sdk/go/keepclient/support.go

   1 package keepclient
   2
   3 import (
   4         "crypto/md5"
   5         "errors"
   6         "fmt"
   7         "io"
   8         "io/ioutil"
   9         "log"
  10         "math/rand"
  11         "net/http"
  12         "os"
  13         "strings"
  14
  15         "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
  16         "git.curoverse.com/arvados.git/sdk/go/streamer"
  17 )
  18
  19 // Function used to emit debug messages. The easiest way to enable
  20 // keepclient debug messages in your application is to assign
  21 // log.Printf to DebugPrintf.
  22 var DebugPrintf = func(string, ...interface{}) {}
  23
  24 func init() {
  25         if arvadosclient.StringBool(os.Getenv("ARVADOS_DEBUG")) {
  26                 DebugPrintf = log.Printf
  27         }
  28 }
  29
  30 type keepService struct {
  31         Uuid     string `json:"uuid"`
  32         Hostname string `json:"service_host"`
  33         Port     int    `json:"service_port"`
  34         SSL      bool   `json:"service_ssl_flag"`
  35         SvcType  string `json:"service_type"`
  36         ReadOnly bool   `json:"read_only"`
  37 }
  38
  39 // Md5String returns md5 hash for the bytes in the given string
  40 func Md5String(s string) string {
  41         return fmt.Sprintf("%x", md5.Sum([]byte(s)))
  42 }
  43
  44 type svcList struct {
  45         Items []keepService `json:"items"`
  46 }
  47
  48 type uploadStatus struct {
  49         err             error
  50         url             string
  51         statusCode      int
  52         replicas_stored int
  53         response        string
  54 }
  55
  56 func (this *KeepClient) uploadToKeepServer(host string, hash string, body io.ReadCloser,
  57         upload_status chan<- uploadStatus, expectedLength int64, requestID int32) {
  58
  59         var req *http.Request
  60         var err error
  61         var url = fmt.Sprintf("%s/%s", host, hash)
  62         if req, err = http.NewRequest("PUT", url, nil); err != nil {
  63                 DebugPrintf("DEBUG: [%08x] Error creating request PUT %v error: %v", requestID, url, err.Error())
  64                 upload_status <- uploadStatus{err, url, 0, 0, ""}
  65                 body.Close()
  66                 return
  67         }
  68
  69         req.ContentLength = expectedLength
  70         if expectedLength > 0 {
  71                 // Do() will close the body ReadCloser when it is done
  72                 // with it.
  73                 req.Body = body
  74         } else {
  75                 // "For client requests, a value of 0 means unknown if Body is
  76                 // not nil."  In this case we do want the body to be empty, so
  77                 // don't set req.Body.  However, we still need to close the
  78                 // body ReadCloser.
  79                 body.Close()
  80         }
  81
  82         req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", this.Arvados.ApiToken))
  83         req.Header.Add("Content-Type", "application/octet-stream")
  84         req.Header.Add(X_Keep_Desired_Replicas, fmt.Sprint(this.Want_replicas))
  85
  86         var resp *http.Response
  87         if resp, err = this.httpClient().Do(req); err != nil {
  88                 DebugPrintf("DEBUG: [%08x] Upload failed %v error: %v", requestID, url, err.Error())
  89                 upload_status <- uploadStatus{err, url, 0, 0, ""}
  90                 return
  91         }
  92
  93         rep := 1
  94         if xr := resp.Header.Get(X_Keep_Replicas_Stored); xr != "" {
  95                 fmt.Sscanf(xr, "%d", &rep)
  96         }
  97
  98         defer resp.Body.Close()
  99         defer io.Copy(ioutil.Discard, resp.Body)
 100
 101         respbody, err2 := ioutil.ReadAll(&io.LimitedReader{R: resp.Body, N: 4096})
 102         response := strings.TrimSpace(string(respbody))
 103         if err2 != nil && err2 != io.EOF {
 104                 DebugPrintf("DEBUG: [%08x] Upload %v error: %v response: %v", requestID, url, err2.Error(), response)
 105                 upload_status <- uploadStatus{err2, url, resp.StatusCode, rep, response}
 106         } else if resp.StatusCode == http.StatusOK {
 107                 DebugPrintf("DEBUG: [%08x] Upload %v success", requestID, url)
 108                 upload_status <- uploadStatus{nil, url, resp.StatusCode, rep, response}
 109         } else {
 110                 if resp.StatusCode >= 300 && response == "" {
 111                         response = resp.Status
 112                 }
 113                 DebugPrintf("DEBUG: [%08x] Upload %v error: %v response: %v", requestID, url, resp.StatusCode, response)
 114                 upload_status <- uploadStatus{errors.New(resp.Status), url, resp.StatusCode, rep, response}
 115         }
 116 }
 117
 118 func (this *KeepClient) putReplicas(
 119         hash string,
 120         tr *streamer.AsyncStream,
 121         expectedLength int64) (locator string, replicas int, err error) {
 122
 123         // Generate an arbitrary ID to identify this specific
 124         // transaction in debug logs.
