1 /* Deals with getting Keep Server blocks from API Server and Keep Servers. */
11 "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
12 "git.curoverse.com/arvados.git/sdk/go/blockdigest"
13 "git.curoverse.com/arvados.git/sdk/go/keepclient"
14 "git.curoverse.com/arvados.git/sdk/go/logger"
15 "git.curoverse.com/arvados.git/services/datamanager/loggerutil"
26 type ServerAddress struct {
27 SSL bool `json:service_ssl_flag`
28 Host string `json:"service_host"`
29 Port int `json:"service_port"`
30 Uuid string `json:"uuid"`
33 // Info about a particular block returned by the server
34 type BlockInfo struct {
35 Digest blockdigest.DigestWithSize
36 Mtime int64 // TODO(misha): Replace this with a timestamp.
39 // Info about a specified block given by a server
40 type BlockServerInfo struct {
42 Mtime int64 // TODO(misha): Replace this with a timestamp.
45 type ServerContents struct {
46 BlockDigestToInfo map[blockdigest.DigestWithSize]BlockInfo
49 type ServerResponse struct {
51 Contents ServerContents
54 type ReadServers struct {
56 KeepServerIndexToAddress []ServerAddress
57 KeepServerAddressToIndex map[ServerAddress]int
58 ServerToContents map[ServerAddress]ServerContents
59 BlockToServers map[blockdigest.DigestWithSize][]BlockServerInfo
60 BlockReplicationCounts map[int]int
63 type GetKeepServersParams struct {
64 Client arvadosclient.ArvadosClient
69 type KeepServiceList struct {
70 ItemsAvailable int `json:"items_available"`
71 KeepServers []ServerAddress `json:"items"`
75 // Don't access the token directly, use getDataManagerToken() to
76 // make sure it's been read.
77 dataManagerToken string
78 dataManagerTokenFile string
79 dataManagerTokenFileReadOnce sync.Once
83 flag.StringVar(&dataManagerTokenFile,
84 "data-manager-token-file",
86 "File with the API token we should use to contact keep servers.")
89 // TODO(misha): Change this to include the UUID as well.
90 func (s ServerAddress) String() string {
94 func (s ServerAddress) URL() string {
96 return fmt.Sprintf("https://%s:%d", s.Host, s.Port)
98 return fmt.Sprintf("http://%s:%d", s.Host, s.Port)
102 func GetDataManagerToken(arvLogger *logger.Logger) string {
103 readDataManagerToken := func() {
104 if dataManagerTokenFile == "" {
106 loggerutil.FatalWithMessage(arvLogger,
107 "Data Manager Token needed, but data manager token file not specified.")
109 rawRead, err := ioutil.ReadFile(dataManagerTokenFile)
111 loggerutil.FatalWithMessage(arvLogger,
112 fmt.Sprintf("Unexpected error reading token file %s: %v",
113 dataManagerTokenFile,
116 dataManagerToken = strings.TrimSpace(string(rawRead))
120 dataManagerTokenFileReadOnce.Do(readDataManagerToken)
121 return dataManagerToken
124 func GetKeepServersAndSummarize(params GetKeepServersParams) (results ReadServers) {
125 results = GetKeepServers(params)
126 log.Printf("Returned %d keep disks", len(results.ServerToContents))
128 results.Summarize(params.Logger)
129 log.Printf("Replication level distribution: %v",
130 results.BlockReplicationCounts)
135 func GetKeepServers(params GetKeepServersParams) (results ReadServers) {
136 if ¶ms.Client == nil {
137 log.Fatalf("params.Client passed to GetKeepServers() should " +
138 "contain a valid ArvadosClient, but instead it is nil.")
141 sdkParams := arvadosclient.Dict{
142 "filters": [][]string{[]string{"service_type", "=", "disk"}},
144 if params.Limit > 0 {
145 sdkParams["limit"] = params.Limit
148 var sdkResponse KeepServiceList
149 err := params.Client.List("keep_services", sdkParams, &sdkResponse)
152 loggerutil.FatalWithMessage(params.Logger,
153 fmt.Sprintf("Error requesting keep disks from API server: %v", err))
156 if params.Logger != nil {
157 params.Logger.Update(func(p map[string]interface{}, e map[string]interface{}) {
158 keepInfo := logger.GetOrCreateMap(p, "keep_info")
159 keepInfo["num_keep_servers_available"] = sdkResponse.ItemsAvailable
160 keepInfo["num_keep_servers_received"] = len(sdkResponse.KeepServers)
161 keepInfo["keep_servers"] = sdkResponse.KeepServers
165 log.Printf("Received keep services list: %+v", sdkResponse)
167 if len(sdkResponse.KeepServers) < sdkResponse.ItemsAvailable {
168 loggerutil.FatalWithMessage(params.Logger,
169 fmt.Sprintf("Did not receive all available keep servers: %+v", sdkResponse))
172 results.KeepServerIndexToAddress = sdkResponse.KeepServers
173 results.KeepServerAddressToIndex = make(map[ServerAddress]int)
174 for i, address := range results.KeepServerIndexToAddress {
175 results.KeepServerAddressToIndex[address] = i
178 log.Printf("Got Server Addresses: %v", results)
180 // This is safe for concurrent use
181 client := http.Client{}
183 // Send off all the index requests concurrently
184 responseChan := make(chan ServerResponse)
185 for _, keepServer := range sdkResponse.KeepServers {
186 // The above keepsServer variable is reused for each iteration, so
187 // it would be shared across all goroutines. This would result in
188 // us querying one server n times instead of n different servers
189 // as we intended. To avoid this we add it as an explicit
190 // parameter which gets copied. This bug and solution is described
191 // in https://golang.org/doc/effective_go.html#channels
192 go func(keepServer ServerAddress) {
193 responseChan <- GetServerContents(params.Logger,
199 results.ServerToContents = make(map[ServerAddress]ServerContents)
200 results.BlockToServers = make(map[blockdigest.DigestWithSize][]BlockServerInfo)
202 // Read all the responses
203 for i := range sdkResponse.KeepServers {
204 _ = i // Here to prevent go from complaining.
