sdk/go/keepclient/support.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: Apache-2.0
   4
   5 package keepclient
   6
   7 import (
   8         "crypto/md5"
   9         "errors"
  10         "fmt"
  11         "io"
  12         "io/ioutil"
  13         "log"
  14         "net/http"
  15         "os"
  16         "strconv"
  17         "strings"
  18
  19         "git.arvados.org/arvados.git/sdk/go/arvadosclient"
  20 )
  21
  22 // DebugPrintf emits debug messages. The easiest way to enable
  23 // keepclient debug messages in your application is to assign
  24 // log.Printf to DebugPrintf.
  25 var DebugPrintf = func(string, ...interface{}) {}
  26
  27 func init() {
  28         if arvadosclient.StringBool(os.Getenv("ARVADOS_DEBUG")) {
  29                 DebugPrintf = log.Printf
  30         }
  31 }
  32
  33 type keepService struct {
  34         Uuid     string `json:"uuid"`
  35         Hostname string `json:"service_host"`
  36         Port     int    `json:"service_port"`
  37         SSL      bool   `json:"service_ssl_flag"`
  38         SvcType  string `json:"service_type"`
  39         ReadOnly bool   `json:"read_only"`
  40 }
  41
  42 // Md5String returns md5 hash for the bytes in the given string
  43 func Md5String(s string) string {
  44         return fmt.Sprintf("%x", md5.Sum([]byte(s)))
  45 }
  46
  47 type svcList struct {
  48         Items []keepService `json:"items"`
  49 }
  50
  51 type uploadStatus struct {
  52         err            error
  53         url            string
  54         statusCode     int
  55         replicasStored int
  56         classesStored  map[string]int
  57         response       string
  58 }
  59
  60 func (kc *KeepClient) uploadToKeepServer(host string, hash string, classesTodo []string, body io.Reader,
  61         uploadStatusChan chan<- uploadStatus, expectedLength int64, reqid string) {
  62
  63         var req *http.Request
  64         var err error
  65         var url = fmt.Sprintf("%s/%s", host, hash)
  66         if req, err = http.NewRequest("PUT", url, nil); err != nil {
  67                 DebugPrintf("DEBUG: [%s] Error creating request PUT %v error: %v", reqid, url, err.Error())
  68                 uploadStatusChan <- uploadStatus{err, url, 0, 0, nil, ""}
  69                 return
  70         }
  71
  72         req.ContentLength = expectedLength
  73         if expectedLength > 0 {
  74                 req.Body = ioutil.NopCloser(body)
  75         } else {
  76                 // "For client requests, a value of 0 means unknown if
  77                 // Body is not nil."  In this case we do want the body
  78                 // to be empty, so don't set req.Body.
  79         }
  80
  81         req.Header.Add("X-Request-Id", reqid)
  82         req.Header.Add("Authorization", "OAuth2 "+kc.Arvados.ApiToken)
  83         req.Header.Add("Content-Type", "application/octet-stream")
  84         req.Header.Add(XKeepDesiredReplicas, fmt.Sprint(kc.Want_replicas))
  85         if len(classesTodo) > 0 {
  86                 req.Header.Add(XKeepStorageClasses, strings.Join(classesTodo, ", "))
  87         }
  88
  89         var resp *http.Response
  90         if resp, err = kc.httpClient().Do(req); err != nil {
  91                 DebugPrintf("DEBUG: [%s] Upload failed %v error: %v", reqid, url, err.Error())
  92                 uploadStatusChan <- uploadStatus{err, url, 0, 0, nil, err.Error()}
  93                 return
  94         }
  95
  96         rep := 1
  97         if xr := resp.Header.Get(XKeepReplicasStored); xr != "" {
  98                 fmt.Sscanf(xr, "%d", &rep)
  99         }
 100         scc := resp.Header.Get(XKeepStorageClassesConfirmed)
 101         classesStored, err := parseStorageClassesConfirmedHeader(scc)
 102         if err != nil {
 103                 DebugPrintf("DEBUG: [%s] Ignoring invalid %s header %q: %s", reqid, XKeepStorageClassesConfirmed, scc, err)
 104         }
 105
 106         defer resp.Body.Close()
 107         defer io.Copy(ioutil.Discard, resp.Body)
 108
 109         respbody, err2 := ioutil.ReadAll(&io.LimitedReader{R: resp.Body, N: 4096})