 125         requestID := rand.Int31()
 126
 127         // Calculate the ordering for uploading to servers
 128         sv := NewRootSorter(this.WritableLocalRoots(), hash).GetSortedRoots()
 129
 130         // The next server to try contacting
 131         next_server := 0
 132
 133         // The number of active writers
 134         active := 0
 135
 136         // Used to communicate status from the upload goroutines
 137         upload_status := make(chan uploadStatus)
 138         defer func() {
 139                 // Wait for any abandoned uploads (e.g., we started
 140                 // two uploads and the first replied with replicas=2)
 141                 // to finish before closing the status channel.
 142                 go func() {
 143                         for active > 0 {
 144                                 <-upload_status
 145                         }
 146                         close(upload_status)
 147                 }()
 148         }()
 149
 150         replicasDone := 0
 151         replicasTodo := this.Want_replicas
 152
 153         replicasPerThread := this.replicasPerService
 154         if replicasPerThread < 1 {
 155                 // unlimited or unknown
 156                 replicasPerThread = replicasTodo
 157         }
 158
 159         retriesRemaining := 1 + this.Retries
 160         var retryServers []string
 161
 162         lastError := make(map[string]string)
 163
 164         for retriesRemaining > 0 {
 165                 retriesRemaining -= 1
 166                 next_server = 0
 167                 retryServers = []string{}
 168                 for replicasTodo > 0 {
 169                         for active*replicasPerThread < replicasTodo {
 170                                 // Start some upload requests
 171                                 if next_server < len(sv) {
 172                                         DebugPrintf("DEBUG: [%08x] Begin upload %s to %s", requestID, hash, sv[next_server])
 173                                         go this.uploadToKeepServer(sv[next_server], hash, tr.MakeStreamReader(), upload_status, expectedLength, requestID)
 174                                         next_server += 1
 175                                         active += 1
 176                                 } else {
 177                                         if active == 0 && retriesRemaining == 0 {
 178                                                 msg := "Could not write sufficient replicas: "
 179                                                 for _, resp := range lastError {
 180                                                         msg += resp + "; "
 181                                                 }
 182                                                 msg = msg[:len(msg)-2]
 183                                                 return locator, replicasDone, InsufficientReplicasError(errors.New(msg))
 184                                         } else {
 185                                                 break
 186                                         }
 187                                 }
 188                         }
 189                         DebugPrintf("DEBUG: [%08x] Replicas remaining to write: %v active uploads: %v",
 190                                 requestID, replicasTodo, active)
 191
 192                         // Now wait for something to happen.
 193                         if active > 0 {
 194                                 status := <-upload_status
 195                                 active -= 1
 196
 197                                 if status.statusCode == 200 {
 198                                         // good news!
 199                                         replicasDone += status.replicas_stored
 200                                         replicasTodo -= status.replicas_stored
 201                                         locator = status.response
 202                                         delete(lastError, status.url)
 203                                 } else {
 204                                         msg := fmt.Sprintf("[%d] %s", status.statusCode, status.response)
 205                                         if len(msg) > 100 {
 206                                                 msg = msg[:100]
 207                                         }
 208                                         lastError[status.url] = msg
 209                                 }
 210
 211                                 if status.statusCode == 0 || status.statusCode == 408 || status.statusCode == 429 ||
 212                                         (status.statusCode >= 500 && status.statusCode != 503) {
 213                                         // Timeout, too many requests, or other server side failure
 214                                         // Do not retry when status code is 503, which means the keep server is full
 215                                         retryServers = append(retryServers, status.url[0:strings.LastIndex(status.url, "/")])
 216                                 }
 217                         } else {
 218                                 break
 219                         }
 220                 }
 221
 222                 sv = retryServers
 223         }
 224
 225         return locator, replicasDone, nil
 226 }