205 response := <-responseChan
206 log.Printf("Received channel response from %v containing %d files",
208 len(response.Contents.BlockDigestToInfo))
209 results.ServerToContents[response.Address] = response.Contents
210 serverIndex := results.KeepServerAddressToIndex[response.Address]
211 for _, blockInfo := range response.Contents.BlockDigestToInfo {
212 results.BlockToServers[blockInfo.Digest] = append(
213 results.BlockToServers[blockInfo.Digest],
214 BlockServerInfo{ServerIndex: serverIndex,
215 Mtime: blockInfo.Mtime})
221 func GetServerContents(arvLogger *logger.Logger,
222 keepServer ServerAddress,
223 client http.Client) (response ServerResponse) {
225 GetServerStatus(arvLogger, keepServer, client)
227 req := CreateIndexRequest(arvLogger, keepServer)
228 resp, err := client.Do(req)
230 loggerutil.FatalWithMessage(arvLogger,
231 fmt.Sprintf("Error fetching %s: %v. Response was %+v",
237 return ReadServerResponse(arvLogger, keepServer, resp)
240 func GetServerStatus(arvLogger *logger.Logger,
241 keepServer ServerAddress,
242 client http.Client) {
243 url := fmt.Sprintf("http://%s:%d/status.json",
247 if arvLogger != nil {
249 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
250 keepInfo := logger.GetOrCreateMap(p, "keep_info")
251 serverInfo := make(map[string]interface{})
252 serverInfo["status_request_sent_at"] = now
253 serverInfo["host"] = keepServer.Host
254 serverInfo["port"] = keepServer.Port
256 keepInfo[keepServer.Uuid] = serverInfo
260 resp, err := client.Get(url)
262 loggerutil.FatalWithMessage(arvLogger,
263 fmt.Sprintf("Error getting keep status from %s: %v", url, err))
264 } else if resp.StatusCode != 200 {
265 loggerutil.FatalWithMessage(arvLogger,
266 fmt.Sprintf("Received error code %d in response to request "+
268 resp.StatusCode, url, resp.Status))
271 var keepStatus map[string]interface{}
272 decoder := json.NewDecoder(resp.Body)
274 err = decoder.Decode(&keepStatus)
276 loggerutil.FatalWithMessage(arvLogger,
277 fmt.Sprintf("Error decoding keep status from %s: %v", url, err))
280 if arvLogger != nil {
282 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
283 keepInfo := logger.GetOrCreateMap(p, "keep_info")
284 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
285 serverInfo["status_response_processed_at"] = now
286 serverInfo["status"] = keepStatus
291 func CreateIndexRequest(arvLogger *logger.Logger,
292 keepServer ServerAddress) (req *http.Request) {
293 url := fmt.Sprintf("http://%s:%d/index", keepServer.Host, keepServer.Port)
294 log.Println("About to fetch keep server contents from " + url)
296 if arvLogger != nil {
298 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
299 keepInfo := logger.GetOrCreateMap(p, "keep_info")
300 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
301 serverInfo["index_request_sent_at"] = now
305 req, err := http.NewRequest("GET", url, nil)
307 loggerutil.FatalWithMessage(arvLogger,
308 fmt.Sprintf("Error building http request for %s: %v", url, err))
311 req.Header.Add("Authorization",
312 fmt.Sprintf("OAuth2 %s", GetDataManagerToken(arvLogger)))
316 func ReadServerResponse(arvLogger *logger.Logger,
317 keepServer ServerAddress,
318 resp *http.Response) (response ServerResponse) {
320 if resp.StatusCode != 200 {
321 loggerutil.FatalWithMessage(arvLogger,
322 fmt.Sprintf("Received error code %d in response to request "+
324 resp.StatusCode, keepServer.String(), resp.Status))
327 if arvLogger != nil {
329 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
330 keepInfo := logger.GetOrCreateMap(p, "keep_info")
331 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
332 serverInfo["index_response_received_at"] = now
336 response.Address = keepServer
337 response.Contents.BlockDigestToInfo =
338 make(map[blockdigest.DigestWithSize]BlockInfo)
339 reader := bufio.NewReader(resp.Body)
340 numLines, numDuplicates, numSizeDisagreements := 0, 0, 0
343 line, err := reader.ReadString('\n')
345 loggerutil.FatalWithMessage(arvLogger,
346 fmt.Sprintf("Index from %s truncated at line %d",
