services/datamanager/keep/keep.go

   1 /* Deals with getting Keep Server blocks from API Server and Keep Servers. */
   2
   3 package keep
   4
   5 import (
   6         "bufio"
   7         "encoding/json"
   8         "errors"
   9         "flag"
  10         "fmt"
  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         "io"
  16         "io/ioutil"
  17         "log"
  18         "net/http"
  19         "strconv"
  20         "strings"
  21         "time"
  22 )
  23
  24 // ServerAddress struct
  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"`
  30         ServiceType string `json:"service_type"`
  31 }
  32
  33 // BlockInfo is 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.
  37 }
  38
  39 // BlockServerInfo is info about a specified block given by a server
  40 type BlockServerInfo struct {
  41         ServerIndex int
  42         Mtime       int64 // TODO(misha): Replace this with a timestamp.
  43 }
  44
  45 // ServerContents struct
  46 type ServerContents struct {
  47         BlockDigestToInfo map[blockdigest.DigestWithSize]BlockInfo
  48 }
  49
  50 // ServerResponse struct
  51 type ServerResponse struct {
  52         Address  ServerAddress
  53         Contents ServerContents
  54         Err      error
  55 }
  56
  57 // ReadServers struct
  58 type ReadServers struct {
  59         ReadAllServers           bool
  60         KeepServerIndexToAddress []ServerAddress
  61         KeepServerAddressToIndex map[ServerAddress]int
  62         ServerToContents         map[ServerAddress]ServerContents
  63         BlockToServers           map[blockdigest.DigestWithSize][]BlockServerInfo
  64         BlockReplicationCounts   map[int]int
  65 }
  66
  67 // GetKeepServersParams struct
  68 type GetKeepServersParams struct {
  69         Client arvadosclient.ArvadosClient
  70         Logger *logger.Logger
  71         Limit  int
  72 }
  73
  74 // ServiceList consists of the addresses of all the available kee servers
  75 type ServiceList struct {
  76         ItemsAvailable int             `json:"items_available"`
  77         KeepServers    []ServerAddress `json:"items"`
  78 }
  79
  80 var serviceType string
  81
  82 func init() {
  83         flag.StringVar(&serviceType,
  84                 "service-type",
  85                 "disk",
  86                 "Operate only on keep_services with the specified service_type, ignoring all others.")
  87 }
  88
  89 // String
  90 // TODO(misha): Change this to include the UUID as well.
  91 func (s ServerAddress) String() string {
  92         return s.URL()
  93 }
  94
  95 // URL of the keep server
  96 func (s ServerAddress) URL() string {
  97         if s.SSL {
  98                 return fmt.Sprintf("https://%s:%d", s.Host, s.Port)
  99         }
 100         return fmt.Sprintf("http://%s:%d", s.Host, s.Port)
 101 }
 102
 103 // GetKeepServersAndSummarize gets keep servers from api
 104 func GetKeepServersAndSummarize(params GetKeepServersParams) (results ReadServers, err error) {
 105         results, err = GetKeepServers(params)
 106         if err != nil {
 107                 return
 108         }
 109         log.Printf("Returned %d keep disks", len(results.ServerToContents))
 110
 111         results.Summarize(params.Logger)
 112         log.Printf("Replication level distribution: %v",
 113                 results.BlockReplicationCounts)
 114
 115         return
 116 }
 117
 118 // GetKeepServers from api server
 119 func GetKeepServers(params GetKeepServersParams) (results ReadServers, err error) {
 120         sdkParams := arvadosclient.Dict{
 121                 "filters": [][]string{[]string{"service_type", "!=", "proxy"}},
 122         }
 123         if params.Limit > 0 {
 124                 sdkParams["limit"] = params.Limit
 125         }
 126
 127         var sdkResponse ServiceList
 128         err = params.Client.List("keep_services", sdkParams, &sdkResponse)
 129
 130         if err != nil {
 131                 return
 132         }
 133
 134         var keepServers []ServerAddress
 135         for _, server := range sdkResponse.KeepServers {
 136                 if server.ServiceType == serviceType {
 137                         keepServers = append(keepServers, server)
 138                 } else {
 139                         log.Printf("Skipping keep_service %q because its service_type %q does not match -service-type=%q", server, server.ServiceType, serviceType)
 140                 }
 141         }
 142
 143         if len(keepServers) == 0 {
 144                 return results, fmt.Errorf("Found no keepservices with the service type %v", serviceType)
