X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/f5ab6191a96ce31defe0269b9bc739c6492c1a90..ed7698518d7eabea0f4e86acfb0660765fe31071:/sdk/go/keepclient/support.go diff --git a/sdk/go/keepclient/support.go b/sdk/go/keepclient/support.go index bd9b2e35f7..542827f5e0 100644 --- a/sdk/go/keepclient/support.go +++ b/sdk/go/keepclient/support.go @@ -1,20 +1,34 @@ +// Copyright (C) The Arvados Authors. All rights reserved. +// +// SPDX-License-Identifier: Apache-2.0 + package keepclient import ( "crypto/md5" - "encoding/json" "errors" "fmt" - "git.curoverse.com/arvados.git/sdk/go/streamer" "io" "io/ioutil" "log" - "net" "net/http" + "os" "strings" - "time" + + "git.curoverse.com/arvados.git/sdk/go/arvadosclient" ) +// 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{}) {} + +func init() { + if arvadosclient.StringBool(os.Getenv("ARVADOS_DEBUG")) { + DebugPrintf = log.Printf + } +} + type keepService struct { Uuid string `json:"uuid"` Hostname string `json:"service_host"` @@ -29,138 +43,10 @@ func Md5String(s string) string { return fmt.Sprintf("%x", md5.Sum([]byte(s))) } -// Set timeouts apply when connecting to keepproxy services (assumed to be over -// the Internet). -func (this *KeepClient) setClientSettingsProxy() { - if this.Client.Timeout == 0 { - // Maximum time to wait for a complete response - this.Client.Timeout = 300 * time.Second - - // TCP and TLS connection settings - this.Client.Transport = &http.Transport{ - Dial: (&net.Dialer{ - // The maximum time to wait to set up - // the initial TCP connection. - Timeout: 30 * time.Second, - - // The TCP keep alive heartbeat - // interval. - KeepAlive: 120 * time.Second, - }).Dial, - - TLSHandshakeTimeout: 10 * time.Second, - } - } -} - -// Set timeouts apply when connecting to keepstore services directly (assumed -// to be on the local network). -func (this *KeepClient) setClientSettingsDisk() { - if this.Client.Timeout == 0 { - // Maximum time to wait for a complete response - this.Client.Timeout = 20 * time.Second - - // TCP and TLS connection timeouts - this.Client.Transport = &http.Transport{ - Dial: (&net.Dialer{ - // The maximum time to wait to set up - // the initial TCP connection. - Timeout: 2 * time.Second, - - // The TCP keep alive heartbeat - // interval. - KeepAlive: 180 * time.Second, - }).Dial, - - TLSHandshakeTimeout: 4 * time.Second, - } - } -} - type svcList struct { Items []keepService `json:"items"` } -// DiscoverKeepServers gets list of available keep services from api server -func (this *KeepClient) DiscoverKeepServers() error { - var list svcList - - // Get keep services from api server - err := this.Arvados.Call("GET", "keep_services", "", "accessible", nil, &list) - if err != nil { - return err - } - - return this.loadKeepServers(list) -} - -// 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) - - // 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" - } - url := fmt.Sprintf("%s://%s:%d", scheme, service.Hostname, service.Port) - - // Skip duplicates - if listed[url] { - continue - } - 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.setClientSettingsDisk() - } - - this.SetServiceRoots(localRoots, writableLocalRoots, gatewayRoots) - return nil -} - type uploadStatus struct { err error url string @@ -169,39 +55,38 @@ 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.Reader, + upload_status chan<- uploadStatus, expectedLength int64, reqid string) { 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()) + DebugPrintf("DEBUG: [%s] Error creating request PUT %v error: %v", reqid, url, err.Error()) upload_status <- uploadStatus{err, url, 0, 0, ""} - body.Close() return } req.ContentLength = expectedLength if expectedLength > 0 { - // http.Client.Do will close the body ReadCloser when it is - // done with it. - req.Body = body + req.Body = ioutil.NopCloser(body) } else { - // "For client requests, a value of 0 means unknown if Body is - // not nil." In this case we do want the body to be empty, so - // don't set req.Body. However, we still need to close the - // body