X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/1a0a58c4f22af82e0a37440af3b0948771bca5e1..224f384d411bb1b4cccc7165c55bb64fd5c695ad:/sdk/go/keepclient/support.go diff --git a/sdk/go/keepclient/support.go b/sdk/go/keepclient/support.go index ce15ce91ad..22447794f1 100644 --- a/sdk/go/keepclient/support.go +++ b/sdk/go/keepclient/support.go @@ -1,121 +1,88 @@ -/* Internal methods to support keepclient.go */ 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" + "math/rand" + "net" "net/http" - "os" - "strconv" "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"` SSL bool `json:"service_ssl_flag"` SvcType string `json:"service_type"` + ReadOnly bool `json:"read_only"` } -func (this *KeepClient) DiscoverKeepServers() error { - if prx := os.Getenv("ARVADOS_KEEP_PROXY"); prx != "" { - this.SetServiceRoots([]string{prx}) - this.Using_proxy = true - return nil - } - - type svcList struct { - Items []keepDisk `json:"items"` - } - var m svcList - - err := this.Arvados.Call("GET", "keep_services", "", "accessible", nil, &m) +// Md5String returns md5 hash for the bytes in the given string +func Md5String(s string) string { + return fmt.Sprintf("%x", md5.Sum([]byte(s))) +} - if err != nil { - if err := this.Arvados.List("keep_disks", nil, &m); err != nil { - return err +// Set timeouts applicable when connecting to non-disk services +// (assumed to be over the Internet). +func (this *KeepClient) setClientSettingsNonDisk() { + 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, } } +} - listed := make(map[string]bool) - service_roots := make([]string, 0, len(m.Items)) - - for _, element := range m.Items { - n := "" - - if element.SSL { - n = "s" - } - - // Construct server URL - url := fmt.Sprintf("http%s://%s:%d", n, element.Hostname, element.Port) - - // Skip duplicates - if !listed[url] { - listed[url] = true - service_roots = append(service_roots, url) - } - if element.SvcType == "proxy" { - this.Using_proxy = true +// Set timeouts applicable 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, } } - - this.SetServiceRoots(service_roots) - - 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 svcList struct { + Items []keepService `json:"items"` } type uploadStatus struct { @@ -126,37 +93,40 @@ type uploadStatus struct { response string } -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) +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 { + DebugPrintf("DEBUG: [%08x] 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)) 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.Body = body + 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 { + DebugPrintf("DEBUG: [%08x] Upload failed %v error: %v", requestID, url, err.Error()) upload_status <- uploadStatus{err, url, 0, 0, ""} - body.Close() return } @@ -168,28 +138,31 @@ 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 { - 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} + DebugPrintf("DEBUG: [%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 { + DebugPrintf("DEBUG: [%08x] 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} + DebugPrintf("DEBUG: [%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) { + // Generate an arbitrary ID to identify this specific + // transaction in debug logs. + requestID := rand.Int31() + // Calculate the ordering for uploading to servers - sv := this.shuffledServiceRoots(hash) + sv := NewRootSorter(this.WritableLocalRoots(), hash).GetSortedRoots() // The next server to try contacting next_server := 0 @@ -199,42 +172,76 @@ 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 + replicasDone := 0 + replicasTodo := this.Want_replicas - remaining_replicas := this.Want_replicas + replicasPerThread := this.replicasPerService + if replicasPerThread < 1 { + // unlimited or unknown + replicasPerThread = replicasTodo + } - 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) - next_server += 1 - active += 1 - } else { - if active == 0 { - return locator, (this.Want_replicas - remaining_replicas), InsufficientReplicasError + retriesRemaining := 1 + this.Retries + var retryServers []string + + for retriesRemaining > 0 { + retriesRemaining -= 1 + next_server = 0 + retryServers = []string{} + for replicasTodo > 0 { + for active*replicasPerThread < replicasTodo { + // Start some upload requests + if next_server < len(sv) { + DebugPrintf("DEBUG: [%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, replicasDone, InsufficientReplicasError + } else { + break + } } } + DebugPrintf("DEBUG: [%08x] Replicas remaining to write: %v active uploads: %v", + requestID, replicasTodo, active) + + // Now wait for something to happen. + if active > 0 { + status := <-upload_status + active -= 1 + + if status.statusCode == 200 { + // good news! + replicasDone += status.replicas_stored + replicasTodo -= 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 + } } - // Now wait for something to happen. - status := <-upload_status - 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) + sv = retryServers } - return locator, this.Want_replicas, nil + return locator, replicasDone, nil }