 145         }
 146
 147         if params.Logger != nil {
 148                 params.Logger.Update(func(p map[string]interface{}, e map[string]interface{}) {
 149                         keepInfo := logger.GetOrCreateMap(p, "keep_info")
 150                         keepInfo["num_keep_servers_available"] = sdkResponse.ItemsAvailable
 151                         keepInfo["num_keep_servers_received"] = len(sdkResponse.KeepServers)
 152                         keepInfo["keep_servers"] = sdkResponse.KeepServers
 153                         keepInfo["indexable_keep_servers"] = keepServers
 154                 })
 155         }
 156
 157         log.Printf("Received keep services list: %+v", sdkResponse)
 158
 159         if len(sdkResponse.KeepServers) < sdkResponse.ItemsAvailable {
 160                 return results, fmt.Errorf("Did not receive all available keep servers: %+v", sdkResponse)
 161         }
 162
 163         results.KeepServerIndexToAddress = keepServers
 164         results.KeepServerAddressToIndex = make(map[ServerAddress]int)
 165         for i, address := range results.KeepServerIndexToAddress {
 166                 results.KeepServerAddressToIndex[address] = i
 167         }
 168
 169         log.Printf("Got Server Addresses: %v", results)
 170
 171         // Send off all the index requests concurrently
 172         responseChan := make(chan ServerResponse)
 173         for _, keepServer := range results.KeepServerIndexToAddress {
 174                 // The above keepsServer variable is reused for each iteration, so
 175                 // it would be shared across all goroutines. This would result in
 176                 // us querying one server n times instead of n different servers
 177                 // as we intended. To avoid this we add it as an explicit
 178                 // parameter which gets copied. This bug and solution is described
 179                 // in https://golang.org/doc/effective_go.html#channels
 180                 go func(keepServer ServerAddress) {
 181                         responseChan <- GetServerContents(params.Logger,
 182                                 keepServer,
 183                                 params.Client)
 184                 }(keepServer)
 185         }
 186
 187         results.ServerToContents = make(map[ServerAddress]ServerContents)
 188         results.BlockToServers = make(map[blockdigest.DigestWithSize][]BlockServerInfo)
 189
 190         // Read all the responses
 191         for i := range results.KeepServerIndexToAddress {
 192                 _ = i // Here to prevent go from complaining.
 193                 response := <-responseChan
 194
 195                 // Check if there were any errors during GetServerContents
 196                 if response.Err != nil {
 197                         return results, response.Err
 198                 }
 199
 200                 log.Printf("Received channel response from %v containing %d files",
 201                         response.Address,
 202                         len(response.Contents.BlockDigestToInfo))
 203                 results.ServerToContents[response.Address] = response.Contents
 204                 serverIndex := results.KeepServerAddressToIndex[response.Address]
 205                 for _, blockInfo := range response.Contents.BlockDigestToInfo {
 206                         results.BlockToServers[blockInfo.Digest] = append(
 207                                 results.BlockToServers[blockInfo.Digest],
 208                                 BlockServerInfo{ServerIndex: serverIndex,
 209                                         Mtime: blockInfo.Mtime})
 210                 }
 211         }
 212         return
 213 }
 214
 215 // GetServerContents of the keep server
 216 func GetServerContents(arvLogger *logger.Logger,
 217         keepServer ServerAddress,
 218         arv arvadosclient.ArvadosClient) (response ServerResponse) {
 219
 220         err := GetServerStatus(arvLogger, keepServer, arv)
 221         if err != nil {
 222                 response.Err = err
 223                 return
 224         }
 225
 226         req, err := CreateIndexRequest(arvLogger, keepServer, arv)
 227         if err != nil {
 228                 response.Err = err
 229                 return
 230         }
 231
 232         resp, err := arv.Client.Do(req)
 233         if err != nil {
 234                 response.Err = err
 235                 return
 236         }
 237
 238         response, err = ReadServerResponse(arvLogger, keepServer, resp)
 239         if err != nil {
