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         "os"
  15         "regexp"
  16         "strings"
  17         "time"
  18 )
  19
  20 // Function used to emit debug messages. The easiest way to enable
  21 // keepclient debug messages in your application is to assign
  22 // log.Printf to DebugPrintf.
  23 var DebugPrintf = func(string, ...interface{}) {}
  24
  25 func init() {
  26         var matchTrue = regexp.MustCompile("^(?i:1|yes|true)$")
  27         if matchTrue.MatchString(os.Getenv("ARVADOS_DEBUG")) {
  28                 DebugPrintf = log.Printf
  29         }
  30 }
  31
  32 type keepService struct {
  33         Uuid     string `json:"uuid"`
  34         Hostname string `json:"service_host"`
  35         Port     int    `json:"service_port"`
  36         SSL      bool   `json:"service_ssl_flag"`
  37         SvcType  string `json:"service_type"`
  38         ReadOnly bool   `json:"read_only"`
  39 }
  40
  41 // Md5String returns md5 hash for the bytes in the given string
  42 func Md5String(s string) string {
  43         return fmt.Sprintf("%x", md5.Sum([]byte(s)))
  44 }
  45
  46 // Set timeouts applicable when connecting to non-disk services
  47 // (assumed to be over the Internet).
  48 func (this *KeepClient) setClientSettingsNonDisk() {
  49         if this.Client.Timeout == 0 {
  50                 // Maximum time to wait for a complete response
  51                 this.Client.Timeout = 300 * time.Second
  52
  53                 // TCP and TLS connection settings
  54                 this.Client.Transport = &http.Transport{
  55                         Dial: (&net.Dialer{
  56                                 // The maximum time to wait to set up
  57                                 // the initial TCP connection.
  58                                 Timeout: 30 * time.Second,
  59
  60                                 // The TCP keep alive heartbeat
  61                                 // interval.
  62                                 KeepAlive: 120 * time.Second,
  63                         }).Dial,
  64
  65                         TLSHandshakeTimeout: 10 * time.Second,
  66                 }
  67         }
  68 }
  69
  70 // Set timeouts applicable when connecting to keepstore services directly
  71 // (assumed to be on the local network).
  72 func (this *KeepClient) setClientSettingsDisk() {
  73         if this.Client.Timeout == 0 {
  74                 // Maximum time to wait for a complete response
  75                 this.Client.Timeout = 20 * time.Second
  76
  77                 // TCP and TLS connection timeouts
  78                 this.Client.Transport = &http.Transport{
  79                         Dial: (&net.Dialer{
  80                                 // The maximum time to wait to set up
  81                                 // the initial TCP connection.
  82                                 Timeout: 2 * time.Second,
  83
  84                                 // The TCP keep alive heartbeat
  85                                 // interval.
  86                                 KeepAlive: 180 * time.Second,
  87                         }).Dial,
  88
  89                         TLSHandshakeTimeout: 4 * time.Second,
  90                 }
  91         }
  92 }
  93
  94 type svcList struct {
  95         Items []keepService `json:"items"`
  96 }
  97
  98 type uploadStatus struct {
  99         err             error
 100         url             string
 101         statusCode      int
 102         replicas_stored int
 103         response        string
 104 }
 105
 106 func (this *KeepClient) uploadToKeepServer(host string, hash string, body io.ReadCloser,
 107         upload_status chan<- uploadStatus, expectedLength int64, requestID int32) {
 108
 109         var req *http.Request
 110         var err error
 111         var url = fmt.Sprintf("%s/%s", host, hash)
 112         if req, err = http.NewRequest("PUT", url, nil); err != nil {
 113                 DebugPrintf("DEBUG: [%08x] Error creating request PUT %v error: %v", requestID, url, err.Error())
 114                 upload_status <- uploadStatus{err, url, 0, 0, ""}
 115                 body.Close()
 116                 return
 117         }
 118
 119         req.ContentLength = expectedLength
 120         if expectedLength > 0 {
 121                 // http.Client.Do will close the body ReadCloser when it is
 122                 // done with it.
 123                 req.Body = body
 124         } else {
 125                 // "For client requests, a value of 0 means unknown if Body is
 126                 // not nil."  In this case we do want the body to be empty, so
 127                 // don't set req.Body.  However, we still need to close the
 128                 // body ReadCloser.
