sdk/go/keepclient/support.go

   1 package keepclient
   2
   3 import (
   4         "crypto/md5"
   5         "errors"
   6         "fmt"
   7         "git.curoverse.com/arvados.git/sdk/go/streamer"
   8         "io"
   9         "io/ioutil"
  10         "log"
  11         "math/rand"
  12         "net"
  13         "net/http"
  14         "strings"
  15         "time"
  16 )
  17
  18 // Function used to emit debug messages. The easiest way to enable
  19 // keepclient debug messages in your application is to assign
  20 // log.Printf to DebugPrintf.
  21 var DebugPrintf = func(string, ...interface{}) {}
  22
  23 type keepService struct {
  24         Uuid     string `json:"uuid"`
  25         Hostname string `json:"service_host"`
  26         Port     int    `json:"service_port"`
  27         SSL      bool   `json:"service_ssl_flag"`
  28         SvcType  string `json:"service_type"`
  29         ReadOnly bool   `json:"read_only"`
  30 }
  31
  32 // Md5String returns md5 hash for the bytes in the given string
  33 func Md5String(s string) string {
  34         return fmt.Sprintf("%x", md5.Sum([]byte(s)))
  35 }
  36
  37 // Set timeouts applicable when connecting to non-disk services
  38 // (assumed to be over the Internet).
  39 func (this *KeepClient) setClientSettingsNonDisk() {
  40         if this.Client.Timeout == 0 {
  41                 // Maximum time to wait for a complete response
  42                 this.Client.Timeout = 300 * time.Second
  43
  44                 // TCP and TLS connection settings
  45                 this.Client.Transport = &http.Transport{
  46                         Dial: (&net.Dialer{
  47                                 // The maximum time to wait to set up
  48                                 // the initial TCP connection.
  49                                 Timeout: 30 * time.Second,
  50
  51                                 // The TCP keep alive heartbeat
  52                                 // interval.
  53                                 KeepAlive: 120 * time.Second,
  54                         }).Dial,
  55
  56                         TLSHandshakeTimeout: 10 * time.Second,
  57                 }
  58         }
  59 }
  60
  61 // Set timeouts applicable when connecting to keepstore services directly
  62 // (assumed to be on the local network).
  63 func (this *KeepClient) setClientSettingsDisk() {
  64         if this.Client.Timeout == 0 {
  65                 // Maximum time to wait for a complete response
  66                 this.Client.Timeout = 20 * time.Second
  67
  68                 // TCP and TLS connection timeouts
  69                 this.Client.Transport = &http.Transport{
  70                         Dial: (&net.Dialer{
  71                                 // The maximum time to wait to set up
  72                                 // the initial TCP connection.
  73                                 Timeout: 2 * time.Second,
  74
  75                                 // The TCP keep alive heartbeat
  76                                 // interval.
  77                                 KeepAlive: 180 * time.Second,
  78                         }).Dial,
  79
  80                         TLSHandshakeTimeout: 4 * time.Second,
  81                 }
  82         }
  83 }
  84
  85 type svcList struct {
  86         Items []keepService `json:"items"`
  87 }
  88
  89 type uploadStatus struct {
  90         err             error
  91         url             string
  92         statusCode      int
  93         replicas_stored int
  94         response        string
  95 }
  96
  97 func (this *KeepClient) uploadToKeepServer(host string, hash string, body io.ReadCloser,
  98         upload_status chan<- uploadStatus, expectedLength int64, requestID int32) {
  99
 100         var req *http.Request
 101         var err error
 102         var url = fmt.Sprintf("%s/%s", host, hash)
 103         if req, err = http.NewRequest("PUT", url, nil); err != nil {
 104                 log.Printf("[%08x] Error creating request PUT %v error: %v", requestID, url, err.Error())
 105                 upload_status <- uploadStatus{err, url, 0, 0, ""}
 106                 body.Close()
 107                 return
 108         }
 109
 110         req.ContentLength = expectedLength
 111         if expectedLength > 0 {
 112                 // http.Client.Do will close the body ReadCloser when it is
 113                 // done with it.
 114                 req.Body = body
 115         } else {
 116                 // "For client requests, a value of 0 means unknown if Body is
 117                 // not nil."  In this case we do want the body to be empty, so
 118                 // don't set req.Body.  However, we still need to close the
 119                 // body ReadCloser.