 240                 response.Err = err
 241                 return
 242         }
 243
 244         return
 245 }
 246
 247 // GetServerStatus get keep server status by invoking /status.json
 248 func GetServerStatus(arvLogger *logger.Logger,
 249         keepServer ServerAddress,
 250         arv arvadosclient.ArvadosClient) error {
 251         url := fmt.Sprintf("http://%s:%d/status.json",
 252                 keepServer.Host,
 253                 keepServer.Port)
 254
 255         if arvLogger != nil {
 256                 now := time.Now()
 257                 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
 258                         keepInfo := logger.GetOrCreateMap(p, "keep_info")
 259                         serverInfo := make(map[string]interface{})
 260                         serverInfo["status_request_sent_at"] = now
 261                         serverInfo["host"] = keepServer.Host
 262                         serverInfo["port"] = keepServer.Port
 263
 264                         keepInfo[keepServer.UUID] = serverInfo
 265                 })
 266         }
 267
 268         resp, err := arv.Client.Get(url)
 269         if err != nil {
 270                 return fmt.Errorf("Error getting keep status from %s: %v", url, err)
 271         } else if resp.StatusCode != 200 {
 272                 return fmt.Errorf("Received error code %d in response to request "+
 273                         "for %s status: %s",
 274                         resp.StatusCode, url, resp.Status)
 275         }
 276
 277         var keepStatus map[string]interface{}
 278         decoder := json.NewDecoder(resp.Body)
 279         decoder.UseNumber()
 280         err = decoder.Decode(&keepStatus)
 281         if err != nil {
 282                 return fmt.Errorf("Error decoding keep status from %s: %v", url, err)
 283         }
 284
 285         if arvLogger != nil {
 286                 now := time.Now()
 287                 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
 288                         keepInfo := logger.GetOrCreateMap(p, "keep_info")
 289                         serverInfo := keepInfo[keepServer.UUID].(map[string]interface{})
 290                         serverInfo["status_response_processed_at"] = now
 291                         serverInfo["status"] = keepStatus
 292                 })
 293         }
 294
 295         return nil
 296 }
 297
 298 // CreateIndexRequest to the keep server
 299 func CreateIndexRequest(arvLogger *logger.Logger,
 300         keepServer ServerAddress,
 301         arv arvadosclient.ArvadosClient) (req *http.Request, err error) {
 302         url := fmt.Sprintf("http://%s:%d/index", keepServer.Host, keepServer.Port)
 303         log.Println("About to fetch keep server contents from " + url)
 304
 305         if arvLogger != nil {
 306                 now := time.Now()
 307                 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
 308                         keepInfo := logger.GetOrCreateMap(p, "keep_info")
 309                         serverInfo := keepInfo[keepServer.UUID].(map[string]interface{})
 310                         serverInfo["index_request_sent_at"] = now
 311                 })
 312         }
 313
 314         req, err = http.NewRequest("GET", url, nil)
 315         if err != nil {
 316                 return req, fmt.Errorf("Error building http request for %s: %v", url, err)
 317         }
 318
 319         req.Header.Add("Authorization", "OAuth2 "+arv.ApiToken)
 320         return req, err
 321 }
 322
 323 // ReadServerResponse reads reasponse from keep server
 324 func ReadServerResponse(arvLogger *logger.Logger,
 325         keepServer ServerAddress,
 326         resp *http.Response) (response ServerResponse, err error) {
 327
 328         if resp.StatusCode != 200 {
 329                 return response, fmt.Errorf("Received error code %d in response to index request for %s: %s",
 330                         resp.StatusCode, keepServer.String(), resp.Status)
 331         }
 332
 333         if arvLogger != nil {
 334                 now := time.Now()
 335                 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
 336                         keepInfo := logger.GetOrCreateMap(p, "keep_info")
 337                         serverInfo := keepInfo[keepServer.UUID].(map[string]interface{})
 338                         serverInfo["index_response_received_at"] = now
 339                 })
 340         }
 341
 342         response.Address = keepServer
 343         response.Contents.BlockDigestToInfo =
 344                 make(map[blockdigest.DigestWithSize]BlockInfo)
