X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/512e652c33bb3a03b46ec630521fb9d6df6585b7..9daf42fbdb868939653c6e3ca8a4fffd1cf94e31:/services/datamanager/keep/keep.go diff --git a/services/datamanager/keep/keep.go b/services/datamanager/keep/keep.go index d760bc4377..93246bc472 100644 --- a/services/datamanager/keep/keep.go +++ b/services/datamanager/keep/keep.go @@ -3,77 +3,442 @@ package keep import ( - //"git.curoverse.com/arvados.git/sdk/go/keepclient" + "bufio" + "encoding/json" + "flag" + "fmt" "git.curoverse.com/arvados.git/sdk/go/arvadosclient" - "git.curoverse.com/arvados.git/sdk/go/util" + "git.curoverse.com/arvados.git/sdk/go/blockdigest" + "git.curoverse.com/arvados.git/sdk/go/logger" + "git.curoverse.com/arvados.git/sdk/go/manifest" + "git.curoverse.com/arvados.git/services/datamanager/loggerutil" + "io/ioutil" "log" + "net/http" + "strconv" + "strings" + "sync" + "time" ) type ServerAddress struct { Host string `json:"service_host"` - Port int `json:"service_port"` + Port int `json:"service_port"` + Uuid string `json:"uuid"` +} + +// Info about a particular block returned by the server +type BlockInfo struct { + Digest blockdigest.BlockDigest + Size int + Mtime int64 // TODO(misha): Replace this with a timestamp. +} + +// Info about a specified block given by a server +type BlockServerInfo struct { + ServerIndex int + Size int + Mtime int64 // TODO(misha): Replace this with a timestamp. } type ServerContents struct { - BlockDigestToSize map[string]int + BlockDigestToInfo map[blockdigest.BlockDigest]BlockInfo +} + +type ServerResponse struct { + Address ServerAddress + Contents ServerContents } type ReadServers struct { - ReadAllServers bool - AddressToContents map[ServerAddress]ServerContents + ReadAllServers bool + KeepServerIndexToAddress []ServerAddress + KeepServerAddressToIndex map[ServerAddress]int + ServerToContents map[ServerAddress]ServerContents + BlockToServers map[blockdigest.BlockDigest][]BlockServerInfo + BlockReplicationCounts map[int]int } type GetKeepServersParams struct { Client arvadosclient.ArvadosClient - Limit int + Logger *logger.Logger + Limit int } type KeepServiceList struct { - ItemsAvailable int `json:"items_available"` - Items []ServerAddress `json:"items"` + ItemsAvailable int `json:"items_available"` + KeepServers []ServerAddress `json:"items"` } -// Methods to implement util.SdkListResponse Interface -func (k KeepServiceList) NumItemsAvailable() (numAvailable int, err error) { - return k.ItemsAvailable, nil +var ( + // Don't access the token directly, use getDataManagerToken() to + // make sure it's been read. + dataManagerToken string + dataManagerTokenFile string + dataManagerTokenFileReadOnce sync.Once +) + +func init() { + flag.StringVar(&dataManagerTokenFile, + "data-manager-token-file", + "", + "File with the API token we should use to contact keep servers.") } -func (k KeepServiceList) NumItemsContained() (numContained int, err error) { - return len(k.Items), nil +// TODO(misha): Change this to include the UUID as well. +func (s ServerAddress) String() string { + return fmt.Sprintf("%s:%d", s.Host, s.Port) } +func getDataManagerToken(arvLogger *logger.Logger) string { + readDataManagerToken := func() { + if dataManagerTokenFile == "" { + flag.Usage() + loggerutil.FatalWithMessage(arvLogger, + "Data Manager Token needed, but data manager token file not specified.") + } else { + rawRead, err := ioutil.ReadFile(dataManagerTokenFile) + if err != nil { + loggerutil.FatalWithMessage(arvLogger, + fmt.Sprintf("Unexpected error reading token file %s: %v", + dataManagerTokenFile, + err)) + } + dataManagerToken = strings.TrimSpace(string(rawRead)) + } + } + + dataManagerTokenFileReadOnce.Do(readDataManagerToken) + return