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         "net"
  13         "net/http"
  14         "strings"
  15         "time"
  16 )
  17
  18 type keepService struct {
  19         Uuid     string `json:"uuid"`
  20         Hostname string `json:"service_host"`
  21         Port     int    `json:"service_port"`
  22         SSL      bool   `json:"service_ssl_flag"`
  23         SvcType  string `json:"service_type"`
  24         ReadOnly bool   `json:"read_only"`
  25 }
  26
  27 // Md5String returns md5 hash for the bytes in the given string
  28 func Md5String(s string) string {
  29         return fmt.Sprintf("%x", md5.Sum([]byte(s)))
  30 }
  31
  32 // Set timeouts apply when connecting to keepproxy services (assumed to be over
  33 // the Internet).
  34 func (this *KeepClient) setClientSettingsProxy() {
  35         if this.Client.Timeout == 0 {
  36                 // Maximum time to wait for a complete response
  37                 this.Client.Timeout = 300 * time.Second
  38
  39                 // TCP and TLS connection settings
  40                 this.Client.Transport = &http.Transport{
  41                         Dial: (&net.Dialer{
  42                                 // The maximum time to wait to set up
  43                                 // the initial TCP connection.
  44                                 Timeout: 30 * time.Second,
  45
  46                                 // The TCP keep alive heartbeat
  47                                 // interval.
  48                                 KeepAlive: 120 * time.Second,
  49                         }).Dial,
  50
  51                         TLSHandshakeTimeout: 10 * time.Second,
  52                 }
  53         }
  54 }
  55
  56 // Set timeouts apply when connecting to keepstore services directly (assumed
  57 // to be on the local network).
  58 func (this *KeepClient) setClientSettingsDisk() {
  59         if this.Client.Timeout == 0 {
  60                 // Maximum time to wait for a complete response
  61                 this.Client.Timeout = 20 * time.Second
  62
  63                 // TCP and TLS connection timeouts
  64                 this.Client.Transport = &http.Transport{
  65                         Dial: (&net.Dialer{
  66                                 // The maximum time to wait to set up
  67                                 // the initial TCP connection.
  68                                 Timeout: 2 * time.Second,
  69
  70                                 // The TCP keep alive heartbeat
  71                                 // interval.
  72                                 KeepAlive: 180 * time.Second,
  73                         }).Dial,
  74
  75                         TLSHandshakeTimeout: 4 * time.Second,
  76                 }
  77         }
  78 }
  79
  80 type svcList struct {
  81         Items []keepService `json:"items"`
  82 }
  83
  84 // DiscoverKeepServers gets list of available keep services from api server
  85 func (this *KeepClient) DiscoverKeepServers() error {
  86         var list svcList
  87
  88         // Get keep services from api server
  89         err := this.Arvados.Call("GET", "keep_services", "", "accessible", nil, &list)
  90         if err != nil {
  91                 return err
  92         }
  93
  94         return this.loadKeepServers(list)
  95 }
  96
  97 // LoadKeepServicesFromJSON gets list of available keep services from given JSON
  98 func (this *KeepClient) LoadKeepServicesFromJSON(services string) error {
  99         var list svcList
 100
 101         // Load keep services from given json
 102         dec := json.NewDecoder(strings.NewReader(services))
 103         if err := dec.Decode(&list); err != nil {
 104                 return err
 105         }
 106
 107         return this.loadKeepServers(list)
 108 }
 109
 110 // loadKeepServers
 111 func (this *KeepClient) loadKeepServers(list svcList) error {
 112         listed := make(map[string]bool)
 113         localRoots := make(map[string]string)
 114         gatewayRoots := make(map[string]string)
 115         writableLocalRoots := make(map[string]string)
 116
 117         // replicasPerService is 1 for disks; unknown or unlimited otherwise
 118         this.replicasPerService = 1
 119         this.Using_proxy = false
 120
 121         for _, service := range list.Items {
 122                 scheme := "http"
 123                 if service.SSL {
 124                         scheme = "https"
 125                 }
 126                 url := fmt.Sprintf("%s://%s:%d", scheme, service.Hostname, service.Port)
 127
 128                 // Skip duplicates
 129                 if listed[url] {
 130                         continue
 131                 }
 132                 listed[url] = true
 133
 134                 localRoots[service.Uuid] = url
 135                 if service.SvcType == "proxy" {
 136                         this.Using_proxy = true
 137                 }
 138
 139                 if service.ReadOnly == false {
 140                         writableLocalRoots[service.Uuid] = url
 141                         if service.SvcType != "disk" {
 142                                 this.replicasPerService = 0
 143                         }
 144                 }
 145
 146                 // Gateway services are only used when specified by
 147                 // UUID, so there's nothing to gain by filtering them
 148                 // by service type. Including all accessible services
 149                 // (gateway and otherwise) merely accommodates more
 150                 // service configurations.
