sdk/go/keepclient/support.go

   1 package keepclient
   2
   3 import (
   4         "crypto/md5"
   5         "encoding/json"
   6         "errors"
   7         "fmt"
   8         "git.curoverse.com/arvados.git/sdk/go/streamer"
   9         "io"
  10         "io/ioutil"
  11         "log"
  12         "math/rand"
  13         "net"
  14         "net/http"
  15         "strings"
  16         "time"
  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 type keepService struct {
  25         Uuid     string `json:"uuid"`
  26         Hostname string `json:"service_host"`
  27         Port     int    `json:"service_port"`
  28         SSL      bool   `json:"service_ssl_flag"`
  29         SvcType  string `json:"service_type"`
  30         ReadOnly bool   `json:"read_only"`
  31 }
  32
  33 // Md5String returns md5 hash for the bytes in the given string
  34 func Md5String(s string) string {
  35         return fmt.Sprintf("%x", md5.Sum([]byte(s)))
  36 }
  37
  38 // Set timeouts apply when connecting to keepproxy services (assumed to be over
  39 // the Internet).
  40 func (this *KeepClient) setClientSettingsProxy() {
  41         if this.Client.Timeout == 0 {
  42                 // Maximum time to wait for a complete response
  43                 this.Client.Timeout = 300 * time.Second
  44
  45                 // TCP and TLS connection settings
  46                 this.Client.Transport = &http.Transport{
  47                         Dial: (&net.Dialer{
  48                                 // The maximum time to wait to set up
  49                                 // the initial TCP connection.
  50                                 Timeout: 30 * time.Second,
  51
  52                                 // The TCP keep alive heartbeat
  53                                 // interval.
  54                                 KeepAlive: 120 * time.Second,
  55                         }).Dial,
  56
  57                         TLSHandshakeTimeout: 10 * time.Second,
  58                 }
  59         }
  60 }
  61
  62 // Set timeouts apply when connecting to keepstore services directly (assumed
  63 // to be on the local network).
  64 func (this *KeepClient) setClientSettingsDisk() {
  65         if this.Client.Timeout == 0 {
  66                 // Maximum time to wait for a complete response
  67                 this.Client.Timeout = 20 * time.Second
  68
  69                 // TCP and TLS connection timeouts
  70                 this.Client.Transport = &http.Transport{
  71                         Dial: (&net.Dialer{
  72                                 // The maximum time to wait to set up
  73                                 // the initial TCP connection.
  74                                 Timeout: 2 * time.Second,
  75
  76                                 // The TCP keep alive heartbeat
  77                                 // interval.
  78                                 KeepAlive: 180 * time.Second,
  79                         }).Dial,
  80
  81                         TLSHandshakeTimeout: 4 * time.Second,
  82                 }
  83         }
  84 }
  85
  86 type svcList struct {
  87         Items []keepService `json:"items"`
  88 }
  89
  90 // DiscoverKeepServers gets list of available keep services from api server
  91 func (this *KeepClient) DiscoverKeepServers() error {
  92         var list svcList
  93
  94         // Get keep services from api server
  95         err := this.Arvados.Call("GET", "keep_services", "", "accessible", nil, &list)
  96         if err != nil {
  97                 return err
  98         }
  99
 100         return this.loadKeepServers(list)
 101 }
 102
 103 // LoadKeepServicesFromJSON gets list of available keep services from given JSON
 104 func (this *KeepClient) LoadKeepServicesFromJSON(services string) error {
 105         var list svcList
 106
 107         // Load keep services from given json
 108         dec := json.NewDecoder(strings.NewReader(services))
 109         if err := dec.Decode(&list); err != nil {
 110                 return err
 111         }
 112
 113         return this.loadKeepServers(list)
 114 }
 115
 116 // loadKeepServers
 117 func (this *KeepClient) loadKeepServers(list svcList) error {
 118         listed := make(map[string]bool)
 119         localRoots := make(map[string]string)
 120         gatewayRoots := make(map[string]string)
 121         writableLocalRoots := make(map[string]string)
 122
 123         // replicasPerService is 1 for disks; unknown or unlimited otherwise
 124         this.replicasPerService = 1
 125         this.Using_proxy = false
 126
 127         for _, service := range list.Items {
 128                 scheme := "http"
 129                 if service.SSL {
 130                         scheme = "https"
 131                 }
 132                 url := fmt.Sprintf("%s://%s:%d", scheme, service.Hostname, service.Port)
 133
 134                 // Skip duplicates
 135                 if listed[url] {
 136                         continue
 137                 }
 138                 listed[url] = true
 139
 140                 localRoots[service.Uuid] = url
 141                 if service.SvcType == "proxy" {
 142                         this.Using_proxy = true
 143                 }
 144
 145                 if service.ReadOnly == false {
 146                         writableLocalRoots[service.Uuid] = url
 147                         if service.SvcType != "disk" {
 148                                 this.replicasPerService = 0
 149                         }
 150                 }
 151
 152                 // Gateway services are only used when specified by
 153                 // UUID, so there's nothing to gain by filtering them
 154                 // by service type. Including all accessible services
 155                 // (gateway and otherwise) merely accommodates more
 156                 // service configurations.
