5824: Add some clarifying comments and golint/vet/fmt fixes.
[arvados.git] / services / datamanager / keep / keep.go
index 39ac30a1b9938247ee537fe77824bcf07647d066..3a9c21a43855a472c4f5b43aa9b651fd85f506d4 100644 (file)
@@ -4,145 +4,114 @@ package keep
 
 import (
        "bufio"
-       "flag"
+       "encoding/json"
+       "errors"
        "fmt"
        "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
        "git.curoverse.com/arvados.git/sdk/go/blockdigest"
+       "git.curoverse.com/arvados.git/sdk/go/keepclient"
        "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"
        "io/ioutil"
        "log"
        "net/http"
        "strconv"
        "strings"
-       "sync"
        "time"
 )
 
+// ServerAddress struct
 type ServerAddress struct {
-       Host string `json:"service_host"`
-       Port int    `json:"service_port"`
+       SSL         bool   `json:"service_ssl_flag"`
+       Host        string `json:"service_host"`
+       Port        int    `json:"service_port"`
+       UUID        string `json:"uuid"`
+       ServiceType string `json:"service_type"`
 }
 
-// Info about a particular block returned by the server
+// BlockInfo is info about a particular block returned by the server
 type BlockInfo struct {
-       Digest blockdigest.BlockDigest
-       Size   int
+       Digest blockdigest.DigestWithSize
        Mtime  int64 // TODO(misha): Replace this with a timestamp.
 }
 
-// Info about a specified block given by a server
+// BlockServerInfo is 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.
 }
 
+// ServerContents struct
 type ServerContents struct {
-       BlockDigestToInfo map[blockdigest.BlockDigest]BlockInfo
+       BlockDigestToInfo map[blockdigest.DigestWithSize]BlockInfo
 }
 
+// ServerResponse struct
 type ServerResponse struct {
        Address  ServerAddress
        Contents ServerContents
 }
 
+// ReadServers struct
 type ReadServers struct {
        ReadAllServers           bool
        KeepServerIndexToAddress []ServerAddress
        KeepServerAddressToIndex map[ServerAddress]int
        ServerToContents         map[ServerAddress]ServerContents
-       BlockToServers           map[blockdigest.BlockDigest][]BlockServerInfo
+       BlockToServers           map[blockdigest.DigestWithSize][]BlockServerInfo
        BlockReplicationCounts   map[int]int
 }
 
+// GetKeepServersParams struct
 type GetKeepServersParams struct {
        Client arvadosclient.ArvadosClient
        Logger *logger.Logger
        Limit  int
 }
 
-type KeepServiceList struct {
+// ServiceList consists of the addresses of all the available kee servers
+type ServiceList struct {
        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
-}
-
-func (k KeepServiceList) NumItemsContained() (numContained int, err error) {
-       return len(k.KeepServers), 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.")
-}
-
+// String
+// TODO(misha): Change this to include the UUID as well.
 func (s ServerAddress) String() string {
-       return fmt.Sprintf("%s:%d", s.Host, s.Port)
+       return s.URL()
 }
 
-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))
-               }
+// URL of the keep server
+func (s ServerAddress) URL() string {
+       if s.SSL {
+               return fmt.Sprintf("https://%s:%d", s.Host, s.Port)
        }
-
-       dataManagerTokenFileReadOnce.Do(readDataManagerToken)
-       return dataManagerToken
+       return fmt.Sprintf("http://%s:%d", s.Host, s.Port)
 }
 
+// GetKeepServersAndSummarize gets keep servers from api
 func GetKeepServersAndSummarize(params GetKeepServersParams) (results ReadServers) {
        results = GetKeepServers(params)
        log.Printf("Returned %d keep disks", len(results.ServerToContents))
 
