6221: Add trash_list_test.go
[arvados.git] / services / datamanager / keep / keep.go
index 97bbf9d437f1c5e58d71d4688c5769c283b08c1e..871acc8e793929b909b1256b3ff7e14a90059eba 100644 (file)
@@ -4,15 +4,15 @@ package keep
 
 import (
        "bufio"
+       "encoding/json"
        "flag"
        "fmt"
-       //"git.curoverse.com/arvados.git/sdk/go/keepclient"
        "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/sdk/go/util"
        "git.curoverse.com/arvados.git/services/datamanager/loggerutil"
+       "io"
        "io/ioutil"
        "log"
        "net/http"
@@ -23,26 +23,26 @@ import (
 )
 
 type ServerAddress struct {
+       SSL  bool   `json:service_ssl_flag`
        Host string `json:"service_host"`
        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  int // TODO(misha): Replace this with a timestamp.
+       Digest blockdigest.DigestWithSize
+       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       int // TODO(misha): Replace this with a timestamp.
+       Mtime       int64 // TODO(misha): Replace this with a timestamp.
 }
 
 type ServerContents struct {
-       BlockDigestToInfo map[blockdigest.BlockDigest]BlockInfo
+       BlockDigestToInfo map[blockdigest.DigestWithSize]BlockInfo
 }
 
 type ServerResponse struct {
@@ -55,7 +55,7 @@ type ReadServers struct {
        KeepServerIndexToAddress []ServerAddress
        KeepServerAddressToIndex map[ServerAddress]int
        ServerToContents         map[ServerAddress]ServerContents
-       BlockToServers           map[blockdigest.BlockDigest][]BlockServerInfo
+       BlockToServers           map[blockdigest.DigestWithSize][]BlockServerInfo
        BlockReplicationCounts   map[int]int
 }
 
@@ -70,15 +70,6 @@ type KeepServiceList struct {
        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.
@@ -94,8 +85,17 @@ func init() {
                "File with the API token we should use to contact keep servers.")
 }
 
+// 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 (s ServerAddress) URL() string {
+       if s.SSL {
+               return fmt.Sprintf("https://%s:%d", s.Host, s.Port)
+       } else {
+               return fmt.Sprintf("http://%s:%d", s.Host, s.Port)
+       }
 }
 
 func getDataManagerToken(arvLogger *logger.Logger) string {
@@ -124,7 +124,7 @@ func GetKeepServersAndSummarize(params GetKeepServersParams) (results ReadServer
        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)
 
@@ -137,48 +137,37 @@ func GetKeepServers(params GetKeepServersParams) (results ReadServers) {
                        "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.Call("GET", "keep_services", "", "accessible", sdkParams, &sdkResponse)
+       err := params.Client.List("keep_services", sdkParams, &sdkResponse)
 
        if err != nil {
                loggerutil.FatalWithMessage(params.Logger,
                        fmt.Sprintf("Error requesting keep disks from API server: %v", err))
        }
 
-       // TODO(misha): Rewrite this block, stop using ContainsAllAvailableItems()
-       {
-               var numReceived, numAvailable int
-               results.ReadAllServers, numReceived, numAvailable =
-                       util.ContainsAllAvailableItems(sdkResponse)
-
-               if !results.ReadAllServers {
-                       log.Printf("ERROR: Did not receive all keep server addresses.")
-               }
-               log.Printf("Received %d of %d available keep server addresses.",
-                       numReceived,
-                       numAvailable)
-       }
-
        if params.Logger != nil {
-               properties, _ := params.Logger.Edit()
-               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
+               params.Logger.Update(func(p map[string]interface{}, e 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
+               })
+       }
 
-               properties["keep_info"] = keepInfo
+       log.Printf("Received keep services list: %+v", sdkResponse)
 
-               params.Logger.Record()
+       if len(sdkResponse.KeepServers) < sdkResponse.ItemsAvailable {
+               loggerutil.FatalWithMessage(params.Logger,
+                       fmt.Sprintf("Did not receive all available keep servers: %+v", sdkResponse))
        }
 
-       log.Printf("Received keep services list: %v", sdkResponse)
-
        results.KeepServerIndexToAddress = sdkResponse.KeepServers
        results.KeepServerAddressToIndex = make(map[ServerAddress]int)
        for i, address := range results.KeepServerIndexToAddress {
@@ -207,7 +196,7 @@ func GetKeepServers(params GetKeepServersParams) (results ReadServers) {
        }
 
