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/keepclient"
13 "git.curoverse.com/arvados.git/sdk/go/logger"
14 "git.curoverse.com/arvados.git/services/datamanager/loggerutil"
25 type ServerAddress struct {
26 SSL bool `json:service_ssl_flag`
27 Host string `json:"service_host"`
28 Port int `json:"service_port"`
29 Uuid string `json:"uuid"`
32 // Info about a particular block returned by the server
33 type BlockInfo struct {
34 Digest blockdigest.DigestWithSize
35 Mtime int64 // TODO(misha): Replace this with a timestamp.
38 // Info about a specified block given by a server
39 type BlockServerInfo struct {
41 Mtime int64 // TODO(misha): Replace this with a timestamp.
44 type ServerContents struct {
45 BlockDigestToInfo map[blockdigest.DigestWithSize]BlockInfo
48 type ServerResponse struct {
50 Contents ServerContents
53 type ReadServers struct {
55 KeepServerIndexToAddress []ServerAddress
56 KeepServerAddressToIndex map[ServerAddress]int
57 ServerToContents map[ServerAddress]ServerContents
58 BlockToServers map[blockdigest.DigestWithSize][]BlockServerInfo
59 BlockReplicationCounts map[int]int
62 type GetKeepServersParams struct {
63 Client arvadosclient.ArvadosClient
68 type KeepServiceList struct {
69 ItemsAvailable int `json:"items_available"`
70 KeepServers []ServerAddress `json:"items"`
74 // Don't access the token directly, use getDataManagerToken() to
75 // make sure it's been read.
76 dataManagerToken string
77 dataManagerTokenFile string
78 dataManagerTokenFileReadOnce sync.Once
82 flag.StringVar(&dataManagerTokenFile,
83 "data-manager-token-file",
85 "File with the API token we should use to contact keep servers.")
88 // TODO(misha): Change this to include the UUID as well.
89 func (s ServerAddress) String() string {
93 func (s ServerAddress) URL() string {
95 return fmt.Sprintf("https://%s:%d", s.Host, s.Port)
97 return fmt.Sprintf("http://%s:%d", s.Host, s.Port)
101 func getDataManagerToken(arvLogger *logger.Logger) string {
102 readDataManagerToken := func() {
103 if dataManagerTokenFile == "" {
105 loggerutil.FatalWithMessage(arvLogger,
106 "Data Manager Token needed, but data manager token file not specified.")
108 rawRead, err := ioutil.ReadFile(dataManagerTokenFile)
110 loggerutil.FatalWithMessage(arvLogger,
111 fmt.Sprintf("Unexpected error reading token file %s: %v",
112 dataManagerTokenFile,
115 dataManagerToken = strings.TrimSpace(string(rawRead))
119 dataManagerTokenFileReadOnce.Do(readDataManagerToken)
120 return dataManagerToken
123 func GetKeepServersAndSummarize(params GetKeepServersParams) (results ReadServers) {
124 results = GetKeepServers(params)
125 log.Printf("Returned %d keep disks", len(results.ServerToContents))
127 results.Summarize(params.Logger)
128 log.Printf("Replication level distribution: %v",
129 results.BlockReplicationCounts)
134 func GetKeepServers(params GetKeepServersParams) (results ReadServers) {
135 if ¶ms.Client == nil {
136 log.Fatalf("params.Client passed to GetKeepServers() should " +
137 "contain a valid ArvadosClient, but instead it is nil.")
140 sdkParams := arvadosclient.Dict{
141 "filters": [][]string{[]string{"service_type", "=", "disk"}},
143 if params.Limit > 0 {
144 sdkParams["limit"] = params.Limit
147 var sdkResponse KeepServiceList
148 err := params.Client.List("keep_services", sdkParams, &sdkResponse)
151 loggerutil.FatalWithMessage(params.Logger,
152 fmt.Sprintf("Error requesting keep disks from API server: %v", err))
155 if params.Logger != nil {
156 params.Logger.Update(func(p map[string]interface{}, e map[string]interface{}) {
157 keepInfo := logger.GetOrCreateMap(p, "keep_info")
158 keepInfo["num_keep_servers_available"] = sdkResponse.ItemsAvailable
159 keepInfo["num_keep_servers_received"] = len(sdkResponse.KeepServers)
160 keepInfo["keep_servers"] = sdkResponse.KeepServers
164 log.Printf("Received keep services list: %+v", sdkResponse)
166 if len(sdkResponse.KeepServers) < sdkResponse.ItemsAvailable {
167 loggerutil.FatalWithMessage(params.Logger,
168 fmt.Sprintf("Did not receive all available keep servers: %+v", sdkResponse))
171 results.KeepServerIndexToAddress = sdkResponse.KeepServers
172 results.KeepServerAddressToIndex = make(map[ServerAddress]int)
173 for i, address := range results.KeepServerIndexToAddress {
174 results.KeepServerAddressToIndex[address] = i
177 log.Printf("Got Server Addresses: %v", results)
179 // This is safe for concurrent use
180 client := http.Client{}
182 // Send off all the index requests concurrently
183 responseChan := make(chan ServerResponse)
184 for _, keepServer := range sdkResponse.KeepServers {
185 // The above keepsServer variable is reused for each iteration, so
186 // it would be shared across all goroutines. This would result in
187 // us querying one server n times instead of n different servers
188 // as we intended. To avoid this we add it as an explicit
189 // parameter which gets copied. This bug and solution is described
190 // in https://golang.org/doc/effective_go.html#channels
191 go func(keepServer ServerAddress) {
192 responseChan <- GetServerContents(params.Logger,
198 results.ServerToContents = make(map[ServerAddress]ServerContents)
199 results.BlockToServers = make(map[blockdigest.DigestWithSize][]BlockServerInfo)
201 // Read all the responses
202 for i := range sdkResponse.KeepServers {
203 _ = i // Here to prevent go from complaining.