 110         response := strings.TrimSpace(string(respbody))
 111         if err2 != nil && err2 != io.EOF {
 112                 DebugPrintf("DEBUG: [%s] Upload %v error: %v response: %v", reqid, url, err2.Error(), response)
 113                 uploadStatusChan <- uploadStatus{err2, url, resp.StatusCode, rep, classesStored, response}
 114         } else if resp.StatusCode == http.StatusOK {
 115                 DebugPrintf("DEBUG: [%s] Upload %v success", reqid, url)
 116                 uploadStatusChan <- uploadStatus{nil, url, resp.StatusCode, rep, classesStored, response}
 117         } else {
 118                 if resp.StatusCode >= 300 && response == "" {
 119                         response = resp.Status
 120                 }
 121                 DebugPrintf("DEBUG: [%s] Upload %v error: %v response: %v", reqid, url, resp.StatusCode, response)
 122                 uploadStatusChan <- uploadStatus{errors.New(resp.Status), url, resp.StatusCode, rep, classesStored, response}
 123         }
 124 }
 125
 126 func (kc *KeepClient) putReplicas(
 127         hash string,
 128         getReader func() io.Reader,
 129         expectedLength int64) (locator string, replicas int, err error) {
 130
 131         reqid := kc.getRequestID()
 132
 133         // Calculate the ordering for uploading to servers
 134         sv := NewRootSorter(kc.WritableLocalRoots(), hash).GetSortedRoots()
 135
 136         // The next server to try contacting
 137         nextServer := 0
 138
 139         // The number of active writers
 140         active := 0
 141
 142         // Used to communicate status from the upload goroutines
 143         uploadStatusChan := make(chan uploadStatus)
 144         defer func() {
 145                 // Wait for any abandoned uploads (e.g., we started
 146                 // two uploads and the first replied with replicas=2)
 147                 // to finish before closing the status channel.
 148                 go func() {
 149                         for active > 0 {
 150                                 <-uploadStatusChan
 151                         }
 152                         close(uploadStatusChan)
 153                 }()
 154         }()
 155
 156         replicasWanted := kc.Want_replicas
 157         replicasTodo := map[string]int{}
 158         for _, c := range kc.StorageClasses {
 159                 replicasTodo[c] = replicasWanted
 160         }
 161         replicasDone := 0
 162
 163         replicasPerThread := kc.replicasPerService
 164         if replicasPerThread < 1 {
 165                 // unlimited or unknown
 166                 replicasPerThread = replicasWanted
 167         }
 168
 169         retriesRemaining := 1 + kc.Retries
 170         var retryServers []string
 171
 172         lastError := make(map[string]string)
 173         trackingClasses := len(replicasTodo) > 0
 174
 175         for retriesRemaining > 0 {
 176                 retriesRemaining--
 177                 nextServer = 0
 178                 retryServers = []string{}
 179                 for {
 180                         var classesTodo []string
 181                         var maxConcurrency int
 182                         for sc, r := range replicasTodo {
 183                                 classesTodo = append(classesTodo, sc)
 184                                 if maxConcurrency == 0 || maxConcurrency > r {
 185                                         // Having more than r
 186                                         // writes in flight
 187                                         // would overreplicate
 188                                         // class sc.
 189                                         maxConcurrency = r
 190                                 }
 191                         }
 192                         if !trackingClasses {
 193                                 maxConcurrency = replicasWanted - replicasDone
 194                         }
 195                         if maxConcurrency < 1 {
 196                                 // If there are no non-zero entries in
 197                                 // replicasTodo, we're done.