        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 {
@@ -222,7 +211,6 @@ func GetKeepServers(params GetKeepServersParams) (results ReadServers) {
                        results.BlockToServers[blockInfo.Digest] = append(
                                results.BlockToServers[blockInfo.Digest],
                                BlockServerInfo{ServerIndex: serverIndex,
-                                       Size:  blockInfo.Size,
                                        Mtime: blockInfo.Mtime})
                }
        }
@@ -233,30 +221,84 @@ 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))
+                       fmt.Sprintf("Error fetching %s: %v. Response was %+v",
+                               req.URL.String(),
+                               err,
+                               resp))
        }
 
        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 := 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 := 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
+               })
+       }
+}
+
 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 {
-               properties, _ := arvLogger.Edit()
-               keepInfo := properties["keep_info"].(map[string]interface{})
-               serverInfo := make(map[string]interface{})
-               serverInfo["request_sent"] = time.Now()
-
-               keepInfo[keepServer.String()] = serverInfo
-
-               arvLogger.Record()
+               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["index_request_sent_at"] = now
+               })
        }
 
        req, err := http.NewRequest("GET", url, nil)
@@ -282,23 +324,46 @@ func ReadServerResponse(arvLogger *logger.Logger,
        }
 
        if arvLogger != nil {
-               properties, _ := arvLogger.Edit()
-               keepInfo := properties["keep_info"].(map[string]interface{})
-               serverInfo := keepInfo[keepServer.String()].(map[string]interface{})
-
-               serverInfo["response_received"] = time.Now()
-
-               arvLogger.Record()
+               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["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 "+
@@ -310,64 +375,33 @@ 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 {
-                                       properties, _ := arvLogger.Edit()
-                                       keepInfo := properties["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)
-                                       arvLogger.Record()
-                               }
-                       }
-                       // 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) {
+                       // 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 {
-                       properties, _ := arvLogger.Edit()
-                       keepInfo := properties["keep_info"].(map[string]interface{})
-                       serverInfo := keepInfo[keepServer.String()].(map[string]interface{})
-
-                       serverInfo["processing_finished"] = time.Now()
+
+       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["dupicates_seen"] = numDuplicates
+                       serverInfo["duplicates_seen"] = numDuplicates
                        serverInfo["size_disagreements_seen"] = numSizeDisagreements
-
-                       arvLogger.Record()
-               }
+               })
        }
        resp.Body.Close()
        return
@@ -381,8 +415,10 @@ 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 {
@@ -392,19 +428,84 @@ func parseBlockInfoFromIndexLine(indexLine string) (blockInfo BlockInfo, err err
                return
        }
 
-       blockInfo.Mtime, err = strconv.Atoi(tokens[1])
+       blockInfo.Mtime, err = strconv.ParseInt(tokens[1], 10, 64)
        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) {
+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
        }
+
+       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)
+               })
+       }
+
+}
+
+type TrashRequest struct {
+       Locator    string `json:"locator"`
+       BlockMtime int64  `json:"block_mtime"`
+}
+
+type TrashList []TrashRequest
+
+func SendTrashLists(arvLogger *logger.Logger, kc *keepclient.KeepClient, spl map[string]TrashList) {
+       count := 0
+       rendezvous := make(chan bool)
+
+       for url, v := range spl {
+               count += 1
+               log.Printf("Sending trash list to %v", url)
+
+               go (func(url string, v TrashList) {
+                       defer (func() {
+                               rendezvous <- true
+                       })()
+
+                       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())
+                               return
+                       }
+
+                       // Add api token header
+                       req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", getDataManagerToken(arvLogger)))
+
+                       // Make the request
+                       var resp *http.Response
+                       if resp, err = kc.Client.Do(req); err != nil {
+                               log.Printf("Error sending trash list to %v error: %v", url, err.Error())
+                               return
+                       }
+
+                       log.Printf("Sent trash list to %v: response was HTTP %d", url, resp.Status)
+
+                       io.Copy(ioutil.Discard, resp.Body)
+                       resp.Body.Close()
+               })(url, v)
+
+       }
+
+       for i := 0; i < count; i += 1 {
+               <-rendezvous
+       }
 }