 120                 body.Close()
 121         }
 122
 123         req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", this.Arvados.ApiToken))
 124         req.Header.Add("Content-Type", "application/octet-stream")
 125         req.Header.Add(X_Keep_Desired_Replicas, fmt.Sprint(this.Want_replicas))
 126
 127         var resp *http.Response
 128         if resp, err = this.Client.Do(req); err != nil {
 129                 log.Printf("[%08x] Upload failed %v error: %v", requestID, url, err.Error())
 130                 upload_status <- uploadStatus{err, url, 0, 0, ""}
 131                 return
 132         }
 133
 134         rep := 1
 135         if xr := resp.Header.Get(X_Keep_Replicas_Stored); xr != "" {
 136                 fmt.Sscanf(xr, "%d", &rep)
 137         }
 138
 139         defer resp.Body.Close()
 140         defer io.Copy(ioutil.Discard, resp.Body)
 141
 142         respbody, err2 := ioutil.ReadAll(&io.LimitedReader{R: resp.Body, N: 4096})
 143         response := strings.TrimSpace(string(respbody))
 144         if err2 != nil && err2 != io.EOF {
 145                 log.Printf("[%08x] Upload %v error: %v response: %v", requestID, url, err2.Error(), response)
 146                 upload_status <- uploadStatus{err2, url, resp.StatusCode, rep, response}
 147         } else if resp.StatusCode == http.StatusOK {
 148                 log.Printf("[%08x] Upload %v success", requestID, url)
 149                 upload_status <- uploadStatus{nil, url, resp.StatusCode, rep, response}
 150         } else {
 151                 log.Printf("[%08x] Upload %v error: %v response: %v", requestID, url, resp.StatusCode, response)
 152                 upload_status <- uploadStatus{errors.New(resp.Status), url, resp.StatusCode, rep, response}
 153         }
 154 }
 155
 156 func (this *KeepClient) putReplicas(
 157         hash string,
 158         tr *streamer.AsyncStream,
 159         expectedLength int64) (locator string, replicas int, err error) {
 160
 161         // Generate an arbitrary ID to identify this specific
 162         // transaction in debug logs.
 163         requestID := rand.Int31()
 164
 165         // Calculate the ordering for uploading to servers
 166         sv := NewRootSorter(this.WritableLocalRoots(), hash).GetSortedRoots()
 167
 168         // The next server to try contacting
 169         next_server := 0
 170
 171         // The number of active writers
 172         active := 0
 173
 174         // Used to communicate status from the upload goroutines
 175         upload_status := make(chan uploadStatus)
 176         defer func() {
 177                 // Wait for any abandoned uploads (e.g., we started
 178                 // two uploads and the first replied with replicas=2)
 179                 // to finish before closing the status channel.
 180                 go func() {
 181                         for active > 0 {
 182                                 <-upload_status
 183                         }
 184                         close(upload_status)
 185                 }()
 186         }()
 187
 188         // Desired number of replicas
 189         remaining_replicas := this.Want_replicas
 190
 191         replicasPerThread := this.replicasPerService
 192         if replicasPerThread < 1 {
 193                 // unlimited or unknown
 194                 replicasPerThread = remaining_replicas
 195         }
 196
 197         retriesRemaining := 1 + this.Retries
 198         var retryServers []string
 199
 200         for retriesRemaining > 0 {
 201                 retriesRemaining -= 1
 202                 next_server = 0
 203                 retryServers = []string{}
 204                 for remaining_replicas > 0 {
 205                         for active*replicasPerThread < remaining_replicas {
 206                                 // Start some upload requests
 207                                 if next_server < len(sv) {
 208                                         log.Printf("[%08x] Begin upload %s to %s", requestID, hash, sv[next_server])
 209                                         go this.uploadToKeepServer(sv[next_server], hash, tr.MakeStreamReader(), upload_status, expectedLength, requestID)
 210                                         next_server += 1
 211                                         active += 1
 212                                 } else {
 213                                         if active == 0 && retriesRemaining == 0 {
 214                                                 return locator, (this.Want_replicas - remaining_replicas), InsufficientReplicasError
 215                                         } else {
 216                                                 break
 217                                         }
 218                                 }
 219                         }
 220                         log.Printf("[%08x] Replicas remaining to write: %v active uploads: %v",
 221                                 requestID, remaining_replicas, active)
 222
 223                         // Now wait for something to happen.
 224                         if active > 0 {
 225                                 status := <-upload_status
 226                                 active -= 1
 227
 228                                 if status.statusCode == 200 {
 229                                         // good news!
 230                                         remaining_replicas -= status.replicas_stored
 231                                         locator = status.response
 232                                 } else if status.statusCode == 0 || status.statusCode == 408 || status.statusCode == 429 ||
 233                                         (status.statusCode >= 500 && status.statusCode != 503) {
 234                                         // Timeout, too many requests, or other server side failure
 235                                         // Do not retry when status code is 503, which means the keep server is full
 236                                         retryServers = append(retryServers, status.url[0:strings.LastIndex(status.url, "/")])
 237                                 }
 238                         } else {
 239                                 break
 240                         }
 241                 }
 242
 243                 sv = retryServers
 244         }
 245
 246         return locator, this.Want_replicas, nil
 247 }