dataManagerToken +} + +func GetKeepServersAndSummarize(params GetKeepServersParams) (results ReadServers) { + results = GetKeepServers(params) + log.Printf("Returned %d keep disks", len(results.ServerToContents)) + + ComputeBlockReplicationCounts(&results) + log.Printf("Replication level distribution: %v", + results.BlockReplicationCounts) + + return +} -// TODO(misha): Send Keep requests in parallel func GetKeepServers(params GetKeepServersParams) (results ReadServers) { if ¶ms.Client == nil { log.Fatalf("params.Client passed to GetKeepServers() should " + "contain a valid ArvadosClient, but instead it is nil.") } - sdkParams := arvadosclient.Dict{} + sdkParams := arvadosclient.Dict{ + "filters": [][]string{[]string{"service_type", "=", "disk"}}, + } if params.Limit > 0 { sdkParams["limit"] = params.Limit } var sdkResponse KeepServiceList err := params.Client.List("keep_services", sdkParams, &sdkResponse) + if err != nil { - log.Fatalf("Error requesting keep disks from API server: %v", err) + loggerutil.FatalWithMessage(params.Logger, + fmt.Sprintf("Error requesting keep disks from API server: %v", err)) } - log.Printf("Received keep services list: %v", sdkResponse) + if params.Logger != nil { + params.Logger.Update(func(p map[string]interface{}, e map[string]interface{}) { + keepInfo := make(map[string]interface{}) + + keepInfo["num_keep_servers_available"] = sdkResponse.ItemsAvailable + keepInfo["num_keep_servers_received"] = len(sdkResponse.KeepServers) + keepInfo["keep_servers"] = sdkResponse.KeepServers - { - var numReceived, numAvailable int - results.ReadAllServers, numReceived, numAvailable = - util.ContainsAllAvailableItems(sdkResponse) + p["keep_info"] = keepInfo + }) + } - if (!results.ReadAllServers) { - log.Printf("ERROR: Did not receive all keep server addresses.") + log.Printf("Received keep services list: %+v", sdkResponse) + + if len(sdkResponse.KeepServers) < sdkResponse.ItemsAvailable { + loggerutil.FatalWithMessage(params.Logger, + fmt.Sprintf("Did not receive all available keep servers: %+v", sdkResponse)) + } + + results.KeepServerIndexToAddress = sdkResponse.KeepServers + results.KeepServerAddressToIndex = make(map[ServerAddress]int) + for i, address := range results.KeepServerIndexToAddress { + results.KeepServerAddressToIndex[address] = i + } + + log.Printf("Got Server Addresses: %v", results) + + // This is safe for concurrent use + client := http.Client{} + + // Send off all the index requests concurrently + responseChan := make(chan ServerResponse) + for _, keepServer := range sdkResponse.KeepServers { + // The above keepsServer variable is reused for each iteration, so + // it would be shared across all goroutines. This would result in + // us querying one server n times instead of n different servers + // as we intended. To avoid this we add it as an explicit + // parameter which gets copied. This bug and solution is described + // in https://golang.org/doc/effective_go.html#channels + go func(keepServer ServerAddress) { + responseChan <- GetServerContents(params.Logger, + keepServer, + client) + }(keepServer) + } + + results.ServerToContents = make(map[ServerAddress]ServerContents) + results.BlockToServers = make(map[blockdigest.BlockDigest][]BlockServerInfo) + + // Read all the responses + for i := range sdkResponse.KeepServers { + _ = i // Here to prevent go from complaining. + response := <-responseChan + log.Printf("Received channel response from %v containing %d files", + response.Address, + len(response.Contents.BlockDigestToInfo)) + results.ServerToContents[response.Address] = response.Contents + serverIndex := results.KeepServerAddressToIndex[response.Address] + for _, blockInfo := range response.Contents.BlockDigestToInfo { + results.BlockToServers[blockInfo.Digest] = append( + results.BlockToServers[blockInfo.Digest], + BlockServerInfo{ServerIndex: serverIndex, + Size: blockInfo.Size, + Mtime: blockInfo.Mtime}) } - log.Printf("Received %d of %d available keep server addresses.", - numReceived, - numAvailable) } return } + +func GetServerContents(arvLogger *logger.Logger, + keepServer ServerAddress, + client http.Client) (response ServerResponse) { + + GetServerStatus(arvLogger, keepServer, client) + + req := CreateIndexRequest(arvLogger, keepServer) + resp, err := client.Do(req) + if err != nil { + loggerutil.FatalWithMessage(arvLogger, + fmt.Sprintf("Error fetching %s: %v", req.URL.String(), err)) + } + + return ReadServerResponse(arvLogger, keepServer, resp) +} + +func GetServerStatus(arvLogger *logger.Logger, + keepServer ServerAddress, + client http.Client) { + url := fmt.Sprintf("http://%s:%d/status.json", + keepServer.Host, + keepServer.Port) + + if arvLogger != nil { + now := time.Now() + arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) { + keepInfo := p["keep_info"].(map[string]interface{}) + serverInfo := make(map[string]interface{}) + serverInfo["status_request_sent_at"] = now + serverInfo["host"] = keepServer.Host + serverInfo["port"] = keepServer.Port + + keepInfo[keepServer.Uuid] = serverInfo + }) + } + + resp, err := client.Get(url) + if err != nil { + loggerutil.FatalWithMessage(arvLogger, + fmt.Sprintf("Error getting keep status from %s: %v", url, err)) + } else if resp.StatusCode != 200 { + loggerutil.FatalWithMessage(arvLogger, + fmt.Sprintf("Received error code %d in response to request "+ + "for %s status: %s", + resp.StatusCode, url, resp.Status)) + } + + var keepStatus map[string]interface{} + decoder := json.NewDecoder(resp.Body) + decoder.UseNumber() + err = decoder.Decode(&keepStatus) + if err != nil { + loggerutil.FatalWithMessage(arvLogger, + fmt.Sprintf("Error decoding keep status from %s: %v", url, err)) + } + + if arvLogger != nil { + now := time.Now() + arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) { + keepInfo := p["keep_info"].(map[string]interface{}) + serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{}) + serverInfo["status_response_processed_at"] = now + serverInfo["status"] = keepStatus + }) + } +} + +func CreateIndexRequest(arvLogger *logger.Logger, + keepServer ServerAddress) (req *http.Request) { + url := fmt.Sprintf("http://%s:%d/index", keepServer.Host, keepServer.Port) + log.Println("About to fetch keep server contents from " + url) + + if arvLogger != nil { + now := time.Now() + arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) { + keepInfo := p["keep_info"].(map[string]interface{}) + serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{}) + serverInfo["index_request_sent_at"] = now + }) + } + + req, err := http.NewRequest("GET", url, nil) + if err != nil { + loggerutil.FatalWithMessage(arvLogger, + fmt.Sprintf("Error building http request for %s: %v", url, err)) + } + + req.Header.Add("Authorization", + fmt.Sprintf("OAuth2 %s", getDataManagerToken(arvLogger))) + return +} + +func ReadServerResponse(arvLogger *logger.Logger, + keepServer ServerAddress, + resp *http.Response) (response ServerResponse) { + + if resp.StatusCode != 200 { + loggerutil.FatalWithMessage(arvLogger, + fmt.Sprintf("Received error code %d in response to request "+ + "for %s index: %s", + resp.StatusCode, keepServer.String(), resp.Status)) + } + + if arvLogger != nil { + now := time.Now() + arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) { + keepInfo := p["keep_info"].