X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/b0b276ff6121aace3c52ee855752df6852120343..6613ec1e9c705fb5b950611fd160d4a2babed251:/services/datamanager/datamanager.go

diff --git a/services/datamanager/datamanager.go b/services/datamanager/datamanager.go
index 28d558bb8d..5250d175ff 100644
--- a/services/datamanager/datamanager.go
+++ b/services/datamanager/datamanager.go
@@ -3,6 +3,7 @@
 package main
 
 import (
+	"errors"
 	"flag"
 	"fmt"
 	"git.curoverse.com/arvados.git/sdk/go/arvadosclient"
@@ -21,6 +22,8 @@ var (
 	logEventTypePrefix  string
 	logFrequencySeconds int
 	minutesBetweenRuns  int
+	collectionBatchSize int
+	dryRun              bool
 )
 
 func init() {
@@ -35,39 +38,59 @@ func init() {
 	flag.IntVar(&minutesBetweenRuns,
 		"minutes-between-runs",
 		0,
-		"How many minutes we wait betwen data manager runs. 0 means run once and exit.")
+		"How many minutes we wait between data manager runs. 0 means run once and exit.")
+	flag.IntVar(&collectionBatchSize,
+		"collection-batch-size",
+		1000,
+		"How many collections to request in each batch.")
+	flag.BoolVar(&dryRun,
+		"dry-run",
+		false,
+		"Perform a dry run. Log how many blocks would be deleted/moved, but do not issue any changes to keepstore.")
 }
 
 func main() {
 	flag.Parse()
+
 	if minutesBetweenRuns == 0 {
-		singlerun()
+		arv, err := arvadosclient.MakeArvadosClient()
+		if err != nil {
+			loggerutil.FatalWithMessage(arvLogger, fmt.Sprintf("Error making arvados client: %v", err))
+		}
+		err = singlerun(arv)
+		if err != nil {
+			loggerutil.FatalWithMessage(arvLogger, fmt.Sprintf("singlerun: %v", err))
+		}
 	} else {
 		waitTime := time.Minute * time.Duration(minutesBetweenRuns)
 		for {
 			log.Println("Beginning Run")
-			singlerun()
+			arv, err := arvadosclient.MakeArvadosClient()
+			if err != nil {
+				loggerutil.FatalWithMessage(arvLogger, fmt.Sprintf("Error making arvados client: %v", err))
+			}
+			err = singlerun(arv)
+			if err != nil {
+				log.Printf("singlerun: %v", err)
+			}
 			log.Printf("Sleeping for %d minutes", minutesBetweenRuns)
 			time.Sleep(waitTime)
 		}
 	}
 }
 
-func singlerun() {
-	arv, err := arvadosclient.MakeArvadosClient()
-	if err != nil {
-		log.Fatalf("Error setting up arvados client %s", err.Error())
-	}
+var arvLogger *logger.Logger
 
-	if is_admin, err := util.UserIsAdmin(arv); err != nil {
-		log.Fatalf("Error querying current arvados user %s", err.Error())
-	} else if !is_admin {
-		log.Fatalf("Current user is not an admin. Datamanager can only be run by admins.")
+func singlerun(arv *arvadosclient.ArvadosClient) error {
+	var err error
+	if isAdmin, err := util.UserIsAdmin(arv); err != nil {
+		return errors.New("Error verifying admin token: " + err.Error())
+	} else if !isAdmin {
+		return errors.New("Current user is not an admin. Datamanager requires a privileged token.")
 	}
 
-	var arvLogger *logger.Logger
 	if logEventTypePrefix != "" {
-		arvLogger = logger.NewLogger(logger.LoggerParams{
+		arvLogger, err = logger.NewLogger(logger.LoggerParams{
 			Client:          arv,
 			EventTypePrefix: logEventTypePrefix,
 			WriteInterval:   time.Second * time.Duration(logFrequencySeconds)})
@@ -90,9 +113,15 @@ func singlerun() {
 		dataFetcher = BuildDataFetcher(arv)
 	}
 
-	dataFetcher(arvLogger, &readCollections, &keepServerInfo)
+	err = dataFetcher(arvLogger, &readCollections, &keepServerInfo)
+	if err != nil {
+		return err
+	}
 