-       ComputeBlockReplicationCounts(&results)
+       results.Summarize(params.Logger)
        log.Printf("Replication level distribution: %v",
                results.BlockReplicationCounts)
 
        return
 }
 
+// GetKeepServers from api server
 func GetKeepServers(params GetKeepServersParams) (results ReadServers) {
-       if &params.Client == nil {
-               log.Fatalf("params.Client passed to GetKeepServers() should " +
-                       "contain a valid ArvadosClient, but instead it is nil.")
-       }
-
        sdkParams := arvadosclient.Dict{
-               "filters": [][]string{[]string{"service_type", "=", "disk"}},
+               "filters": [][]string{[]string{"service_type", "!=", "proxy"}},
        }
        if params.Limit > 0 {
                sdkParams["limit"] = params.Limit
        }
 
-       var sdkResponse KeepServiceList
+       var sdkResponse ServiceList
        err := params.Client.List("keep_services", sdkParams, &sdkResponse)
 
        if err != nil {
@@ -150,15 +119,20 @@ func GetKeepServers(params GetKeepServersParams) (results ReadServers) {
                        fmt.Sprintf("Error requesting keep disks from API server: %v", err))
        }
 
+       // Currently, only "disk" types are supported. Stop if any other service types are found.
+       for _, server := range sdkResponse.KeepServers {
+               if server.ServiceType != "disk" {
+                       loggerutil.FatalWithMessage(params.Logger,
+                               fmt.Sprintf("Unsupported service type %q found for: %v", server.ServiceType, server))
+               }
+       }
+
        if params.Logger != nil {
                params.Logger.Update(func(p map[string]interface{}, e map[string]interface{}) {
-                       keepInfo := make(map[string]interface{})
-
+                       keepInfo := logger.GetOrCreateMap(p, "keep_info")
                        keepInfo["num_keep_servers_available"] = sdkResponse.ItemsAvailable
                        keepInfo["num_keep_servers_received"] = len(sdkResponse.KeepServers)
                        keepInfo["keep_servers"] = sdkResponse.KeepServers
-
-                       p["keep_info"] = keepInfo
                })
        }
 
@@ -177,9 +151,6 @@ func GetKeepServers(params GetKeepServersParams) (results ReadServers) {
 
        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 {
@@ -192,12 +163,12 @@ func GetKeepServers(params GetKeepServersParams) (results ReadServers) {
                go func(keepServer ServerAddress) {
                        responseChan <- GetServerContents(params.Logger,
                                keepServer,
-                               client)
+                               params.Client)
                }(keepServer)
        }
 
        results.ServerToContents = make(map[ServerAddress]ServerContents)
-       results.BlockToServers = make(map[blockdigest.BlockDigest][]BlockServerInfo)
+       results.BlockToServers = make(map[blockdigest.DigestWithSize][]BlockServerInfo)
 
        // Read all the responses
        for i := range sdkResponse.KeepServers {
@@ -212,39 +183,97 @@ func GetKeepServers(params GetKeepServersParams) (results ReadServers) {
                        results.BlockToServers[blockInfo.Digest] = append(
                                results.BlockToServers[blockInfo.Digest],
                                BlockServerInfo{ServerIndex: serverIndex,
-                                       Size:  blockInfo.Size,
                                        Mtime: blockInfo.Mtime})
                }
        }
        return
 }
 
+// GetServerContents of the keep server
 func GetServerContents(arvLogger *logger.Logger,
        keepServer ServerAddress,
-       client http.Client) (response ServerResponse) {
+       arv arvadosclient.ArvadosClient) (response ServerResponse) {
+
+       GetServerStatus(arvLogger, keepServer, arv)
 
-       req := CreateIndexRequest(arvLogger, keepServer)
-       resp, err := client.Do(req)
+       req := CreateIndexRequest(arvLogger, keepServer, arv)
+       resp, err := arv.Client.Do(req)
        if err != nil {
                loggerutil.FatalWithMessage(arvLogger,
-                       fmt.Sprintf("Error fetching %s: %v", req.URL.String(), err))
+                       fmt.Sprintf("Error fetching %s: %v. Response was %+v",
+                               req.URL.String(),
+                               err,
+                               resp))
        }
 