 345         reader := bufio.NewReader(resp.Body)
 346         numLines, numDuplicates, numSizeDisagreements := 0, 0, 0
 347         for {
 348                 numLines++
 349                 line, err := reader.ReadString('\n')
 350                 if err == io.EOF {
 351                         return response, fmt.Errorf("Index from %s truncated at line %d",
 352                                 keepServer.String(), numLines)
 353                 } else if err != nil {
 354                         return response, fmt.Errorf("Error reading index response from %s at line %d: %v",
 355                                 keepServer.String(), numLines, err)
 356                 }
 357                 if line == "\n" {
 358                         if _, err := reader.Peek(1); err == nil {
 359                                 extra, _ := reader.ReadString('\n')
 360                                 return response, fmt.Errorf("Index from %s had trailing data at line %d after EOF marker: %s",
 361                                         keepServer.String(), numLines+1, extra)
 362                         } else if err != io.EOF {
 363                                 return response, fmt.Errorf("Index from %s had read error after EOF marker at line %d: %v",
 364                                         keepServer.String(), numLines, err)
 365                         }
 366                         numLines--
 367                         break
 368                 }
 369                 blockInfo, err := parseBlockInfoFromIndexLine(line)
 370                 if err != nil {
 371                         return response, fmt.Errorf("Error parsing BlockInfo from index line "+
 372                                 "received from %s: %v",
 373                                 keepServer.String(),
 374                                 err)
 375                 }
 376
 377                 if storedBlock, ok := response.Contents.BlockDigestToInfo[blockInfo.Digest]; ok {
 378                         // This server returned multiple lines containing the same block digest.
 379                         numDuplicates++
 380                         // Keep the block that's newer.
 381                         if storedBlock.Mtime < blockInfo.Mtime {
 382                                 response.Contents.BlockDigestToInfo[blockInfo.Digest] = blockInfo
 383                         }
 384                 } else {
 385                         response.Contents.BlockDigestToInfo[blockInfo.Digest] = blockInfo
 386                 }
 387         }
 388
 389         log.Printf("%s index contained %d lines with %d duplicates with "+
 390                 "%d size disagreements",
 391                 keepServer.String(),
 392                 numLines,
 393                 numDuplicates,
 394                 numSizeDisagreements)
 395
 396         if arvLogger != nil {
 397                 now := time.Now()
 398                 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
 399                         keepInfo := logger.GetOrCreateMap(p, "keep_info")
 400                         serverInfo := keepInfo[keepServer.UUID].(map[string]interface{})
 401
 402                         serverInfo["processing_finished_at"] = now
 403                         serverInfo["lines_received"] = numLines
 404                         serverInfo["duplicates_seen"] = numDuplicates
 405                         serverInfo["size_disagreements_seen"] = numSizeDisagreements
 406                 })
 407         }
 408         resp.Body.Close()
 409         return
 410 }
 411
 412 func parseBlockInfoFromIndexLine(indexLine string) (blockInfo BlockInfo, err error) {
 413         tokens := strings.Fields(indexLine)
 414         if len(tokens) != 2 {
 415                 err = fmt.Errorf("Expected 2 tokens per line but received a "+
 416                         "line containing %#q instead.",
 417                         tokens)
 418         }
 419
 420         var locator blockdigest.BlockLocator
 421         if locator, err = blockdigest.ParseBlockLocator(tokens[0]); err != nil {
 422                 err = fmt.Errorf("%v Received error while parsing line \"%#q\"",
 423                         err, indexLine)
 424                 return
 425         }
 426         if len(locator.Hints) > 0 {
 427                 err = fmt.Errorf("Block locator in index line should not contain hints "+
 428                         "but it does: %#q",
 429                         locator)
 430                 return
 431         }
 432
 433         blockInfo.Mtime, err = strconv.ParseInt(tokens[1], 10, 64)
 434         if err != nil {
