks
+	for _, host := range hosts {
+		httpHost := fmt.Sprintf("ws://%s:%d", host, 8546)
+		log.Trace("Calling `Has` on host", "httpHost", httpHost)
+		rpcClient, err := rpc.Dial(httpHost)
+		if err != nil {
+			log.Trace("Error dialing host", "err", err)
+			return err
+		}
+		log.Trace("rpc dial ok")
+		var hasInfo []api.HasInfo
+		err = rpcClient.Call(&hasInfo, "bzz_has", addrs)
+		if err != nil {
+			log.Trace("Error calling host", "err", err)
+			return err
+		}
+		log.Trace("rpc call ok")
+		count := 0
+		for _, info := range hasInfo {
+			if !info.Has {
+				count++
+				log.Error("Host does not have chunk", "host", httpHost, "chunk", info.Addr)
+			}
+		}
+		if count == 0 {
+			log.Info("Host reported to have all chunks", "host", httpHost)
+		}
+	}
+	return nil
+}
+
+func getAllRefs(testData []byte) (storage.AddressCollection, error) {
+	log.Trace("Getting all references for given root hash")
+	datadir, err := ioutil.TempDir("", "chunk-debug")
+	if err != nil {
+		return nil, fmt.Errorf("unable to create temp dir: %v", err)
+	}
+	defer os.RemoveAll(datadir)
+	fileStore, err := storage.NewLocalFileStore(datadir, make([]byte, 32))
+	if err != nil {
+		return nil, err
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), time.Duration(getAllRefsTimeout)*time.Second)
+	defer cancel()
+
+	reader := bytes.NewReader(testData)
+	return fileStore.GetAllReferences(ctx, reader, false)
+}
+
 func uplaodAndSync(c *cli.Context, randomBytes []byte, tuid string) error {
 	log.Info("uploading to "+httpEndpoint(hosts[0])+" and syncing", "tuid", tuid, "seed", seed)