X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/ca35c58c19b28dcdbbff71fe3da0ee16ca760673..af6d31cba8346ac86bc0027eb0f675144fb43056:/sdk/go/keepclient/support.go diff --git a/sdk/go/keepclient/support.go b/sdk/go/keepclient/support.go index b467d06b21..8414afab1e 100644 --- a/sdk/go/keepclient/support.go +++ b/sdk/go/keepclient/support.go @@ -2,19 +2,26 @@ package keepclient import ( "crypto/md5" + "encoding/json" "errors" "fmt" "git.curoverse.com/arvados.git/sdk/go/streamer" "io" "io/ioutil" "log" + "math/rand" "net" "net/http" "strings" "time" ) -type keepDisk struct { +// Function used to emit debug messages. The easiest way to enable +// keepclient debug messages in your application is to assign +// log.Printf to DebugPrintf. +var DebugPrintf = func(string, ...interface{}) {} + +type keepService struct { Uuid string `json:"uuid"` Hostname string `json:"service_host"` Port int `json:"service_port"` @@ -23,6 +30,7 @@ type keepDisk struct { 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))) } @@ -49,12 +57,11 @@ 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 @@ -76,26 +83,48 @@ 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 { - type svcList struct { - Items []keepDisk `json:"items"` + var list svcList + + // 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) +} - if err != nil { - if err := this.Arvados.List("keep_disks", nil, &m); err != nil { - return err - } +// LoadKeepServicesFromJSON gets list of available keep services from given JSON +func (this *KeepClient) LoadKeepServicesFromJSON(services string) error { + var list svcList + + // Load keep services from given json + dec := json.NewDecoder(strings.NewReader(services)) + if err := dec.Decode(&list); err != nil { + return err } + return this.loadKeepServers(list) +} + +// 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) - for _, service := range m.Items { + // 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" @@ -108,16 +137,16 @@ func (this *KeepClient) DiscoverKeepServers() error { } listed[url] = true - switch service.SvcType { - case "disk": - localRoots[service.Uuid] = url - case "proxy": - localRoots[service.Uuid] = url + 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 @@ -131,7 +160,7 @@ func (this *KeepClient) DiscoverKeepServers() error { if this.Using_proxy { this.setClientSettingsProxy() } else { - this.setClientSettingsStore() + this.setClientSettingsDisk() } this.SetServiceRoots(localRoots, writableLocalRoots, gatewayRoots) @@ -146,14 +175,14 @@ type uploadStatus struct { response string } -func (this KeepClient) uploadToKeepServer(host string, hash string, body io.ReadCloser, - upload_status chan<- uploadStatus, expectedLength int64, requestId string) { +func (this *KeepClient) uploadToKeepServer(host string, hash string, body io.ReadCloser, + upload_status chan<- uploadStatus, expectedLength int64, requestID int32) { var req *http.Request var err error var url = fmt.Sprintf("%s/%s", host, hash) if req, err = http.NewRequest("PUT", url, nil); err != nil { - log.Printf("[%v] Error creating request PUT %v error: %v", requestId, url, err.Error()) + log.Printf("[%08x] Error creating request PUT %v error: %v", requestID, url, err.Error()) upload_status <- uploadStatus{err, url, 0, 0, ""} body.Close() return @@ -174,14 +203,11 @@ 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 { - log.Printf("[%v] Upload failed %v error: %v", requestId, url, err.Error()) + log.Printf("[%08x] Upload failed %v error: %v", requestID, url, err.Error()) upload_status <- uploadStatus{err, url, 0, 0, ""} return } @@ -194,28 +220,28 @@ func (this KeepClient) uploadToKeepServer(host string, hash string, body io.Read defer resp.Body.Close() defer io.Copy(ioutil.Discard, resp.Body) - respbody, err2 := ioutil.ReadAll(&io.LimitedReader{resp.Body, 4096}) + respbody, err2 := ioutil.ReadAll(&io.LimitedReader{R: resp.Body, N: 4096}) response := strings.TrimSpace(string(respbody)) if err2 != nil && err2 != io.EOF { - log.Printf("[%v] Upload %v error: %v response: %v", requestId, url, err2.Error(), response) + log.Printf("[%08x] Upload %v error: %v response: %v", requestID, url, err2.Error(), response) upload_status <- uploadStatus{err2, url, resp.StatusCode, rep, response} } else if resp.StatusCode == http.StatusOK { - log.Printf("[%v] Upload %v success", requestId, url) + log.Printf("[%08x] Upload %v success", requestID, url) upload_status <- uploadStatus{nil, url, resp.StatusCode, rep, response} } else { - log.Printf("[%v] Upload %v error: %v response: %v", requestId, url, resp.StatusCode, response) + log.Printf("[%08x] Upload %v error: %v response: %v", requestID, url, resp.StatusCode, response) upload_status <- uploadStatus{errors.New(resp.Status), url, resp.StatusCode, rep, response} } } -func (this KeepClient) putReplicas( +func (this *KeepClient) putReplicas( hash string, tr *streamer.AsyncStream, expectedLength int64) (locator string, replicas int, err error) { - // Take the hash of locator and timestamp in order to identify this - // specific transaction in log statements. - requestId := fmt.Sprintf("%x", md5.Sum([]byte(locator+time.Now().String())))[0:8] + // Generate an arbitrary ID to identify this specific + // transaction in debug logs. + requestID := rand.Int31() // Calculate the ordering for uploading to servers sv := NewRootSorter(this.WritableLocalRoots(), hash).GetSortedRoots() @@ -228,39 +254,74 @@ 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 - for remaining_replicas > 0 { - for active < remaining_replicas { - // Start some upload requests - if next_server < len(sv) { - log.Printf("[%v] Begin upload %s to %s", requestId, hash, sv[next_server]) - go this.uploadToKeepServer(sv[next_server], hash, tr.MakeStreamReader(), upload_status, expectedLength, requestId) - next_server += 1 - active += 1 - } else { - if active == 0 { - return locator, (this.Want_replicas - remaining_replicas), InsufficientReplicasError + replicasPerThread := this.replicasPerService + if replicasPerThread < 1 { + // unlimited or unknown + replicasPerThread = remaining_replicas + } + + retriesRemaining := 1 + this.Retries + var retryServers []string + + for retriesRemaining > 0 { + retriesRemaining -= 1 + next_server = 0 + retryServers = []string{} + for remaining_replicas > 0 { + for active*replicasPerThread < remaining_replicas { + // Start some upload requests + if next_server < len(sv) { + log.Printf("[%08x] Begin upload %s to %s", requestID, hash, sv[next_server]) + go this.uploadToKeepServer(sv[next_server], hash, tr.MakeStreamReader(), upload_status, expectedLength, requestID) + next_server += 1 + active += 1 } else { - break + if active == 0 && retriesRemaining == 0 { + return locator, (this.Want_replicas - remaining_replicas), InsufficientReplicasError + } else { + break + } + } + } + log.Printf("[%08x] Replicas remaining to write: %v active uploads: %v", + requestID, remaining_replicas, active) + + // Now wait for something to happen. + if active > 0 { + status := <-upload_status + active -= 1 + + if status.statusCode == 200 { + // good news! + remaining_replicas -= status.replicas_stored + locator = status.response + } else if status.statusCode == 0 || status.statusCode == 408 || status.statusCode == 429 || + (status.statusCode >= 500 && status.statusCode != 503) { + // Timeout, too many requests, or other server side failure + // Do not retry when status code is 503, which means the keep server is full + retryServers = append(retryServers, status.url[0:strings.LastIndex(status.url, "/")]) } + } else { + break } } - log.Printf("[%v] Replicas remaining to write: %v active uploads: %v", - requestId, remaining_replicas, active) - - // Now wait for something to happen. - status := <-upload_status - active -= 1 - if status.statusCode == 200 { - // good news! - remaining_replicas -= status.replicas_stored - locator = status.response - } + sv = retryServers } return locator, this.Want_replicas, nil