(map[string]interface{}) + serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{}) + serverInfo["index_response_received_at"] = now + }) + } + + response.Address = keepServer + response.Contents.BlockDigestToInfo = + make(map[blockdigest.BlockDigest]BlockInfo) + scanner := bufio.NewScanner(resp.Body) + numLines, numDuplicates, numSizeDisagreements := 0, 0, 0 + for scanner.Scan() { + numLines++ + blockInfo, err := parseBlockInfoFromIndexLine(scanner.Text()) + if err != nil { + loggerutil.FatalWithMessage(arvLogger, + fmt.Sprintf("Error parsing BlockInfo from index line "+ + "received from %s: %v", + keepServer.String(), + err)) + } + + if storedBlock, ok := response.Contents.BlockDigestToInfo[blockInfo.Digest]; ok { + // This server returned multiple lines containing the same block digest. + numDuplicates += 1 + if storedBlock.Size != blockInfo.Size { + numSizeDisagreements += 1 + // TODO(misha): Consider failing here. + message := fmt.Sprintf("Saw different sizes for the same block "+ + "on %s: %+v %+v", + keepServer.String(), + storedBlock, + blockInfo) + log.Println(message) + if arvLogger != nil { + arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) { + keepInfo := p["keep_info"].(map[string]interface{}) + serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{}) + var error_list []string + read_error_list, has_list := serverInfo["error_list"] + if has_list { + error_list = read_error_list.([]string) + } // If we didn't have the list, error_list is already an empty list + serverInfo["error_list"] = append(error_list, message) + }) + } + } + // Keep the block that is bigger, or the block that's newer in + // the case of a size tie. + if storedBlock.Size < blockInfo.Size || + (storedBlock.Size == blockInfo.Size && + storedBlock.Mtime < blockInfo.Mtime) { + response.Contents.BlockDigestToInfo[blockInfo.Digest] = blockInfo + } + } else { + response.Contents.BlockDigestToInfo[blockInfo.Digest] = blockInfo + } + } + if err := scanner.Err(); err != nil { + loggerutil.FatalWithMessage(arvLogger, + fmt.Sprintf("Received error scanning index response from %s: %v", + keepServer.String(), + err)) + } else { + log.Printf("%s index contained %d lines with %d duplicates with "+ + "%d size disagreements", + keepServer.String(), + numLines, + numDuplicates, + numSizeDisagreements) + + if arvLogger != nil { + now := time.Now() + arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) { + keepInfo := p["keep_info"].(map[string]interface{}) + serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{}) + + serverInfo["processing_finished_at"] = now + serverInfo["lines_received"] = numLines + serverInfo["duplicates_seen"] = numDuplicates + serverInfo["size_disagreements_seen"] = numSizeDisagreements + }) + } + } + resp.Body.Close() + return +} + +func parseBlockInfoFromIndexLine(indexLine string) (blockInfo BlockInfo, err error) { + tokens := strings.Fields(indexLine) + if len(tokens) != 2 { + err = fmt.Errorf("Expected 2 tokens per line but received a "+ + "line containing %v instead.", + tokens) + } + + var locator manifest.BlockLocator + if locator, err = manifest.ParseBlockLocator(tokens[0]); err != nil { + return + } + if len(locator.Hints) > 0 { + err = fmt.Errorf("Block locator in index line should not contain hints "+ + "but it does: %v", + locator) + return + } + + blockInfo.Mtime, err = strconv.ParseInt(tokens[1], 10, 64) + if err != nil { + return + } + blockInfo.Digest = locator.Digest + blockInfo.Size = locator.Size + return +} + +func ComputeBlockReplicationCounts(readServers *ReadServers) { + readServers.BlockReplicationCounts = make(map[int]int) + for _, infos := range readServers.BlockToServers { + replication := len(infos) + readServers.BlockReplicationCounts[replication] += 1 + } +}