 129                 body.Close()
 130         }
 131
 132         req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", this.Arvados.ApiToken))
 133         req.Header.Add("Content-Type", "application/octet-stream")
 134         req.Header.Add(X_Keep_Desired_Replicas, fmt.Sprint(this.Want_replicas))
 135
 136         var resp *http.Response
 137         if resp, err = this.Client.Do(req); err != nil {
 138                 DebugPrintf("DEBUG: [%08x] Upload failed %v error: %v", requestID, url, err.Error())
 139                 upload_status <- uploadStatus{err, url, 0, 0, ""}
 140                 return
 141         }
 142
 143         rep := 1
 144         if xr := resp.Header.Get(X_Keep_Replicas_Stored); xr != "" {
 145                 fmt.Sscanf(xr, "%d", &rep)
 146         }
 147
 148         defer resp.Body.Close()
 149         defer io.Copy(ioutil.Discard, resp.Body)
 150
 151         respbody, err2 := ioutil.ReadAll(&io.LimitedReader{R: resp.Body, N: 4096})
 152         response := strings.TrimSpace(string(respbody))
 153         if err2 != nil && err2 != io.EOF {
 154                 DebugPrintf("DEBUG: [%08x] Upload %v error: %v response: %v", requestID, url, err2.Error(), response)
 155                 upload_status <- uploadStatus{err2, url, resp.StatusCode, rep, response}
 156         } else if resp.StatusCode == http.StatusOK {
 157                 DebugPrintf("DEBUG: [%08x] Upload %v success", requestID, url)
 158                 upload_status <- uploadStatus{nil, url, resp.StatusCode, rep, response}
 159         } else {
 160                 DebugPrintf("DEBUG: [%08x] Upload %v error: %v response: %v", requestID, url, resp.StatusCode, response)
 161                 upload_status <- uploadStatus{errors.New(resp.Status), url, resp.StatusCode, rep, response}
 162         }
 163 }
 164
 165 func (this *KeepClient) putReplicas(
 166         hash string,
 167         tr *streamer.AsyncStream,
 168         expectedLength int64) (locator string, replicas int, err error) {
 169
 170         // Generate an arbitrary ID to identify this specific
 171         // transaction in debug logs.
 172         requestID := rand.Int31()
 173
 174         // Calculate the ordering for uploading to servers
 175         sv := NewRootSorter(this.WritableLocalRoots(), hash).GetSortedRoots()
 176
 177         // The next server to try contacting
 178         next_server := 0
 179
 180         // The number of active writers
 181         active := 0
 182
 183         // Used to communicate status from the upload goroutines
 184         upload_status := make(chan uploadStatus)
 185         defer func() {
 186                 // Wait for any abandoned uploads (e.g., we started
 187                 // two uploads and the first replied with replicas=2)
 188                 // to finish before closing the status channel.
 189                 go func() {
 190                         for active > 0 {
 191                                 <-upload_status
 192                         }
 193                         close(upload_status)
 194                 }()
 195         }()
 196
 197         replicasDone := 0
 198         replicasTodo := this.Want_replicas
 199
 200         replicasPerThread := this.replicasPerService
 201         if replicasPerThread < 1 {
 202                 // unlimited or unknown
 203                 replicasPerThread = replicasTodo
 204         }
 205
 206         retriesRemaining := 1 + this.Retries
 207         var retryServers []string
 208
 209         for retriesRemaining > 0 {
 210                 retriesRemaining -= 1
 211                 next_server = 0
 212                 retryServers = []string{}
 213                 for replicasTodo > 0 {
 214                         for active*replicasPerThread < replicasTodo {
 215                                 // Start some upload requests
 216                                 if next_server < len(sv) {
 217                                         DebugPrintf("DEBUG: [%08x] Begin upload %s to %s", requestID, hash, sv[next_server])
 218                                         go this.uploadToKeepServer(sv[next_server], hash, tr.MakeStreamReader(), upload_status, expectedLength, requestID)
 219                                         next_server += 1
 220                                         active += 1
 221                                 } else {
 222                                         if active == 0 && retriesRemaining == 0 {
 223                                                 return locator, replicasDone, InsufficientReplicasError
 224                                         } else {
 225                                                 break
 226                                         }
 227                                 }
 228                         }
 229                         DebugPrintf("DEBUG: [%08x] Replicas remaining to write: %v active uploads: %v",
 230                                 requestID, replicasTodo, active)
 231
 232                         // Now wait for something to happen.
 233                         if active > 0 {
 234                                 status := <-upload_status
 235                                 active -= 1
 236
 237                                 if status.statusCode == 200 {
 238                                         // good news!
 239                                         replicasDone += status.replicas_stored
 240                                         replicasTodo -= status.replicas_stored
 241                                         locator = status.response
 242                                 } else if status.statusCode == 0 || status.statusCode == 408 || status.statusCode == 429 ||
 243                                         (status.statusCode >= 500 && status.statusCode != 503) {
 244                                         // Timeout, too many requests, or other server side failure
 245                                         // Do not retry when status code is 503, which means the keep server is full
 246                                         retryServers = append(retryServers, status.url[0:strings.LastIndex(status.url, "/")])
 247                                 }
 248                         } else {
 249                                 break
 250                         }
 251                 }
 252
 253                 sv = retryServers
 254         }
 255
 256         return locator, replicasDone, nil
 257 }