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