        return ReadServerResponse(arvLogger, keepServer, resp)
 }
 
+// GetServerStatus get keep server status by invoking /status.json
+func GetServerStatus(arvLogger *logger.Logger,
+       keepServer ServerAddress,
+       arv arvadosclient.ArvadosClient) {
+       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 := logger.GetOrCreateMap(p, "keep_info")
+                       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 := arv.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 := logger.GetOrCreateMap(p, "keep_info")
+                       serverInfo := keepInfo[keepServer.UUID].(map[string]interface{})
+                       serverInfo["status_response_processed_at"] = now
+                       serverInfo["status"] = keepStatus
+               })
+       }
+}
+
+// CreateIndexRequest to the keep server
 func CreateIndexRequest(arvLogger *logger.Logger,
-       keepServer ServerAddress) (req *http.Request) {
+       keepServer ServerAddress,
+       arv arvadosclient.ArvadosClient) (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 := make(map[string]interface{})
-                       serverInfo["request_sent"] = time.Now()
-
-                       keepInfo[keepServer.String()] = serverInfo
+                       keepInfo := logger.GetOrCreateMap(p, "keep_info")
+                       serverInfo := keepInfo[keepServer.UUID].(map[string]interface{})
+                       serverInfo["index_request_sent_at"] = now
                })
        }
 
@@ -254,11 +283,11 @@ func CreateIndexRequest(arvLogger *logger.Logger,
                        fmt.Sprintf("Error building http request for %s: %v", url, err))
        }
 
-       req.Header.Add("Authorization",
-               fmt.Sprintf("OAuth2 %s", getDataManagerToken(arvLogger)))
+       req.Header.Add("Authorization", "OAuth2 "+arv.ApiToken)
        return
 }
 
+// ReadServerResponse reads reasponse from keep server
 func ReadServerResponse(arvLogger *logger.Logger,
        keepServer ServerAddress,
        resp *http.Response) (response ServerResponse) {
@@ -271,22 +300,46 @@ func ReadServerResponse(arvLogger *logger.Logger,
        }
 
        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.String()].(map[string]interface{})
-
-                       serverInfo["response_received"] = time.Now()
+                       keepInfo := logger.GetOrCreateMap(p, "keep_info")
+                       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)
+               make(map[blockdigest.DigestWithSize]BlockInfo)
+       reader := bufio.NewReader(resp.Body)
        numLines, numDuplicates, numSizeDisagreements := 0, 0, 0
-       for scanner.Scan() {
+       for {
                numLines++
-               blockInfo, err := parseBlockInfoFromIndexLine(scanner.Text())
+               line, err := reader.ReadString('\n')
+               if err == io.EOF {
+                       loggerutil.FatalWithMessage(arvLogger,
+                               fmt.Sprintf("Index from %s truncated at line %d",
+                                       keepServer.String(), numLines))
+               } else if err != nil {
+                       loggerutil.FatalWithMessage(arvLogger,
+                               fmt.Sprintf("Error reading index response from %s at line %d: %v",
+                                       keepServer.String(), numLines, err))
+               }
+               if line == "\n" {
+                       if _, err := reader.Peek(1); err == nil {
+                               extra, _ := reader.ReadString('\n')
+                               loggerutil.FatalWithMessage(arvLogger,
+                                       fmt.Sprintf("Index from %s had trailing data at line %d after EOF marker: %s",
+                                               keepServer.String(), numLines+1, extra))
+                       } else if err != io.EOF {
+                               loggerutil.FatalWithMessage(arvLogger,
+                                       fmt.Sprintf("Index from %s had read error after EOF marker at line %d: %v",
+                                               keepServer.String(), numLines, err))
+                       }
+                       numLines--
+                       break
+               }
+               blockInfo, err := parseBlockInfoFromIndexLine(line)
                if err != nil {
                        loggerutil.FatalWithMessage(arvLogger,
                                fmt.Sprintf("Error parsing BlockInfo from index line "+
@@ -297,64 +350,34 @@ func ReadServerResponse(arvLogger *logger.Logger,
 
