1 /* Deals with getting Keep Server blocks from API Server and Keep Servers. */
10 "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
11 "git.curoverse.com/arvados.git/sdk/go/blockdigest"
12 "git.curoverse.com/arvados.git/sdk/go/logger"
13 "git.curoverse.com/arvados.git/services/datamanager/loggerutil"
23 type ServerAddress struct {
24 Host string `json:"service_host"`
25 Port int `json:"service_port"`
26 Uuid string `json:"uuid"`
29 // Info about a particular block returned by the server
30 type BlockInfo struct {
31 Digest blockdigest.DigestWithSize
32 Mtime int64 // TODO(misha): Replace this with a timestamp.
35 // Info about a specified block given by a server
36 type BlockServerInfo struct {
38 Mtime int64 // TODO(misha): Replace this with a timestamp.
41 type ServerContents struct {
42 BlockDigestToInfo map[blockdigest.DigestWithSize]BlockInfo
45 type ServerResponse struct {
47 Contents ServerContents
50 type ReadServers struct {
52 KeepServerIndexToAddress []ServerAddress
53 KeepServerAddressToIndex map[ServerAddress]int
54 ServerToContents map[ServerAddress]ServerContents
55 BlockToServers map[blockdigest.DigestWithSize][]BlockServerInfo
56 BlockReplicationCounts map[int]int
59 type GetKeepServersParams struct {
60 Client arvadosclient.ArvadosClient
65 type KeepServiceList struct {
66 ItemsAvailable int `json:"items_available"`
67 KeepServers []ServerAddress `json:"items"`
71 // Don't access the token directly, use getDataManagerToken() to
72 // make sure it's been read.
73 dataManagerToken string
74 dataManagerTokenFile string
75 dataManagerTokenFileReadOnce sync.Once
79 flag.StringVar(&dataManagerTokenFile,
80 "data-manager-token-file",
82 "File with the API token we should use to contact keep servers.")
85 // TODO(misha): Change this to include the UUID as well.
86 func (s ServerAddress) String() string {
90 func (s ServerAddress) HostPort() string {
91 return fmt.Sprintf("%s:%d", s.Host, s.Port)
94 func getDataManagerToken(arvLogger *logger.Logger) string {
95 readDataManagerToken := func() {
96 if dataManagerTokenFile == "" {
98 loggerutil.FatalWithMessage(arvLogger,
99 "Data Manager Token needed, but data manager token file not specified.")
101 rawRead, err := ioutil.ReadFile(dataManagerTokenFile)
103 loggerutil.FatalWithMessage(arvLogger,
104 fmt.Sprintf("Unexpected error reading token file %s: %v",
105 dataManagerTokenFile,
108 dataManagerToken = strings.TrimSpace(string(rawRead))
112 dataManagerTokenFileReadOnce.Do(readDataManagerToken)
113 return dataManagerToken
116 func GetKeepServersAndSummarize(params GetKeepServersParams) (results ReadServers) {
117 results = GetKeepServers(params)
118 log.Printf("Returned %d keep disks", len(results.ServerToContents))
120 results.Summarize(params.Logger)
121 log.Printf("Replication level distribution: %v",
122 results.BlockReplicationCounts)
127 func GetKeepServers(params GetKeepServersParams) (results ReadServers) {
128 if ¶ms.Client == nil {
129 log.Fatalf("params.Client passed to GetKeepServers() should " +
130 "contain a valid ArvadosClient, but instead it is nil.")