 151                 gatewayRoots[service.Uuid] = url
 152         }
 153
 154         if this.Using_proxy {
 155                 this.setClientSettingsProxy()
 156         } else {
 157                 this.setClientSettingsDisk()
 158         }
 159
 160         this.SetServiceRoots(localRoots, writableLocalRoots, gatewayRoots)
 161         return nil
 162 }
 163
 164 type uploadStatus struct {
 165         err             error
 166         url             string
 167         statusCode      int
 168         replicas_stored int
 169         response        string
 170 }
 171
 172 func (this *KeepClient) uploadToKeepServer(host string, hash string, body io.ReadCloser,
 173         upload_status chan<- uploadStatus, expectedLength int64, requestId string) {
 174
 175         var req *http.Request
 176         var err error
 177         var url = fmt.Sprintf("%s/%s", host, hash)
 178         if req, err = http.NewRequest("PUT", url, nil); err != nil {
 179                 log.Printf("[%v] Error creating request PUT %v error: %v", requestId, url, err.Error())
 180                 upload_status <- uploadStatus{err, url, 0, 0, ""}
 181                 body.Close()
 182                 return
 183         }
 184
 185         req.ContentLength = expectedLength
 186         if expectedLength > 0 {
 187                 // http.Client.Do will close the body ReadCloser when it is
 188                 // done with it.
 189                 req.Body = body
 190         } else {
 191                 // "For client requests, a value of 0 means unknown if Body is
 192                 // not nil."  In this case we do want the body to be empty, so
 193                 // don't set req.Body.  However, we still need to close the
 194                 // body ReadCloser.
 195                 body.Close()
 196         }
 197
 198         req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", this.Arvados.ApiToken))
 199         req.Header.Add("Content-Type", "application/octet-stream")
 200         req.Header.Add(X_Keep_Desired_Replicas, fmt.Sprint(this.Want_replicas))
 201
 202         var resp *http.Response
 203         if resp, err = this.Client.Do(req); err != nil {
 204                 log.Printf("[%v] Upload failed %v error: %v", requestId, url, err.Error())
 205                 upload_status <- uploadStatus{err, url, 0, 0, ""}
 206                 return
 207         }
 208
 209         rep := 1
 210         if xr := resp.Header.Get(X_Keep_Replicas_Stored); xr != "" {
 211                 fmt.Sscanf(xr, "%d", &rep)
 212         }
 213
 214         defer resp.Body.Close()
 215         defer io.Copy(ioutil.Discard, resp.Body)
 216
 217         respbody, err2 := ioutil.ReadAll(&io.LimitedReader{R: resp.Body, N: 4096})
 218         response := strings.TrimSpace(string(respbody))
 219         if err2 != nil && err2 != io.EOF {
 220                 log.Printf("[%v] Upload %v error: %v response: %v", requestId, url, err2.Error(), response)
 221                 upload_status <- uploadStatus{err2, url, resp.StatusCode, rep, response}
 222         } else if resp.StatusCode == http.StatusOK {
 223                 log.Printf("[%v] Upload %v success", requestId, url)
 224                 upload_status <- uploadStatus{nil, url, resp.StatusCode, rep, response}
 225         } else {
 226                 log.Printf("[%v] Upload %v error: %v response: %v", requestId, url, resp.StatusCode, response)
 227                 upload_status <- uploadStatus{errors.New(resp.Status), url, resp.StatusCode, rep, response}
 228         }
 229 }
 230
 231 func (this *KeepClient) putReplicas(
 232         hash string,
 233         tr *streamer.AsyncStream,
 234         expectedLength int64) (locator string, replicas int, err error) {
 235
 236         // Take the hash of locator and timestamp in order to identify this
 237         // specific transaction in log statements.
