X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/9c8bad50218730ac2c640f8e1c00fd12f2f174e4..27daf08f38eec505c224e7776678b32d50241e13:/services/datamanager/datamanager.go

diff --git a/services/datamanager/datamanager.go b/services/datamanager/datamanager.go
index a8e506eacb..70a9ae7859 100644
--- a/services/datamanager/datamanager.go
+++ b/services/datamanager/datamanager.go
@@ -4,12 +4,15 @@ package main
 
 import (
 	"flag"
+	"fmt"
 	"git.curoverse.com/arvados.git/sdk/go/arvadosclient"
+	"git.curoverse.com/arvados.git/sdk/go/keepclient"
 	"git.curoverse.com/arvados.git/sdk/go/logger"
 	"git.curoverse.com/arvados.git/sdk/go/util"
 	"git.curoverse.com/arvados.git/services/datamanager/collection"
 	"git.curoverse.com/arvados.git/services/datamanager/keep"
 	"git.curoverse.com/arvados.git/services/datamanager/loggerutil"
+	"git.curoverse.com/arvados.git/services/datamanager/summary"
 	"log"
 	"time"
 )
@@ -38,19 +41,25 @@ func init() {
 func main() {
 	flag.Parse()
 	if minutesBetweenRuns == 0 {
-		singlerun()
+		err := singlerun()
+		if err != nil {
+			log.Fatalf("Got an error: %v", err)
+		}
 	} else {
 		waitTime := time.Minute * time.Duration(minutesBetweenRuns)
 		for {
 			log.Println("Beginning Run")
-			singlerun()
+			err := singlerun()
+			if err != nil {
+				log.Printf("Got an error: %v", err)
+			}
 			log.Printf("Sleeping for %d minutes", minutesBetweenRuns)
 			time.Sleep(waitTime)
 		}
 	}
 }
 
-func singlerun() {
+func singlerun() error {
 	arv, err := arvadosclient.MakeArvadosClient()
 	if err != nil {
 		log.Fatalf("Error setting up arvados client %s", err.Error())
@@ -64,7 +73,8 @@ func singlerun() {
 
 	var arvLogger *logger.Logger
 	if logEventTypePrefix != "" {
-		arvLogger = logger.NewLogger(logger.LoggerParams{Client: arv,
+		arvLogger = logger.NewLogger(logger.LoggerParams{
+			Client:          arv,
 			EventTypePrefix: logEventTypePrefix,
 			WriteInterval:   time.Second * time.Duration(logFrequencySeconds)})
 	}
@@ -74,28 +84,103 @@ func singlerun() {
 		arvLogger.AddWriteHook(loggerutil.LogMemoryAlloc)
 	}
 
-	collectionChannel := make(chan collection.ReadCollections)
+	var (
+		dataFetcher     summary.DataFetcher
+		readCollections collection.ReadCollections
+		keepServerInfo  keep.ReadServers
+	)
+
+	if summary.ShouldReadData() {
+		dataFetcher = summary.ReadData
+	} else {
+		dataFetcher = BuildDataFetcher(arv)
+	}
 
-	go func() {
-		collectionChannel <- collection.GetCollectionsAndSummarize(
-			collection.GetCollectionsParams{
-				Client: arv, Logger: arvLogger, BatchSize: 50})
-	}()
+	dataFetcher(arvLogger, &readCollections, &keepServerInfo)
 
-	keepServerInfo := keep.GetKeepServersAndSummarize(
-		keep.GetKeepServersParams{Client: arv, Logger: arvLogger, Limit: 1000})
+	summary.MaybeWriteData(arvLogger, readCollections, keepServerInfo)
 
-	readCollections := <-collectionChannel
+	buckets := summary.BucketReplication(readCollections, keepServerInfo)
+	bucketCounts := buckets.Counts()
 
-	// TODO(misha): Use these together to verify replication.
-	_ = readCollections
-	_ = keepServerInfo
+	replicationSummary := buckets.SummarizeBuckets(readCollections)
+	replicationCounts := replicationSummary.ComputeCounts()
+
+	log.Printf("Blocks In Collections: %d, "+
+		"\nBlocks In Keep: %d.",
+		len(readCollections.BlockToDesiredReplication),
+		len(keepServerInfo.BlockToServers))
+	log.Println(replicationCounts.PrettyPrint())
+
+	log.Printf("Blocks Histogram:")
+	for _, rlbss := range bucketCounts {
+		log.Printf("%+v: %10d",
+			rlbss.Levels,
+			rlbss.Count)
+	}
+
+	kc, err := keepclient.MakeKeepClient(&arv)
+	if err != nil {
+		loggerutil.FatalWithMessage(arvLogger,
+			fmt.Sprintf("Error setting up keep client %s", err.Error()))
+	}
 
 	// Log that we're finished. We force the recording, since go will
-	// not wait for the timer before exiting.
+	// 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
+			p["summary_info"] = summaryInfo
+
 			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)
+
+	summary.WritePullLists(arvLogger, pullLists)
+
+	if trashErr != nil {
+		return err
+	} else {
+		keep.SendTrashLists(keep.GetDataManagerToken(arvLogger), kc, trashLists)
+	}
+
+	return nil
+}
+
+// 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)
+
+		go func() {
+			collectionChannel <- collection.GetCollectionsAndSummarize(
+				collection.GetCollectionsParams{
+					Client:    arv,
+					Logger:    arvLogger,
+					BatchSize: 50})
+		}()
+
+		*keepServerInfo = keep.GetKeepServersAndSummarize(
+			keep.GetKeepServersParams{
+				Client: arv,
+				Logger: arvLogger,
+				Limit:  1000})
+
+		*readCollections = <-collectionChannel
+	}
 }