Merge branch 'master' into 7167-keep-rsync-test-setup
[arvados.git] / sdk / go / keepclient / support.go
index c24849e687a8d11cf2e5d2154fdd62d0e470ec83..8fabb308f7be306ea4de494d2698666a20061022 100644 (file)
@@ -1,8 +1,8 @@
-/* Internal methods to support keepclient.go */
 package keepclient
 
 import (
        "crypto/md5"
+       "encoding/json"
        "errors"
        "fmt"
        "git.curoverse.com/arvados.git/sdk/go/streamer"
@@ -11,19 +11,20 @@ import (
        "log"
        "net"
        "net/http"
-       "os"
        "strings"
        "time"
 )
 
-type keepDisk struct {
+type keepService struct {
        Uuid     string `json:"uuid"`
        Hostname string `json:"service_host"`
        Port     int    `json:"service_port"`
        SSL      bool   `json:"service_ssl_flag"`
        SvcType  string `json:"service_type"`
+       ReadOnly bool   `json:"read_only"`
 }
 
+// Md5String returns md5 hash for the bytes in the given string
 func Md5String(s string) string {
        return fmt.Sprintf("%x", md5.Sum([]byte(s)))
 }
@@ -50,12 +51,11 @@ func (this *KeepClient) setClientSettingsProxy() {
                        TLSHandshakeTimeout: 10 * time.Second,
                }
        }
-
 }
 
 // Set timeouts apply when connecting to keepstore services directly (assumed
 // to be on the local network).
-func (this *KeepClient) setClientSettingsStore() {
+func (this *KeepClient) setClientSettingsDisk() {
        if this.Client.Timeout == 0 {
                // Maximum time to wait for a complete response
                this.Client.Timeout = 20 * time.Second
@@ -77,59 +77,87 @@ func (this *KeepClient) setClientSettingsStore() {
        }
 }
 
+type svcList struct {
+       Items []keepService `json:"items"`
+}
+
+// DiscoverKeepServers gets list of available keep services from api server
 func (this *KeepClient) DiscoverKeepServers() error {
-       if prx := os.Getenv("ARVADOS_KEEP_PROXY"); prx != "" {
-               sr := map[string]string{"proxy": prx}
-               this.SetServiceRoots(sr)
-               this.Using_proxy = true
-               this.setClientSettingsProxy()
-               return nil
-       }
+       var list svcList
 
-       type svcList struct {
-               Items []keepDisk `json:"items"`
+       // Get keep services from api server
+       err := this.Arvados.Call("GET", "keep_services", "", "accessible", nil, &list)
+       if err != nil {
+               return err
        }
-       var m svcList
 
-       err := this.Arvados.Call("GET", "keep_services", "", "accessible", nil, &m)
+       return this.loadKeepServers(list)
+}
+
+// DiscoverKeepServersFromJSON gets list of available keep services from given JSON
+func (this *KeepClient) DiscoverKeepServersFromJSON(services string) error {
+       var list svcList
 
-       if err != nil {
-               if err := this.Arvados.List("keep_disks", nil, &m); err != nil {
-                       return err
-               }
+       // Load keep services from given json
+       dec := json.NewDecoder(strings.NewReader(services))
+       if err := dec.Decode(&list); err != nil {
+               return err
        }
 
-       listed := make(map[string]bool)
-       service_roots := make(map[string]string)
-
-       for _, element := range m.Items {
-               n := ""
+       return this.loadKeepServers(list)
+}
 
-               if element.SSL {
-                       n = "s"
+// loadKeepServers
+func (this *KeepClient) loadKeepServers(list svcList) error {
+       listed := make(map[string]bool)
+       localRoots := make(map[string]string)
+       gatewayRoots := make(map[string]string)
+       writableLocalRoots := make(map[string]string)
+
+       // replicasPerService is 1 for disks; unknown or unlimited otherwise
+       this.replicasPerService = 1
+       this.Using_proxy = false
+
+       for _, service := range list.Items {
+               scheme := "http"
+               if service.SSL {
+                       scheme = "https"
                }
-
-               // Construct server URL
-               url := fmt.Sprintf("http%s://%s:%d", n, element.Hostname, element.Port)
+               url := fmt.Sprintf("%s://%s:%d", scheme, service.Hostname, service.Port)
 
                // Skip duplicates
-               if !listed[url] {
-                       listed[url] = true
-                       service_roots[element.Uuid] = url
+               if listed[url] {
+                       continue
                }
-               if element.SvcType == "proxy" {
+               listed[url] = true
+
+               localRoots[service.Uuid] = url
+               if service.SvcType == "proxy" {
                        this.Using_proxy = true
                }
+
+               if service.ReadOnly == false {
+                       writableLocalRoots[service.Uuid] = url
+                       if service.SvcType != "disk" {
+                               this.replicasPerService = 0
+                       }
+               }
+
+               // Gateway services are only used when specified by
+               // UUID, so there's nothing to gain by filtering them
+               // by service type. Including all accessible services
+               // (gateway and otherwise) merely accommodates more
+               // service configurations.
+               gatewayRoots[service.Uuid] = url
        }
 
        if this.Using_proxy {
                this.setClientSettingsProxy()
        } else {
-               this.setClientSettingsStore()
+               this.setClientSettingsDisk()
        }
 
-       this.SetServiceRoots(service_roots)
-
+       this.SetServiceRoots(localRoots, writableLocalRoots, gatewayRoots)
        return nil
 }
 
@@ -169,10 +197,7 @@ func (this KeepClient) uploadToKeepServer(host string, hash string, body io.Read
 
        req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", this.Arvados.ApiToken))
        req.Header.Add("Content-Type", "application/octet-stream")
-
-       if this.Using_proxy {
-               req.Header.Add(X_Keep_Desired_Replicas, fmt.Sprint(this.Want_replicas))
-       }
+       req.Header.Add(X_Keep_Desired_Replicas, fmt.Sprint(this.Want_replicas))
 
        var resp *http.Response
        if resp, err = this.Client.Do(req); err != nil {
@@ -213,7 +238,7 @@ func (this KeepClient) putReplicas(
        requestId := fmt.Sprintf("%x", md5.Sum([]byte(locator+time.Now().String())))[0:8]
 
        // Calculate the ordering for uploading to servers
-       sv := NewRootSorter(this.ServiceRoots(), hash).GetSortedRoots()
+       sv := NewRootSorter(this.WritableLocalRoots(), hash).GetSortedRoots()
 
        // The next server to try contacting
        next_server := 0
@@ -223,13 +248,29 @@ func (this KeepClient) putReplicas(
 
        // Used to communicate status from the upload goroutines
        upload_status := make(chan uploadStatus)
-       defer close(upload_status)
+       defer func() {
+               // Wait for any abandoned uploads (e.g., we started
+               // two uploads and the first replied with replicas=2)
+               // to finish before closing the status channel.
+               go func() {
+                       for active > 0 {
+                               <-upload_status
+                       }
+                       close(upload_status)
+               }()
+       }()
 
        // Desired number of replicas
        remaining_replicas := this.Want_replicas
 
+       replicasPerThread := this.replicasPerService
+       if replicasPerThread < 1 {
+               // unlimited or unknown
+               replicasPerThread = remaining_replicas
+       }
+
        for remaining_replicas > 0 {
-               for active < remaining_replicas {
+               for active*replicasPerThread < remaining_replicas {
                        // Start some upload requests
                        if next_server < len(sv) {
                                log.Printf("[%v] Begin upload %s to %s", requestId, hash, sv[next_server])