 435                 return
 436         }
 437         blockInfo.Digest =
 438                 blockdigest.DigestWithSize{Digest: locator.Digest,
 439                         Size: uint32(locator.Size)}
 440         return
 441 }
 442
 443 // Summarize results from keep server
 444 func (readServers *ReadServers) Summarize(arvLogger *logger.Logger) {
 445         readServers.BlockReplicationCounts = make(map[int]int)
 446         for _, infos := range readServers.BlockToServers {
 447                 replication := len(infos)
 448                 readServers.BlockReplicationCounts[replication]++
 449         }
 450
 451         if arvLogger != nil {
 452                 arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
 453                         keepInfo := logger.GetOrCreateMap(p, "keep_info")
 454                         keepInfo["distinct_blocks_stored"] = len(readServers.BlockToServers)
 455                 })
 456         }
 457 }
 458
 459 // TrashRequest struct
 460 type TrashRequest struct {
 461         Locator    string `json:"locator"`
 462         BlockMtime int64  `json:"block_mtime"`
 463 }
 464
 465 // TrashList is an array of TrashRequest objects
 466 type TrashList []TrashRequest
 467
 468 // SendTrashLists to trash queue
 469 func SendTrashLists(arvLogger *logger.Logger, kc *keepclient.KeepClient, spl map[string]TrashList, dryRun bool) (errs []error) {
 470         count := 0
 471         barrier := make(chan error)
 472
 473         client := kc.Client
 474
 475         for url, v := range spl {
 476                 if arvLogger != nil {
 477                         // We need a local variable because Update doesn't call our mutator func until later,
 478                         // when our list variable might have been reused by the next loop iteration.
 479                         url := url
 480                         trashLen := len(v)
 481                         arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
 482                                 trashListInfo := logger.GetOrCreateMap(p, "trash_list_len")
 483                                 trashListInfo[url] = trashLen
 484                         })
 485                 }
 486
 487                 if dryRun {
 488                         log.Printf("dry run, not sending trash list to service %s with %d blocks", url, len(v))
 489                         continue
 490                 }
 491
 492                 count++
 493                 log.Printf("Sending trash list to %v", url)
 494
 495                 go (func(url string, v TrashList) {
 496                         pipeReader, pipeWriter := io.Pipe()
 497                         go (func() {
 498                                 enc := json.NewEncoder(pipeWriter)
 499                                 enc.Encode(v)
 500                                 pipeWriter.Close()
 501                         })()
 502
 503                         req, err := http.NewRequest("PUT", fmt.Sprintf("%s/trash", url), pipeReader)
 504                         if err != nil {
 505                                 log.Printf("Error creating trash list request for %v error: %v", url, err.Error())
 506                                 barrier <- err
 507                                 return
 508                         }
 509
 510                         req.Header.Add("Authorization", "OAuth2 "+kc.Arvados.ApiToken)
 511
 512                         // Make the request
 513                         var resp *http.Response
 514                         if resp, err = client.Do(req); err != nil {
 515                                 log.Printf("Error sending trash list to %v error: %v", url, err.Error())
 516                                 barrier <- err
 517                                 return
 518                         }
 519
 520                         log.Printf("Sent trash list to %v: response was HTTP %v", url, resp.Status)
 521
 522                         io.Copy(ioutil.Discard, resp.Body)
 523                         resp.Body.Close()
 524
 525                         if resp.StatusCode != 200 {
 526                                 barrier <- errors.New(fmt.Sprintf("Got HTTP code %v", resp.StatusCode))
 527                         } else {
 528                                 barrier <- nil
 529                         }
 530                 })(url, v)
 531         }
 532
 533         for i := 0; i < count; i++ {
 534                 b := <-barrier
 535                 if b != nil {
 536                         errs = append(errs, b)
 537                 }
 538         }
 539
 540         return errs
 541 }