204 response := <-responseChan
205 log.Printf("Received channel response from %v containing %d files",
207 len(response.Contents.BlockDigestToInfo))
208 results.ServerToContents[response.Address] = response.Contents
209 serverIndex := results.KeepServerAddressToIndex[response.Address]
210 for _, blockInfo := range response.Contents.BlockDigestToInfo {
211 results.BlockToServers[blockInfo.Digest] = append(
212 results.BlockToServers[blockInfo.Digest],
213 BlockServerInfo{ServerIndex: serverIndex,
214 Mtime: blockInfo.Mtime})
220 func GetServerContents(arvLogger *logger.Logger,
221 keepServer ServerAddress,
222 client http.Client) (response ServerResponse) {
224 GetServerStatus(arvLogger, keepServer, client)
226 req := CreateIndexRequest(arvLogger, keepServer)
227 resp, err := client.Do(req)
229 loggerutil.FatalWithMessage(arvLogger,
230 fmt.Sprintf("Error fetching %s: %v. Response was %+v",
236 return ReadServerResponse(arvLogger, keepServer, resp)
239 func GetServerStatus(arvLogger *logger.Logger,
240 keepServer ServerAddress,
241 client http.Client) {
242 url := fmt.Sprintf("http://%s:%d/status.json",
246 if arvLogger != nil {
248 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
249 keepInfo := logger.GetOrCreateMap(p, "keep_info")
250 serverInfo := make(map[string]interface{})
251 serverInfo["status_request_sent_at"] = now
252 serverInfo["host"] = keepServer.Host
253 serverInfo["port"] = keepServer.Port
255 keepInfo[keepServer.Uuid] = serverInfo
259 resp, err := client.Get(url)
261 loggerutil.FatalWithMessage(arvLogger,
262 fmt.Sprintf("Error getting keep status from %s: %v", url, err))
263 } else if resp.StatusCode != 200 {
264 loggerutil.FatalWithMessage(arvLogger,
265 fmt.Sprintf("Received error code %d in response to request "+
267 resp.StatusCode, url, resp.Status))
270 var keepStatus map[string]interface{}
271 decoder := json.NewDecoder(resp.Body)
273 err = decoder.Decode(&keepStatus)
275 loggerutil.FatalWithMessage(arvLogger,
276 fmt.Sprintf("Error decoding keep status from %s: %v", url, err))
279 if arvLogger != nil {
281 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
282 keepInfo := logger.GetOrCreateMap(p, "keep_info")
283 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
284 serverInfo["status_response_processed_at"] = now
285 serverInfo["status"] = keepStatus
290 func CreateIndexRequest(arvLogger *logger.Logger,
291 keepServer ServerAddress) (req *http.Request) {
292 url := fmt.Sprintf("http://%s:%d/index", keepServer.Host, keepServer.Port)
293 log.Println("About to fetch keep server contents from " + url)
295 if arvLogger != nil {
297 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
298 keepInfo := logger.GetOrCreateMap(p, "keep_info")
299 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
300 serverInfo["index_request_sent_at"] = now
304 req, err := http.NewRequest("GET", url, nil)
306 loggerutil.FatalWithMessage(arvLogger,
307 fmt.Sprintf("Error building http request for %s: %v", url, err))
310 req.Header.Add("Authorization",
311 fmt.Sprintf("OAuth2 %s", getDataManagerToken(arvLogger)))
315 func ReadServerResponse(arvLogger *logger.Logger,
316 keepServer ServerAddress,
317 resp *http.Response) (response ServerResponse) {
319 if resp.StatusCode != 200 {
320 loggerutil.FatalWithMessage(arvLogger,
321 fmt.Sprintf("Received error code %d in response to request "+
323 resp.StatusCode, keepServer.String(), resp.Status))
326 if arvLogger != nil {
328 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
329 keepInfo := logger.GetOrCreateMap(p, "keep_info")
330 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
331 serverInfo["index_response_received_at"] = now
335 response.Address = keepServer
336 response.Contents.BlockDigestToInfo =
337 make(map[blockdigest.DigestWithSize]BlockInfo)
338 reader := bufio.NewReader(resp.Body)
339 numLines, numDuplicates, numSizeDisagreements := 0, 0, 0
342 line, err := reader.ReadString('\n')
344 loggerutil.FatalWithMessage(arvLogger,
345 fmt.Sprintf("Index from %s truncated at line %d",