347 keepServer.String(), numLines))
348 } else if err != nil {
349 loggerutil.FatalWithMessage(arvLogger,
350 fmt.Sprintf("Error reading index response from %s at line %d: %v",
351 keepServer.String(), numLines, err))
354 if _, err := reader.Peek(1); err == nil {
355 extra, _ := reader.ReadString('\n')
356 loggerutil.FatalWithMessage(arvLogger,
357 fmt.Sprintf("Index from %s had trailing data at line %d after EOF marker: %s",
358 keepServer.String(), numLines+1, extra))
359 } else if err != io.EOF {
360 loggerutil.FatalWithMessage(arvLogger,
361 fmt.Sprintf("Index from %s had read error after EOF marker at line %d: %v",
362 keepServer.String(), numLines, err))
367 blockInfo, err := parseBlockInfoFromIndexLine(line)
369 loggerutil.FatalWithMessage(arvLogger,
370 fmt.Sprintf("Error parsing BlockInfo from index line "+
371 "received from %s: %v",
376 if storedBlock, ok := response.Contents.BlockDigestToInfo[blockInfo.Digest]; ok {
377 // This server returned multiple lines containing the same block digest.
379 // Keep the block that's newer.
380 if storedBlock.Mtime < blockInfo.Mtime {
381 response.Contents.BlockDigestToInfo[blockInfo.Digest] = blockInfo
384 response.Contents.BlockDigestToInfo[blockInfo.Digest] = blockInfo
388 log.Printf("%s index contained %d lines with %d duplicates with "+
389 "%d size disagreements",
393 numSizeDisagreements)
395 if arvLogger != nil {
397 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
398 keepInfo := logger.GetOrCreateMap(p, "keep_info")
399 serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
401 serverInfo["processing_finished_at"] = now
402 serverInfo["lines_received"] = numLines
403 serverInfo["duplicates_seen"] = numDuplicates
404 serverInfo["size_disagreements_seen"] = numSizeDisagreements
411 func parseBlockInfoFromIndexLine(indexLine string) (blockInfo BlockInfo, err error) {
412 tokens := strings.Fields(indexLine)
413 if len(tokens) != 2 {
414 err = fmt.Errorf("Expected 2 tokens per line but received a "+
415 "line containing %v instead.",
419 var locator blockdigest.BlockLocator
420 if locator, err = blockdigest.ParseBlockLocator(tokens[0]); err != nil {
421 err = fmt.Errorf("%v Received error while parsing line \"%s\"",
425 if len(locator.Hints) > 0 {
426 err = fmt.Errorf("Block locator in index line should not contain hints "+
432 blockInfo.Mtime, err = strconv.ParseInt(tokens[1], 10, 64)
437 blockdigest.DigestWithSize{Digest: locator.Digest,
438 Size: uint32(locator.Size)}
442 func (readServers *ReadServers) Summarize(arvLogger *logger.Logger) {
443 readServers.BlockReplicationCounts = make(map[int]int)
444 for _, infos := range readServers.BlockToServers {
445 replication := len(infos)
446 readServers.BlockReplicationCounts[replication] += 1
449 if arvLogger != nil {
450 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
451 keepInfo := logger.GetOrCreateMap(p, "keep_info")
452 keepInfo["distinct_blocks_stored"] = len(readServers.BlockToServers)
458 type TrashRequest struct {
459 Locator string `json:"locator"`
460 BlockMtime int64 `json:"block_mtime"`
463 type TrashList []TrashRequest
465 func SendTrashLists(dataManagerToken string, kc *keepclient.KeepClient, spl map[string]TrashList) (errs []error) {
467 barrier := make(chan error)
471 for url, v := range spl {
473 log.Printf("Sending trash list to %v", url)
475 go (func(url string, v TrashList) {
476 pipeReader, pipeWriter := io.Pipe()
478 enc := json.NewEncoder(pipeWriter)
483 req, err := http.NewRequest("PUT", fmt.Sprintf("%s/trash", url), pipeReader)
485 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", dataManagerToken))
494 var resp *http.Response
495 if resp, err = client.Do(req); err != nil {
496 log.Printf("Error sending trash list to %v error: %v", url, err.Error())
501 log.Printf("Sent trash list to %v: response was HTTP %v", url, resp.Status)
503 io.Copy(ioutil.Discard, resp.Body)
506 if resp.StatusCode != 200 {
507 barrier <- errors.New(fmt.Sprintf("Got HTTP code %v", resp.StatusCode))
515 for i := 0; i < count; i += 1 {
518 errs = append(errs, b)