-	summary.MaybeWriteData(arvLogger, readCollections, keepServerInfo)
+	err = summary.MaybeWriteData(arvLogger, readCollections, keepServerInfo)
+	if err != nil {
+		return err
+	}
 
 	buckets := summary.BucketReplication(readCollections, keepServerInfo)
 	bucketCounts := buckets.Counts()
@@ -113,30 +142,15 @@ func singlerun() {
 			rlbss.Count)
 	}
 
-	kc, err := keepclient.MakeKeepClient(&arv)
+	kc, err := keepclient.MakeKeepClient(arv)
 	if err != nil {
-		loggerutil.FatalWithMessage(arvLogger,
-			fmt.Sprintf("Error setting up keep client %s", err.Error()))
+		return fmt.Errorf("Error setting up keep client %v", err.Error())
 	}
 
-	pullServers := summary.ComputePullServers(kc,
-		&keepServerInfo,
-		readCollections.BlockToDesiredReplication,
-		replicationSummary.UnderReplicatedBlocks)
-
-	pullLists := summary.BuildPullLists(pullServers)
-	trashLists := summary.BuildTrashLists(kc,
-		&keepServerInfo,
-		replicationSummary.KeepBlocksNotInCollections)
-
-	summary.WritePullLists(arvLogger, pullLists)
-
-	keep.SendTrashLists(arvLogger, kc, trashLists)
-
 	// Log that we're finished. We force the recording, since go will
 	// not wait for the write timer before exiting.
 	if arvLogger != nil {
-		arvLogger.FinalUpdate(func(p map[string]interface{}, e map[string]interface{}) {
+		defer arvLogger.FinalUpdate(func(p map[string]interface{}, e map[string]interface{}) {
 			summaryInfo := logger.GetOrCreateMap(p, "summary_info")
 			summaryInfo["block_replication_counts"] = bucketCounts
 			summaryInfo["replication_summary"] = replicationCounts
@@ -145,29 +159,62 @@ func singlerun() {
 			p["run_info"].(map[string]interface{})["finished_at"] = time.Now()
 		})
 	}
+
+	pullServers := summary.ComputePullServers(kc,
+		&keepServerInfo,
+		readCollections.BlockToDesiredReplication,
+		replicationSummary.UnderReplicatedBlocks)
+
+	pullLists := summary.BuildPullLists(pullServers)
+
+	trashLists, trashErr := summary.BuildTrashLists(kc,
+		&keepServerInfo,
+		replicationSummary.KeepBlocksNotInCollections)
+
+	err = summary.WritePullLists(arvLogger, pullLists, dryRun)
+	if err != nil {
+		return err
+	}
+
+	if trashErr != nil {
+		return err
+	}
+	keep.SendTrashLists(arvLogger, kc, trashLists, dryRun)
+
+	return nil
 }
 
-// Returns a data fetcher that fetches data from remote servers.
-func BuildDataFetcher(arv arvadosclient.ArvadosClient) summary.DataFetcher {
-	return func(arvLogger *logger.Logger,
+// BuildDataFetcher returns a data fetcher that fetches data from remote servers.
+func BuildDataFetcher(arv *arvadosclient.ArvadosClient) summary.DataFetcher {
+	return func(
+		arvLogger *logger.Logger,
 		readCollections *collection.ReadCollections,
-		keepServerInfo *keep.ReadServers) {
-		collectionChannel := make(chan collection.ReadCollections)
-
+		keepServerInfo *keep.ReadServers,
+	) error {
+		collDone := make(chan struct{})
+		var collErr error
 		go func() {
-			collectionChannel <- collection.GetCollectionsAndSummarize(
+			*readCollections, collErr = collection.GetCollectionsAndSummarize(
 				collection.GetCollectionsParams{
 					Client:    arv,
 					Logger:    arvLogger,
-					BatchSize: 50})
+					BatchSize: collectionBatchSize})
+			collDone <- struct{}{}
 		}()
 
-		*keepServerInfo = keep.GetKeepServersAndSummarize(
+		var keepErr error
+		*keepServerInfo, keepErr = keep.GetKeepServersAndSummarize(
 			keep.GetKeepServersParams{
 				Client: arv,
 				Logger: arvLogger,
 				Limit:  1000})
 
-		*readCollections = <-collectionChannel
+		<-collDone
+
+		// Return a nil error only if both parts succeeded.
+		if collErr != nil {
+			return collErr
+		}
+		return keepErr
 	}
 }