346 keepServer.String(), numLines))
347 } else if err != nil {
348 loggerutil.FatalWithMessage(arvLogger,
349 fmt.Sprintf("Error reading index response from %s at line %d: %v",
350 keepServer.String(), numLines, err))
353 if _, err := reader.Peek(1); err == nil {
354 extra, _ := reader.ReadString('\n')
355 loggerutil.FatalWithMessage(arvLogger,
356 fmt.Sprintf("Index from %s had trailing data at line %d after EOF marker: %s",
357 keepServer.String(), numLines+1, extra))
358 } else if err != io.EOF {
359 loggerutil.FatalWithMessage(arvLogger,
360 fmt.Sprintf("Index from %s had read error after EOF marker at line %d: %v",
361 keepServer.String(), numLines, err))
366 blockInfo, err := parseBlockInfoFromIndexLine(line)
368 loggerutil.FatalWithMessage(arvLogger,
369 fmt.Sprintf("Error parsing BlockInfo from index line "+
370 "received from %s: %v",
375 if storedBlock, ok := response.Contents.BlockDigestToInfo[blockInfo.Digest]; ok {
376 // This server returned multiple lines containing the same block digest.
378 // Keep the block that's newer.
379 if storedBlock.Mtime < blockInfo.Mtime {
380 response.Contents.BlockDigestToInfo[blockInfo.Digest] = blockInfo
383 response.Contents.BlockDigestToInfo[blockInfo.Digest] = blockInfo
387 log.Printf("%s index contained %d lines with %d duplicates with "+
388 "%d size disagreements",
392 numSizeDisagreements)
394 if arvLogger != nil {
396 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
397 keepInfo := logger.GetOrCreateMap(p, "keep_info")
398 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
400 serverInfo["processing_finished_at"] = now
401 serverInfo["lines_received"] = numLines
402 serverInfo["duplicates_seen"] = numDuplicates
403 serverInfo["size_disagreements_seen"] = numSizeDisagreements
410 func parseBlockInfoFromIndexLine(indexLine string) (blockInfo BlockInfo, err error) {
411 tokens := strings.Fields(indexLine)
412 if len(tokens) != 2 {
413 err = fmt.Errorf("Expected 2 tokens per line but received a "+
414 "line containing %v instead.",
418 var locator blockdigest.BlockLocator
419 if locator, err = blockdigest.ParseBlockLocator(tokens[0]); err != nil {
420 err = fmt.Errorf("%v Received error while parsing line \"%s\"",
424 if len(locator.Hints) > 0 {
425 err = fmt.Errorf("Block locator in index line should not contain hints "+
431 blockInfo.Mtime, err = strconv.ParseInt(tokens[1], 10, 64)
436 blockdigest.DigestWithSize{Digest: locator.Digest,
437 Size: uint32(locator.Size)}
441 func (readServers *ReadServers) Summarize(arvLogger *logger.Logger) {
442 readServers.BlockReplicationCounts = make(map[int]int)
443 for _, infos := range readServers.BlockToServers {
444 replication := len(infos)
445 readServers.BlockReplicationCounts[replication] += 1
448 if arvLogger != nil {
449 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
450 keepInfo := logger.GetOrCreateMap(p, "keep_info")
451 keepInfo["distinct_blocks_stored"] = len(readServers.BlockToServers)
457 type TrashRequest struct {
458 Locator string `json:"locator"`
459 BlockMtime int64 `json:"block_mtime"`
462 type TrashList []TrashRequest
464 func SendTrashLists(arvLogger *logger.Logger, kc *keepclient.KeepClient, spl map[string]TrashList) {
466 rendezvous := make(chan bool)
468 for url, v := range spl {
470 log.Printf("Sending trash list to %v", url)
472 go (func(url string, v TrashList) {
477 pipeReader, pipeWriter := io.Pipe()
479 enc := json.NewEncoder(pipeWriter)
484 req, err := http.NewRequest("PUT", fmt.Sprintf("%s/trash", url), pipeReader)
486 log.Printf("Error creating trash list request for %v error: %v", url, err.Error())
490 // Add api token header
491 req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", getDataManagerToken(arvLogger)))
494 var resp *http.Response
495 if resp, err = kc.Client.Do(req); err != nil {
496 log.Printf("Error sending trash list to %v error: %v", url, err.Error())
500 log.Printf("Sent trash list to %v: response was HTTP %d", url, resp.Status)
502 io.Copy(ioutil.Discard, resp.Body)
508 for i := 0; i < count; i += 1 {