 198                                 break
 199                         }
 200                         for active*replicasPerThread < maxConcurrency {
 201                                 // Start some upload requests
 202                                 if nextServer < len(sv) {
 203                                         DebugPrintf("DEBUG: [%s] Begin upload %s to %s", reqid, hash, sv[nextServer])
 204                                         go kc.uploadToKeepServer(sv[nextServer], hash, classesTodo, getReader(), uploadStatusChan, expectedLength, reqid)
 205                                         nextServer++
 206                                         active++
 207                                 } else {
 208                                         if active == 0 && retriesRemaining == 0 {
 209                                                 msg := "Could not write sufficient replicas: "
 210                                                 for _, resp := range lastError {
 211                                                         msg += resp + "; "
 212                                                 }
 213                                                 msg = msg[:len(msg)-2]
 214                                                 return locator, replicasDone, InsufficientReplicasError(errors.New(msg))
 215                                         }
 216                                         break
 217                                 }
 218                         }
 219
 220                         DebugPrintf("DEBUG: [%s] Replicas remaining to write: %v active uploads: %v", reqid, replicasTodo, active)
 221                         if active < 1 {
 222                                 break
 223                         }
 224
 225                         // Wait for something to happen.
 226                         status := <-uploadStatusChan
 227                         active--
 228
 229                         if status.statusCode == http.StatusOK {
 230                                 delete(lastError, status.url)
 231                                 replicasDone += status.replicasStored
 232                                 if len(status.classesStored) == 0 {
 233                                         // Server doesn't report
 234                                         // storage classes. Give up
 235                                         // trying to track which ones
 236                                         // are satisfied; just rely on
 237                                         // total # replicas.
 238                                         trackingClasses = false
 239                                 }
 240                                 for className, replicas := range status.classesStored {
 241                                         if replicasTodo[className] > replicas {
 242                                                 replicasTodo[className] -= replicas
 243                                         } else {
 244                                                 delete(replicasTodo, className)
 245                                         }
 246                                 }
 247                                 locator = status.response
 248                         } else {
 249                                 msg := fmt.Sprintf("[%d] %s", status.statusCode, status.response)
 250                                 if len(msg) > 100 {
 251                                         msg = msg[:100]
 252                                 }
 253                                 lastError[status.url] = msg
 254                         }
 255
 256                         if status.statusCode == 0 || status.statusCode == 408 || status.statusCode == 429 ||
 257                                 (status.statusCode >= 500 && status.statusCode != 503) {
 258                                 // Timeout, too many requests, or other server side failure
 259                                 // Do not retry when status code is 503, which means the keep server is full
 260                                 retryServers = append(retryServers, status.url[0:strings.LastIndex(status.url, "/")])
 261                         }
 262                 }
 263
 264                 sv = retryServers
 265         }
 266
 267         return locator, replicasDone, nil
 268 }
 269
 270 func parseStorageClassesConfirmedHeader(hdr string) (map[string]int, error) {
 271         if hdr == "" {
 272                 return nil, nil
 273         }
 274         classesStored := map[string]int{}
 275         for _, cr := range strings.Split(hdr, ",") {
 276                 cr = strings.TrimSpace(cr)
 277                 if cr == "" {
 278                         continue
 279                 }
 280                 fields := strings.SplitN(cr, "=", 2)
 281                 if len(fields) != 2 {
 282                         return nil, fmt.Errorf("expected exactly one '=' char in entry %q", cr)
 283                 }
 284                 className := fields[0]
 285                 if className == "" {
 286                         return nil, fmt.Errorf("empty class name in entry %q", cr)
 287                 }
 288                 replicas, err := strconv.Atoi(fields[1])
 289                 if err != nil || replicas < 1 {
 290                         return nil, fmt.Errorf("invalid replica count %q", fields[1])
 291                 }
 292                 classesStored[className] = replicas
 293         }
 294         return classesStored, nil
 295 }