 157                 gatewayRoots[service.Uuid] = url
 158         }
 159
 160         if this.Using_proxy {
 161                 this.setClientSettingsProxy()
 162         } else {
 163                 this.setClientSettingsDisk()
 164         }
 165
 166         this.SetServiceRoots(localRoots, writableLocalRoots, gatewayRoots)
 167         return nil
 168 }
 169
 170 type uploadStatus struct {
 171         err             error
 172         url             string
 173         statusCode      int
 174         replicas_stored int
 175         response        string
 176 }
 177
 178 func (this *KeepClient) uploadToKeepServer(host string, hash string, body io.ReadCloser,
 179         upload_status chan<- uploadStatus, expectedLength int64, requestID int32) {
 180
 181         var req *http.Request
 182         var err error
 183         var url = fmt.Sprintf("%s/%s", host, hash)
 184         if req, err = http.NewRequest("PUT", url, nil); err != nil {
 185                 log.Printf("[%08x] Error creating request PUT %v error: %v", requestID, url, err.Error())
 186                 upload_status <- uploadStatus{err, url, 0, 0, ""}
 187                 body.Close()
 188                 return
 189         }
 190
 191         req.ContentLength = expectedLength
 192         if expectedLength > 0 {
 193                 // http.Client.Do will close the body ReadCloser when it is
 194                 // done with it.
 195                 req.Body = body
 196         } else {
 197                 // "For client requests, a value of 0 means unknown if Body is
 198                 // not nil."  In this case we do want the body to be empty, so
 199                 // don't set req.Body.  However, we still need to close the
 200                 // body ReadCloser.
 201                 body.Close()
 202         }
 203
 204         req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", this.Arvados.ApiToken))
 205         req.Header.Add("Content-Type", "application/octet-stream")
 206         req.Header.Add(X_Keep_Desired_Replicas, fmt.Sprint(this.Want_replicas))
 207
 208         var resp *http.Response
 209         if resp, err = this.Client.Do(req); err != nil {
 210                 log.Printf("[%08x] Upload failed %v error: %v", requestID, url, err.Error())
 211                 upload_status <- uploadStatus{err, url, 0, 0, ""}
 212                 return
 213         }
 214
 215         rep := 1
 216         if xr := resp.Header.Get(X_Keep_Replicas_Stored); xr != "" {
 217                 fmt.Sscanf(xr, "%d", &rep)
 218         }
 219
 220         defer resp.Body.Close()
 221         defer io.Copy(ioutil.Discard, resp.Body)
 222
 223         respbody, err2 := ioutil.ReadAll(&io.LimitedReader{R: resp.Body, N: 4096})
 224         response := strings.TrimSpace(string(respbody))
 225         if err2 != nil && err2 != io.EOF {
 226                 log.Printf("[%08x] Upload %v error: %v response: %v", requestID, url, err2.Error(), response)
 227                 upload_status <- uploadStatus{err2, url, resp.StatusCode, rep, response}
 228         } else if resp.StatusCode == http.StatusOK {
 229                 log.Printf("[%08x] Upload %v success", requestID, url)
 230                 upload_status <- uploadStatus{nil, url, resp.StatusCode, rep, response}
 231         } else {
 232                 log.Printf("[%08x] Upload %v error: %v response: %v", requestID, url, resp.StatusCode, response)
 233                 upload_status <- uploadStatus{errors.New(resp.Status), url, resp.StatusCode, rep, response}
 234         }
 235 }
 236
 237 func (this *KeepClient) putReplicas(
 238         hash string,
 239         tr *streamer.AsyncStream,
 240         expectedLength int64) (locator string, replicas int, err error) {
 241
 242         // Generate an arbitrary ID to identify this specific
 243         // transaction in debug logs.