                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.String()].(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) {
+                       numDuplicates++
+                       // Keep the block that's newer.
+                       if 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 {
-                       arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
-                               keepInfo := p["keep_info"].(map[string]interface{})
-                               serverInfo := keepInfo[keepServer.String()].(map[string]interface{})
-
-                               serverInfo["processing_finished"] = time.Now()
-                               serverInfo["lines_received"] = numLines
-                               serverInfo["duplicates_seen"] = numDuplicates
-                               serverInfo["size_disagreements_seen"] = numSizeDisagreements
-                       })
-               }
+
+       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 := logger.GetOrCreateMap(p, "keep_info")
+                       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
@@ -368,8 +391,10 @@ func parseBlockInfoFromIndexLine(indexLine string) (blockInfo BlockInfo, err err
                        tokens)
        }
 
-       var locator manifest.BlockLocator
-       if locator, err = manifest.ParseBlockLocator(tokens[0]); err != nil {
+       var locator blockdigest.BlockLocator
+       if locator, err = blockdigest.ParseBlockLocator(tokens[0]); err != nil {
+               err = fmt.Errorf("%v Received error while parsing line \"%s\"",
+                       err, indexLine)
                return
        }
        if len(locator.Hints) > 0 {
@@ -383,15 +408,93 @@ func parseBlockInfoFromIndexLine(indexLine string) (blockInfo BlockInfo, err err
        if err != nil {
                return
        }
-       blockInfo.Digest = locator.Digest
-       blockInfo.Size = locator.Size
+       blockInfo.Digest =
+               blockdigest.DigestWithSize{Digest: locator.Digest,
+                       Size: uint32(locator.Size)}
        return
 }
 
-func ComputeBlockReplicationCounts(readServers *ReadServers) {
+// Summarize results from keep server
+func (readServers *ReadServers) Summarize(arvLogger *logger.Logger) {
        readServers.BlockReplicationCounts = make(map[int]int)
        for _, infos := range readServers.BlockToServers {
                replication := len(infos)
-               readServers.BlockReplicationCounts[replication] += 1
+               readServers.BlockReplicationCounts[replication]++
+       }
+
+       if arvLogger != nil {
+               arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
+                       keepInfo := logger.GetOrCreateMap(p, "keep_info")
+                       keepInfo["distinct_blocks_stored"] = len(readServers.BlockToServers)
+               })
        }
 }
+
+// TrashRequest struct
+type TrashRequest struct {
+       Locator    string `json:"locator"`
+       BlockMtime int64  `json:"block_mtime"`
+}
+
+// TrashList is an array of TrashRequest objects
+type TrashList []TrashRequest
+
+// SendTrashLists to trash queue
+func SendTrashLists(kc *keepclient.KeepClient, spl map[string]TrashList) (errs []error) {
+       count := 0
+       barrier := make(chan error)
+
+       client := kc.Client
+
+       for url, v := range spl {
+               count++
+               log.Printf("Sending trash list to %v", url)
+
+               go (func(url string, v TrashList) {
+                       pipeReader, pipeWriter := io.Pipe()
+                       go (func() {
+                               enc := json.NewEncoder(pipeWriter)
+                               enc.Encode(v)
+                               pipeWriter.Close()
+                       })()
+
+                       req, err := http.NewRequest("PUT", fmt.Sprintf("%s/trash", url), pipeReader)
+                       if err != nil {
+                               log.Printf("Error creating trash list request for %v error: %v", url, err.Error())
+                               barrier <- err
+                               return
+                       }
+
+                       req.Header.Add("Authorization", "OAuth2 "+kc.Arvados.ApiToken)
+
+                       // Make the request
+                       var resp *http.Response
+                       if resp, err = client.Do(req); err != nil {
+                               log.Printf("Error sending trash list to %v error: %v", url, err.Error())
+                               barrier <- err
+                               return
+                       }
+
+                       log.Printf("Sent trash list to %v: response was HTTP %v", url, resp.Status)
+
+                       io.Copy(ioutil.Discard, resp.Body)
+                       resp.Body.Close()
+
+                       if resp.StatusCode != 200 {
+                               barrier <- errors.New(fmt.Sprintf("Got HTTP code %v", resp.StatusCode))
+                       } else {
+                               barrier <- nil
+                       }
+               })(url, v)
+
+       }
+
+       for i := 0; i < count; i++ {
+               b := <-barrier
+               if b != nil {
+                       errs = append(errs, b)
+               }
+       }
+
+       return errs
+}