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"
logEventTypePrefix string
logFrequencySeconds int
minutesBetweenRuns int
+ collectionBatchSize int
+ dryRun bool
)
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)})
}
var (
+ dataFetcher summary.DataFetcher
readCollections collection.ReadCollections
keepServerInfo keep.ReadServers
)
- if !summary.MaybeReadData(arvLogger, &readCollections, &keepServerInfo) {
- 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})
+ if summary.ShouldReadData() {
+ dataFetcher = summary.ReadData
+ } else {
+ dataFetcher = BuildDataFetcher(arv)
+ }
- readCollections = <-collectionChannel
+ 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()
log.Printf("Blocks In Collections: %d, "+
"\nBlocks In Keep: %d.",
- len(readCollections.BlockToReplication),
+ len(readCollections.BlockToDesiredReplication),
len(keepServerInfo.BlockToServers))
log.Println(replicationCounts.PrettyPrint())
rlbss.Count)
}
+ kc, err := keepclient.MakeKeepClient(&arv)
+ if err != nil {
+ return fmt.Errorf("Error setting up keep client %v", err.Error())
+ }
+
// 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
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
+}
+
+// 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
+ }
}