1 /* Keep Datamanager. Responsible for checking on and reporting on Keep Storage */
9 "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
10 "git.curoverse.com/arvados.git/sdk/go/keepclient"
11 "git.curoverse.com/arvados.git/sdk/go/logger"
12 "git.curoverse.com/arvados.git/sdk/go/util"
13 "git.curoverse.com/arvados.git/services/datamanager/collection"
14 "git.curoverse.com/arvados.git/services/datamanager/keep"
15 "git.curoverse.com/arvados.git/services/datamanager/loggerutil"
16 "git.curoverse.com/arvados.git/services/datamanager/summary"
22 logEventTypePrefix string
23 logFrequencySeconds int
24 minutesBetweenRuns int
28 flag.StringVar(&logEventTypePrefix,
29 "log-event-type-prefix",
30 "experimental-data-manager",
31 "Prefix to use in the event_type of our arvados log entries. Set to empty to turn off logging")
32 flag.IntVar(&logFrequencySeconds,
33 "log-frequency-seconds",
35 "How frequently we'll write log entries in seconds.")
36 flag.IntVar(&minutesBetweenRuns,
37 "minutes-between-runs",
39 "How many minutes we wait betwen data manager runs. 0 means run once and exit.")
44 if minutesBetweenRuns == 0 {
45 arv, err := arvadosclient.MakeArvadosClient()
47 loggerutil.FatalWithMessage(arvLogger, fmt.Sprintf("Error making arvados client: %v", err))
51 loggerutil.FatalWithMessage(arvLogger, fmt.Sprintf("singlerun: %v", err))
54 waitTime := time.Minute * time.Duration(minutesBetweenRuns)
56 log.Println("Beginning Run")
57 arv, err := arvadosclient.MakeArvadosClient()
59 loggerutil.FatalWithMessage(arvLogger, fmt.Sprintf("Error making arvados client: %v", err))
63 log.Printf("singlerun: %v", err)
65 log.Printf("Sleeping for %d minutes", minutesBetweenRuns)
71 var arvLogger *logger.Logger
73 func singlerun(arv arvadosclient.ArvadosClient) error {
75 if isAdmin, err := util.UserIsAdmin(arv); err != nil {
76 return errors.New("Error verifying admin token: " + err.Error())
78 return errors.New("Current user is not an admin. Datamanager requires a privileged token.")
81 if logEventTypePrefix != "" {
82 arvLogger = logger.NewLogger(logger.LoggerParams{
84 EventTypePrefix: logEventTypePrefix,
85 WriteInterval: time.Second * time.Duration(logFrequencySeconds)})
88 loggerutil.LogRunInfo(arvLogger)
90 arvLogger.AddWriteHook(loggerutil.LogMemoryAlloc)
94 dataFetcher summary.DataFetcher
95 readCollections collection.ReadCollections
96 keepServerInfo keep.ReadServers
99 if summary.ShouldReadData() {
100 dataFetcher = summary.ReadData
102 dataFetcher = BuildDataFetcher(arv)
105 dataFetcher(arvLogger, &readCollections, &keepServerInfo)
107 if readCollections.Err != nil {
108 return readCollections.Err
111 err = summary.MaybeWriteData(arvLogger, readCollections, keepServerInfo)
116 buckets := summary.BucketReplication(readCollections, keepServerInfo)
117 bucketCounts := buckets.Counts()
119 replicationSummary := buckets.SummarizeBuckets(readCollections)
120 replicationCounts := replicationSummary.ComputeCounts()
122 log.Printf("Blocks In Collections: %d, "+
123 "\nBlocks In Keep: %d.",
124 len(readCollections.BlockToDesiredReplication),
125 len(keepServerInfo.BlockToServers))
126 log.Println(replicationCounts.PrettyPrint())
128 log.Printf("Blocks Histogram:")
129 for _, rlbss := range bucketCounts {
130 log.Printf("%+v: %10d",
135 kc, err := keepclient.MakeKeepClient(&arv)
137 return fmt.Errorf("Error setting up keep client %v", err.Error())
140 // Log that we're finished. We force the recording, since go will
141 // not wait for the write timer before exiting.
142 if arvLogger != nil {
143 defer arvLogger.FinalUpdate(func(p map[string]interface{}, e map[string]interface{}) {
144 summaryInfo := logger.GetOrCreateMap(p, "summary_info")
145 summaryInfo["block_replication_counts"] = bucketCounts
146 summaryInfo["replication_summary"] = replicationCounts
147 p["summary_info"] = summaryInfo
149 p["run_info"].(map[string]interface{})["finished_at"] = time.Now()
153 pullServers := summary.ComputePullServers(kc,
155 readCollections.BlockToDesiredReplication,
156 replicationSummary.UnderReplicatedBlocks)
158 pullLists := summary.BuildPullLists(pullServers)
160 trashLists, trashErr := summary.BuildTrashLists(kc,
162 replicationSummary.KeepBlocksNotInCollections)
164 err = summary.WritePullLists(arvLogger, pullLists)
172 keep.SendTrashLists(kc, trashLists)
177 // BuildDataFetcher returns a data fetcher that fetches data from remote servers.
178 func BuildDataFetcher(arv arvadosclient.ArvadosClient) summary.DataFetcher {
179 return func(arvLogger *logger.Logger,
180 readCollections *collection.ReadCollections,
181 keepServerInfo *keep.ReadServers) {
182 collectionChannel := make(chan collection.ReadCollections)
185 collectionChannel <- collection.GetCollectionsAndSummarize(
186 collection.GetCollectionsParams{
193 *keepServerInfo, err = keep.GetKeepServersAndSummarize(
194 keep.GetKeepServersParams{
203 *readCollections = <-collectionChannel