X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/1a0a58c4f22af82e0a37440af3b0948771bca5e1..dffb8b5eccb3af901512effe562a8b8944b14b89:/sdk/go/keepclient/support.go diff --git a/sdk/go/keepclient/support.go b/sdk/go/keepclient/support.go index ce15ce91ad..6393503e96 100644 --- a/sdk/go/keepclient/support.go +++ b/sdk/go/keepclient/support.go @@ -2,32 +2,81 @@ package keepclient import ( - "git.curoverse.com/arvados.git/sdk/go/streamer" + "crypto/md5" "errors" "fmt" + "git.curoverse.com/arvados.git/sdk/go/streamer" "io" "io/ioutil" "log" + "net" "net/http" - "os" - "strconv" "strings" + "time" ) type keepDisk 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"` } -func (this *KeepClient) DiscoverKeepServers() error { - if prx := os.Getenv("ARVADOS_KEEP_PROXY"); prx != "" { - this.SetServiceRoots([]string{prx}) - this.Using_proxy = true - return nil +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) setClientSettingsStore() { + 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, + } + } +} + +func (this *KeepClient) DiscoverKeepServers() error { type svcList struct { Items []keepDisk `json:"items"` } @@ -42,82 +91,47 @@ func (this *KeepClient) DiscoverKeepServers() error { } listed := make(map[string]bool) - service_roots := make([]string, 0, len(m.Items)) - - for _, element := range m.Items { - n := "" + localRoots := make(map[string]string) + gatewayRoots := make(map[string]string) - if element.SSL { - n = "s" + for _, service := range m.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 = append(service_roots, url) + if listed[url] { + continue } - if element.SvcType == "proxy" { + listed[url] = true + + switch service.SvcType { + case "disk": + localRoots[service.Uuid] = url + case "proxy": + localRoots[service.Uuid] = url this.Using_proxy = true } + // 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 } - this.SetServiceRoots(service_roots) + if this.Using_proxy { + this.setClientSettingsProxy() + } else { + this.setClientSettingsStore() + } + this.SetServiceRoots(localRoots, gatewayRoots) return nil } -func (this KeepClient) shuffledServiceRoots(hash string) (pseq []string) { - // Build an ordering with which to query the Keep servers based on the - // contents of the hash. "hash" is a hex-encoded number at least 8 - // digits (32 bits) long - - // seed used to calculate the next keep server from 'pool' to be added - // to 'pseq' - seed := hash - - // Keep servers still to be added to the ordering - service_roots := this.ServiceRoots() - pool := make([]string, len(service_roots)) - copy(pool, service_roots) - - // output probe sequence - pseq = make([]string, 0, len(service_roots)) - - // iterate while there are servers left to be assigned - for len(pool) > 0 { - - if len(seed) < 8 { - // ran out of digits in the seed - if len(pseq) < (len(hash) / 4) { - // the number of servers added to the probe - // sequence is less than the number of 4-digit - // slices in 'hash' so refill the seed with the - // last 4 digits. - seed = hash[len(hash)-4:] - } - seed += hash - } - - // Take the next 8 digits (32 bytes) and interpret as an integer, - // then modulus with the size of the remaining pool to get the next - // selected server. - probe, _ := strconv.ParseUint(seed[0:8], 16, 32) - probe %= uint64(len(pool)) - - // Append the selected server to the probe sequence and remove it - // from the pool. - pseq = append(pseq, pool[probe]) - pool = append(pool[:probe], pool[probe+1:]...) - - // Remove the digits just used from the seed - seed = seed[8:] - } - return pseq -} - type uploadStatus struct { err error url string @@ -127,21 +141,29 @@ type uploadStatus struct { } func (this KeepClient) uploadToKeepServer(host string, hash string, body io.ReadCloser, - upload_status chan<- uploadStatus, expectedLength int64) { - - log.Printf("Uploading %s to %s", hash, host) + upload_status chan<- uploadStatus, expectedLength int64, requestId 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()) upload_status <- uploadStatus{err, url, 0, 0, ""} body.Close() return } + req.ContentLength = expectedLength if expectedLength > 0 { - req.ContentLength = expectedLength + // http.Client.Do will close the body ReadCloser when it is + // done with it. + req.Body = 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() } req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", this.Arvados.ApiToken)) @@ -151,12 +173,10 @@ func (this KeepClient) uploadToKeepServer(host string, hash string, body io.Read req.Header.Add(X_Keep_Desired_Replicas, fmt.Sprint(this.Want_replicas)) } - req.Body = body - 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()) upload_status <- uploadStatus{err, url, 0, 0, ""} - body.Close() return } @@ -169,17 +189,16 @@ func (this KeepClient) uploadToKeepServer(host string, hash string, body io.Read defer io.Copy(ioutil.Discard, resp.Body) respbody, err2 := ioutil.ReadAll(&io.LimitedReader{resp.Body, 4096}) + response := strings.TrimSpace(string(respbody)) if err2 != nil && err2 != io.EOF { - upload_status <- uploadStatus{err2, url, resp.StatusCode, rep, string(respbody)} - return - } - - locator := strings.TrimSpace(string(respbody)) - - if resp.StatusCode == http.StatusOK { - upload_status <- uploadStatus{nil, url, resp.StatusCode, rep, locator} + log.Printf("[%v] 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) + upload_status <- uploadStatus{nil, url, resp.StatusCode, rep, response} } else { - upload_status <- uploadStatus{errors.New(resp.Status), url, resp.StatusCode, rep, locator} + log.Printf("[%v] Upload %v error: %v response: %v", requestId, url, resp.StatusCode, response) + upload_status <- uploadStatus{errors.New(resp.Status), url, resp.StatusCode, rep, response} } } @@ -188,8 +207,12 @@ func (this KeepClient) putReplicas( 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] + // Calculate the ordering for uploading to servers - sv := this.shuffledServiceRoots(hash) + sv := NewRootSorter(this.LocalRoots(), hash).GetSortedRoots() // The next server to try contacting next_server := 0 @@ -202,14 +225,14 @@ func (this KeepClient) putReplicas( defer 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) { - go this.uploadToKeepServer(sv[next_server], hash, tr.MakeStreamReader(), upload_status, expectedLength) + 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 { @@ -220,20 +243,18 @@ func (this KeepClient) putReplicas( } } } + 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 - } else { - // writing to keep server failed for some reason - log.Printf("Keep server put to %v failed with '%v'", - status.url, status.err) } - active -= 1 - log.Printf("Upload to %v status code: %v remaining replicas: %v active: %v", status.url, status.statusCode, remaining_replicas, active) } return locator, this.Want_replicas, nil