 238         requestId := fmt.Sprintf("%x", md5.Sum([]byte(hash+time.Now().String())))[0:8]
 239
 240         // Calculate the ordering for uploading to servers
 241         sv := NewRootSorter(this.WritableLocalRoots(), hash).GetSortedRoots()
 242
 243         // The next server to try contacting
 244         next_server := 0
 245
 246         // The number of active writers
 247         active := 0
 248
 249         // Used to communicate status from the upload goroutines
 250         upload_status := make(chan uploadStatus)
 251         defer func() {
 252                 // Wait for any abandoned uploads (e.g., we started
 253                 // two uploads and the first replied with replicas=2)
 254                 // to finish before closing the status channel.
 255                 go func() {
 256                         for active > 0 {
 257                                 <-upload_status
 258                         }
 259                         close(upload_status)
 260                 }()
 261         }()
 262
 263         // Desired number of replicas
 264         remaining_replicas := this.Want_replicas
 265
 266         replicasPerThread := this.replicasPerService
 267         if replicasPerThread < 1 {
 268                 // unlimited or unknown
 269                 replicasPerThread = remaining_replicas
 270         }
 271
 272         retriesRemaining := 1 + this.Retries
 273         var retryServers []string
 274
 275         for retriesRemaining > 0 {
 276                 retriesRemaining -= 1
 277                 next_server = 0
 278                 retryServers = []string{}
 279                 for remaining_replicas > 0 {
 280                         for active*replicasPerThread < remaining_replicas {
 281                                 // Start some upload requests
 282                                 if next_server < len(sv) {
 283                                         log.Printf("[%v] Begin upload %s to %s", requestId, hash, sv[next_server])
 284                                         go this.uploadToKeepServer(sv[next_server], hash, tr.MakeStreamReader(), upload_status, expectedLength, requestId)
 285                                         next_server += 1
 286                                         active += 1
 287                                 } else {
 288                                         if active == 0 && retriesRemaining == 0 {
 289                                                 return locator, (this.Want_replicas - remaining_replicas), InsufficientReplicasError
 290                                         } else {
 291                                                 break
 292                                         }
 293                                 }
 294                         }
 295                         log.Printf("[%v] Replicas remaining to write: %v active uploads: %v",
 296                                 requestId, remaining_replicas, active)
 297
 298                         // Now wait for something to happen.
 299                         if active > 0 {
 300                                 status := <-upload_status
 301                                 active -= 1
 302
 303                                 if status.statusCode == 200 {
 304                                         // good news!
 305                                         remaining_replicas -= status.replicas_stored
 306                                         locator = status.response
 307                                 } else if status.statusCode == 0 || status.statusCode == 408 || status.statusCode == 429 ||
 308                                         (status.statusCode >= 500 && status.statusCode != 503) {
 309                                         // Timeout, too many requests, or other server side failure
 310                                         // Do not retry when status code is 503, which means the keep server is full
 311                                         retryServers = append(retryServers, status.url[0:strings.LastIndex(status.url, "/")])
 312                                 }
 313                         } else {
 314                                 break
 315                         }
 316                 }
 317
 318                 sv = retryServers
 319         }
 320
 321         return locator, this.Want_replicas, nil
 322 }