ReadCloser. - body.Close() + // "For client requests, a value of 0 means unknown if + // Body is not nil." In this case we do want the body + // to be empty, so don't set req.Body. } - req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", this.Arvados.ApiToken)) + req.Header.Add("X-Request-Id", reqid) + req.Header.Add("Authorization", "OAuth2 "+this.Arvados.ApiToken) req.Header.Add("Content-Type", "application/octet-stream") req.Header.Add(X_Keep_Desired_Replicas, fmt.Sprint(this.Want_replicas)) + if len(this.StorageClasses) > 0 { + req.Header.Add("X-Keep-Storage-Classes", strings.Join(this.StorageClasses, ", ")) + } 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()) + if resp, err = this.httpClient().Do(req); err != nil { + DebugPrintf("DEBUG: [%s] Upload failed %v error: %v", reqid, url, err.Error()) upload_status <- uploadStatus{err, url, 0, 0, ""} return } @@ -214,28 +99,29 @@ 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) + DebugPrintf("DEBUG: [%s] Upload %v error: %v response: %v", reqid, 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) + DebugPrintf("DEBUG: [%s] Upload %v success", reqid, 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) + if resp.StatusCode >= 300 && response == "" { + response = resp.Status + } + DebugPrintf("DEBUG: [%s] Upload %v error: %v response: %v", reqid, 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, + getReader func() io.Reader, 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(hash+time.Now().String())))[0:8] + reqid := this.getRequestID() // Calculate the ordering for uploading to servers sv := NewRootSorter(this.WritableLocalRoots(), hash).GetSortedRoots() @@ -260,40 +146,47 @@ func (this KeepClient) putReplicas( }() }() - // Desired number of replicas - remaining_replicas := this.Want_replicas + replicasDone := 0 + replicasTodo := this.Want_replicas replicasPerThread := this.replicasPerService if replicasPerThread < 1 { // unlimited or unknown - replicasPerThread = remaining_replicas + replicasPerThread = replicasTodo } retriesRemaining := 1 + this.Retries var retryServers []string + lastError := make(map[string]string) + for retriesRemaining > 0 { retriesRemaining -= 1 next_server = 0 retryServers = []string{} - for remaining_replicas > 0 { - for active*replicasPerThread < remaining_replicas { + for replicasTodo > 0 { + for active*replicasPerThread < replicasTodo { // 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) + DebugPrintf("DEBUG: [%s] Begin upload %s to %s", reqid, hash, sv[next_server]) + go this.uploadToKeepServer(sv[next_server], hash, getReader(), upload_status, expectedLength, reqid) next_server += 1 active += 1 } else { if active == 0 && retriesRemaining == 0 { - return locator, (this.Want_replicas - remaining_replicas), InsufficientReplicasError + msg := "Could not write sufficient replicas: " + for _, resp := range lastError { + msg += resp + "; " + } + msg = msg[:len(msg)-2] + return locator, replicasDone, InsufficientReplicasError(errors.New(msg)) } else { break } } } - log.Printf("[%v] Replicas remaining to write: %v active uploads: %v", - requestId, remaining_replicas, active) + DebugPrintf("DEBUG: [%s] Replicas remaining to write: %v active uploads: %v", + reqid, replicasTodo, active) // Now wait for something to happen. if active > 0 { @@ -302,9 +195,19 @@ func (this KeepClient) putReplicas( if status.statusCode == 200 { // good news! - remaining_replicas -= status.replicas_stored + replicasDone += status.replicas_stored + replicasTodo -= status.replicas_stored locator = status.response - } else if status.statusCode == 0 || status.statusCode == 408 || status.statusCode == 429 || + delete(lastError, status.url) + } else { + msg := fmt.Sprintf("[%d] %s", status.statusCode, status.response) + if len(msg) > 100 { + msg = msg[:100] + } + lastError[status.url] = msg + } + + 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 @@ -318,5 +221,5 @@ func (this KeepClient) putReplicas( sv = retryServers } - return locator, this.Want_replicas, nil + return locator, replicasDone, nil }