 244         requestID := rand.Int31()
 245
 246         // Calculate the ordering for uploading to servers
 247         sv := NewRootSorter(this.WritableLocalRoots(), hash).GetSortedRoots()
 248
 249         // The next server to try contacting
 250         next_server := 0
 251
 252         // The number of active writers
 253         active := 0
 254
 255         // Used to communicate status from the upload goroutines
 256         upload_status := make(chan uploadStatus)
 257         defer func() {
 258                 // Wait for any abandoned uploads (e.g., we started
 259                 // two uploads and the first replied with replicas=2)
 260                 // to finish before closing the status channel.
 261                 go func() {
 262                         for active > 0 {
 263                                 <-upload_status
 264                         }
 265                         close(upload_status)
 266                 }()
 267         }()
 268
 269         // Desired number of replicas
 270         remaining_replicas := this.Want_replicas
 271
 272         replicasPerThread := this.replicasPerService
 273         if replicasPerThread < 1 {
 274                 // unlimited or unknown
 275                 replicasPerThread = remaining_replicas
 276         }
 277
 278         retriesRemaining := 1 + this.Retries
 279         var retryServers []string
 280
 281         for retriesRemaining > 0 {
 282                 retriesRemaining -= 1
 283                 next_server = 0
 284                 retryServers = []string{}
 285                 for remaining_replicas > 0 {
 286                         for active*replicasPerThread < remaining_replicas {
 287                                 // Start some upload requests
 288                                 if next_server < len(sv) {
 289                                         log.Printf("[%08x] Begin upload %s to %s", requestID, hash, sv[next_server])
 290                                         go this.uploadToKeepServer(sv[next_server], hash, tr.MakeStreamReader(), upload_status, expectedLength, requestID)
 291                                         next_server += 1
 292                                         active += 1
 293                                 } else {
 294                                         if active == 0 && retriesRemaining == 0 {
 295                                                 return locator, (this.Want_replicas - remaining_replicas), InsufficientReplicasError
 296                                         } else {
 297                                                 break
 298                                         }
 299                                 }
 300                         }
 301                         log.Printf("[%08x] Replicas remaining to write: %v active uploads: %v",
 302                                 requestID, remaining_replicas, active)
 303
 304                         // Now wait for something to happen.
 305                         if active > 0 {
 306                                 status := <-upload_status
 307                                 active -= 1
 308
 309                                 if status.statusCode == 200 {
 310                                         // good news!
 311                                         remaining_replicas -= status.replicas_stored
 312                                         locator = status.response
 313                                 } else if status.statusCode == 0 || status.statusCode == 408 || status.statusCode == 429 ||
 314                                         (status.statusCode >= 500 && status.statusCode != 503) {
 315                                         // Timeout, too many requests, or other server side failure
 316                                         // Do not retry when status code is 503, which means the keep server is full
 317                                         retryServers = append(retryServers, status.url[0:strings.LastIndex(status.url, "/")])
 318                                 }
 319                         } else {
 320                                 break
 321                         }
 322                 }
 323
 324                 sv = retryServers
 325         }
 326
 327         return locator, this.Want_replicas, nil
 328 }