133 sdkParams := arvadosclient.Dict{
134 "filters": [][]string{[]string{"service_type", "=", "disk"}},
136 if params.Limit > 0 {
137 sdkParams["limit"] = params.Limit
140 var sdkResponse KeepServiceList
141 err := params.Client.List("keep_services", sdkParams, &sdkResponse)
144 loggerutil.FatalWithMessage(params.Logger,
145 fmt.Sprintf("Error requesting keep disks from API server: %v", err))
148 if params.Logger != nil {
149 params.Logger.Update(func(p map[string]interface{}, e map[string]interface{}) {
150 keepInfo := logger.GetOrCreateMap(p, "keep_info")
151 keepInfo["num_keep_servers_available"] = sdkResponse.ItemsAvailable
152 keepInfo["num_keep_servers_received"] = len(sdkResponse.KeepServers)
153 keepInfo["keep_servers"] = sdkResponse.KeepServers
157 log.Printf("Received keep services list: %+v", sdkResponse)
159 if len(sdkResponse.KeepServers) < sdkResponse.ItemsAvailable {
160 loggerutil.FatalWithMessage(params.Logger,
161 fmt.Sprintf("Did not receive all available keep servers: %+v", sdkResponse))
164 results.KeepServerIndexToAddress = sdkResponse.KeepServers
165 results.KeepServerAddressToIndex = make(map[ServerAddress]int)
166 for i, address := range results.KeepServerIndexToAddress {
167 results.KeepServerAddressToIndex[address] = i
170 log.Printf("Got Server Addresses: %v", results)
172 // This is safe for concurrent use
173 client := http.Client{}
175 // Send off all the index requests concurrently
176 responseChan := make(chan ServerResponse)
177 for _, keepServer := range sdkResponse.KeepServers {
178 // The above keepsServer variable is reused for each iteration, so
179 // it would be shared across all goroutines. This would result in
180 // us querying one server n times instead of n different servers
181 // as we intended. To avoid this we add it as an explicit
182 // parameter which gets copied. This bug and solution is described
183 // in https://golang.org/doc/effective_go.html#channels
184 go func(keepServer ServerAddress) {
185 responseChan <- GetServerContents(params.Logger,
191 results.ServerToContents = make(map[ServerAddress]ServerContents)
192 results.BlockToServers = make(map[blockdigest.DigestWithSize][]BlockServerInfo)
194 // Read all the responses
195 for i := range sdkResponse.KeepServers {
196 _ = i // Here to prevent go from complaining.
197 response := <-responseChan
198 log.Printf("Received channel response from %v containing %d files",
200 len(response.Contents.BlockDigestToInfo))
201 results.ServerToContents[response.Address] = response.Contents
202 serverIndex := results.KeepServerAddressToIndex[response.Address]
203 for _, blockInfo := range response.Contents.BlockDigestToInfo {
204 results.BlockToServers[blockInfo.Digest] = append(
205 results.BlockToServers[blockInfo.Digest],
206 BlockServerInfo{ServerIndex: serverIndex,
207 Mtime: blockInfo.Mtime})
213 func GetServerContents(arvLogger *logger.Logger,
214 keepServer ServerAddress,
215 client http.Client) (response ServerResponse) {
217 GetServerStatus(arvLogger, keepServer, client)
219 req := CreateIndexRequest(arvLogger, keepServer)
220 resp, err := client.Do(req)
222 loggerutil.FatalWithMessage(arvLogger,
223 fmt.Sprintf("Error fetching %s: %v. Response was %+v",
229 return ReadServerResponse(arvLogger, keepServer, resp)
232 func GetServerStatus(arvLogger *logger.Logger,
233 keepServer ServerAddress,
234 client http.Client) {
235 url := fmt.Sprintf("http://%s:%d/status.json",
239 if arvLogger != nil {
241 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
242 keepInfo := logger.GetOrCreateMap(p, "keep_info")
243 serverInfo := make(map[string]interface{})
244 serverInfo["status_request_sent_at"] = now
245 serverInfo["host"] = keepServer.Host
246 serverInfo["port"] = keepServer.Port
248 keepInfo[keepServer.Uuid] = serverInfo
252 resp, err := client.Get(url)
254 loggerutil.FatalWithMessage(arvLogger,
255 fmt.Sprintf("Error getting keep status from %s: %v", url, err))
256 } else if resp.StatusCode != 200 {
257 loggerutil.FatalWithMessage(arvLogger,
258 fmt.Sprintf("Received error code %d in response to request "+
260 resp.StatusCode, url, resp.Status))
263 var keepStatus map[string]interface{}
264 decoder := json.NewDecoder(resp.Body)
266 err = decoder.Decode(&keepStatus)
268 loggerutil.FatalWithMessage(arvLogger,
269 fmt.Sprintf("Error decoding keep status from %s: %v", url, err))
