X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/367a6fbc62b4b20af9f5724359fb0d0e423dc718..88c382d13b3d6e6f3b03ba0d5139ad9552c3c359:/services/datamanager/datamanager.go diff --git a/services/datamanager/datamanager.go b/services/datamanager/datamanager.go index 4c12c45bad..4a3b5627dd 100644 --- a/services/datamanager/datamanager.go +++ b/services/datamanager/datamanager.go @@ -3,109 +3,218 @@ package main import ( + "errors" "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" - "os" "time" ) var ( - logEventType string + logEventTypePrefix string logFrequencySeconds int + minutesBetweenRuns int + collectionBatchSize int + dryRun bool ) func init() { - flag.StringVar(&logEventType, - "log-event-type", - "experimental-data-manager-report", - "event_type to use in our arvados log entries.") - flag.IntVar(&logFrequencySeconds, + flag.StringVar(&logEventTypePrefix, + "log-event-type-prefix", + "experimental-data-manager", + "Prefix to use in the event_type of our arvados log entries. Set to empty to turn off logging") + flag.IntVar(&logFrequencySeconds, "log-frequency-seconds", 20, "How frequently we'll write log entries in seconds.") + flag.IntVar(&minutesBetweenRuns, + "minutes-between-runs", + 0, + "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() - arv, err := arvadosclient.MakeArvadosClient() - if err != nil { - log.Fatalf("Error setting up arvados client %s", err.Error()) + if minutesBetweenRuns == 0 { + 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") + 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) + } } +} + +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.") } - arvLogger := logger.NewLogger(logger.LoggerParams{Client: arv, - EventType: logEventType, - MinimumWriteInterval: time.Second * time.Duration(logFrequencySeconds)}) + if logEventTypePrefix != "" { + arvLogger, err = logger.NewLogger(logger.LoggerParams{ + Client: arv, + EventTypePrefix: logEventTypePrefix, + WriteInterval: time.Second * time.Duration(logFrequencySeconds)}) + } - { - properties, _ := arvLogger.Edit() - properties["start_time"] = time.Now() - properties["args"] = os.Args - hostname, err := os.Hostname() - if err != nil { - properties["hostname_error"] = err.Error() - } else { - properties["hostname"] = hostname - } + loggerutil.LogRunInfo(arvLogger) + if arvLogger != nil { + arvLogger.AddWriteHook(loggerutil.LogMemoryAlloc) } - arvLogger.Record() - // TODO(misha): Read Collections and Keep Contents concurrently as goroutines. - // This requires waiting on them to finish before you let main() exit. + var ( + dataFetcher summary.DataFetcher + readCollections collection.ReadCollections + keepServerInfo keep.ReadServers + ) - RunCollections(collection.GetCollectionsParams{ - Client: arv, Logger: arvLogger, BatchSize: 500}) + if summary.ShouldReadData() { + dataFetcher = summary.ReadData + } else { + dataFetcher = BuildDataFetcher(arv) + } - RunKeep(keep.GetKeepServersParams{Client: arv, Limit: 1000}) -} + err = dataFetcher(arvLogger, &readCollections, &keepServerInfo) + if err != nil { + return err + } -func RunCollections(params collection.GetCollectionsParams) { - readCollections := collection.GetCollections(params) + err = summary.MaybeWriteData(arvLogger, readCollections, keepServerInfo) + if err != nil { + return err + } - UserUsage := ComputeSizeOfOwnedCollections(readCollections) - log.Printf("Uuid to Size used: %v", UserUsage) + buckets := summary.BucketReplication(readCollections, keepServerInfo) + bucketCounts := buckets.Counts() - // TODO(misha): Add a "readonly" flag. If we're in readonly mode, - // lots of behaviors can become warnings (and obviously we can't - // write anything). - // if !readCollections.ReadAllCollections { - // log.Fatalf("Did not read all collections") - // } + replicationSummary := buckets.SummarizeBuckets(readCollections) + replicationCounts := replicationSummary.ComputeCounts() - log.Printf("Read and processed %d collections", - len(readCollections.UuidToCollection)) -} + log.Printf("Blocks In Collections: %d, "+ + "\nBlocks In Keep: %d.", + len(readCollections.BlockToDesiredReplication), + len(keepServerInfo.BlockToServers)) + log.Println(replicationCounts.PrettyPrint()) -func RunKeep(params keep.GetKeepServersParams) { - readServers := keep.GetKeepServers(params) + log.Printf("Blocks Histogram:") + for _, rlbss := range bucketCounts { + log.Printf("%+v: %10d", + rlbss.Levels, + rlbss.Count) + } - log.Printf("Returned %d keep disks", len(readServers.ServerToContents)) + kc, err := keepclient.MakeKeepClient(&arv) + if err != nil { + return fmt.Errorf("Error setting up keep client %v", err.Error()) + } - blockReplicationCounts := make(map[int]int) - for _, infos := range readServers.BlockToServers { - replication := len(infos) - blockReplicationCounts[replication] += 1 + // Log that we're finished. We force the recording, since go will + // not wait for the write timer before exiting. + if arvLogger != nil { + 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() + }) } - log.Printf("Replication level distribution: %v", blockReplicationCounts) + 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 } -func ComputeSizeOfOwnedCollections(readCollections collection.ReadCollections) ( - results map[string]int) { - results = make(map[string]int) - for _, coll := range readCollections.UuidToCollection { - results[coll.OwnerUuid] = results[coll.OwnerUuid] + coll.TotalSize +// 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, + ) error { + collDone := make(chan struct{}) + var collErr error + go func() { + *readCollections, collErr = collection.GetCollectionsAndSummarize( + collection.GetCollectionsParams{ + Client: arv, + Logger: arvLogger, + BatchSize: collectionBatchSize}) + collDone <- struct{}{} + }() + + var keepErr error + *keepServerInfo, keepErr = keep.GetKeepServersAndSummarize( + keep.GetKeepServersParams{ + Client: arv, + Logger: arvLogger, + Limit: 1000}) + + <- collDone + + // Return a nil error only if both parts succeeded. + if collErr != nil { + return collErr + } + return keepErr } - return }