272 if arvLogger != nil {
274 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
275 keepInfo := logger.GetOrCreateMap(p, "keep_info")
276 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
277 serverInfo["status_response_processed_at"] = now
278 serverInfo["status"] = keepStatus
283 func CreateIndexRequest(arvLogger *logger.Logger,
284 keepServer ServerAddress) (req *http.Request) {
285 url := fmt.Sprintf("http://%s:%d/index", keepServer.Host, keepServer.Port)
286 log.Println("About to fetch keep server contents from " + url)
288 if arvLogger != nil {
290 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
291 keepInfo := logger.GetOrCreateMap(p, "keep_info")
292 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
293 serverInfo["index_request_sent_at"] = now
297 req, err := http.NewRequest("GET", url, nil)
299 loggerutil.FatalWithMessage(arvLogger,
300 fmt.Sprintf("Error building http request for %s: %v", url, err))
303 req.Header.Add("Authorization",
304 fmt.Sprintf("OAuth2 %s", getDataManagerToken(arvLogger)))
308 func ReadServerResponse(arvLogger *logger.Logger,
309 keepServer ServerAddress,
310 resp *http.Response) (response ServerResponse) {
312 if resp.StatusCode != 200 {
313 loggerutil.FatalWithMessage(arvLogger,
314 fmt.Sprintf("Received error code %d in response to request "+
316 resp.StatusCode, keepServer.String(), resp.Status))
319 if arvLogger != nil {
321 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
322 keepInfo := logger.GetOrCreateMap(p, "keep_info")
323 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
324 serverInfo["index_response_received_at"] = now
328 response.Address = keepServer
329 response.Contents.BlockDigestToInfo =
330 make(map[blockdigest.DigestWithSize]BlockInfo)
331 scanner := bufio.NewScanner(resp.Body)
332 numLines, numDuplicates, numSizeDisagreements := 0, 0, 0
335 blockInfo, err := parseBlockInfoFromIndexLine(scanner.Text())
337 loggerutil.FatalWithMessage(arvLogger,
338 fmt.Sprintf("Error parsing BlockInfo from index line "+
339 "received from %s: %v",
344 if storedBlock, ok := response.Contents.BlockDigestToInfo[blockInfo.Digest]; ok {
345 // This server returned multiple lines containing the same block digest.
347 // Keep the block that's newer.
348 if storedBlock.Mtime < blockInfo.Mtime {
349 response.Contents.BlockDigestToInfo[blockInfo.Digest] = blockInfo
352 response.Contents.BlockDigestToInfo[blockInfo.Digest] = blockInfo
355 if err := scanner.Err(); err != nil {
356 loggerutil.FatalWithMessage(arvLogger,
357 fmt.Sprintf("Received error scanning index response from %s: %v",
361 log.Printf("%s index contained %d lines with %d duplicates with "+
362 "%d size disagreements",
366 numSizeDisagreements)
368 if arvLogger != nil {
370 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
371 keepInfo := logger.GetOrCreateMap(p, "keep_info")
372 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
374 serverInfo["processing_finished_at"] = now
375 serverInfo["lines_received"] = numLines
376 serverInfo["duplicates_seen"] = numDuplicates
377 serverInfo["size_disagreements_seen"] = numSizeDisagreements
385 func parseBlockInfoFromIndexLine(indexLine string) (blockInfo BlockInfo, err error) {
386 tokens := strings.Fields(indexLine)
387 if len(tokens) != 2 {
388 err = fmt.Errorf("Expected 2 tokens per line but received a "+
389 "line containing %v instead.",
393 var locator blockdigest.BlockLocator
394 if locator, err = blockdigest.ParseBlockLocator(tokens[0]); err != nil {
395 err = fmt.Errorf("%v Received error while parsing line \"%s\"",
399 if len(locator.Hints) > 0 {
400 err = fmt.Errorf("Block locator in index line should not contain hints "+
406 blockInfo.Mtime, err = strconv.ParseInt(tokens[1], 10, 64)
411 blockdigest.DigestWithSize{Digest: locator.Digest,
412 Size: uint32(locator.Size)}
416 func (readServers *ReadServers) Summarize(arvLogger *logger.Logger) {
417 readServers.BlockReplicationCounts = make(map[int]int)
418 for _, infos := range readServers.BlockToServers {
419 replication := len(infos)
420 readServers.BlockReplicationCounts[replication] += 1
423 if arvLogger != nil {
424 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
425 keepInfo := logger.GetOrCreateMap(p, "keep_info")
426 keepInfo["distinct_blocks_stored"] = len(readServers.BlockToServers)