Merge branch '7393-s3-volume' closes #7393
[arvados.git] / sdk / go / keepclient / keepclient.go
index 53dfb2b5384302b0f113ec608c016760a2ca77ce..26aa7177e05c3aa64d3e956f183379614093d341 100644 (file)
@@ -11,9 +11,7 @@ import (
        "git.curoverse.com/arvados.git/sdk/go/streamer"
        "io"
        "io/ioutil"
-       "log"
        "net/http"
-       "os"
        "regexp"
        "strconv"
        "strings"
@@ -23,7 +21,33 @@ import (
 // A Keep "block" is 64MB.
 const BLOCKSIZE = 64 * 1024 * 1024
 
-var BlockNotFound = errors.New("Block not found")
+// Error interface with an error and boolean indicating whether the error is temporary
+type Error interface {
+       error
+       Temporary() bool
+}
+
+// multipleResponseError is of type Error
+type multipleResponseError struct {
+       error
+       isTemp bool
+}
+
+func (e *multipleResponseError) Temporary() bool {
+       return e.isTemp
+}
+
+// BlockNotFound is a multipleResponseError where isTemp is false
+var BlockNotFound = &ErrNotFound{multipleResponseError{
+       error:  errors.New("Block not found"),
+       isTemp: false,
+}}
+
+// ErrNotFound is a multipleResponseError where isTemp can be true or false
+type ErrNotFound struct {
+       multipleResponseError
+}
+
 var InsufficientReplicasError = errors.New("Could not write sufficient replicas")
 var OversizeBlockError = errors.New("Exceeded maximum block size (" + strconv.Itoa(BLOCKSIZE) + ")")
 var MissingArvadosApiHost = errors.New("Missing required environment variable ARVADOS_API_HOST")
@@ -43,30 +67,46 @@ const X_Keep_Replicas_Stored = "X-Keep-Replicas-Stored"
 type KeepClient struct {
        Arvados            *arvadosclient.ArvadosClient
        Want_replicas      int
-       Using_proxy        bool
        localRoots         *map[string]string
        writableLocalRoots *map[string]string
        gatewayRoots       *map[string]string
        lock               sync.RWMutex
        Client             *http.Client
+       Retries            int
 
        // set to 1 if all writable services are of disk type, otherwise 0
        replicasPerService int
+
+       // Any non-disk typed services found in the list of keepservers?
+       foundNonDiskSvc bool
 }
 
-// Create a new KeepClient.  This will contact the API server to discover Keep
-// servers.
+// MakeKeepClient creates a new KeepClient by contacting the API server to discover Keep servers.
 func MakeKeepClient(arv *arvadosclient.ArvadosClient) (*KeepClient, error) {
-       var matchTrue = regexp.MustCompile("^(?i:1|yes|true)$")
-       insecure := matchTrue.MatchString(os.Getenv("ARVADOS_API_HOST_INSECURE"))
+       kc := New(arv)
+       return kc, kc.DiscoverKeepServers()
+}
+
+// New func creates a new KeepClient struct.
+// This func does not discover keep servers. It is the caller's responsibility.
+func New(arv *arvadosclient.ArvadosClient) *KeepClient {
+       defaultReplicationLevel := 2
+       value, err := arv.Discovery("defaultCollectionReplication")
+       if err == nil {
+               v, ok := value.(float64)
+               if ok && v > 0 {
+                       defaultReplicationLevel = int(v)
+               }
+       }
+
        kc := &KeepClient{
                Arvados:       arv,
-               Want_replicas: 2,
-               Using_proxy:   false,
+               Want_replicas: defaultReplicationLevel,
                Client: &http.Client{Transport: &http.Transport{
-                       TLSClientConfig: &tls.Config{InsecureSkipVerify: insecure}}},
+                       TLSClientConfig: &tls.Config{InsecureSkipVerify: arv.ApiInsecure}}},
+               Retries: 2,
        }
-       return kc, kc.DiscoverKeepServers()
+       return kc
 }
 
 // Put a block given the block hash, a reader, and the number of bytes
@@ -127,6 +167,85 @@ func (kc *KeepClient) PutR(r io.Reader) (locator string, replicas int, err error
        }
 }
 
+func (kc *KeepClient) getOrHead(method string, locator string) (io.ReadCloser, int64, string, error) {
+       var errs []string
+
+       tries_remaining := 1 + kc.Retries
+
+       serversToTry := kc.getSortedRoots(locator)
+
+       numServers := len(serversToTry)
+       count404 := 0
+
+       var retryList []string
+
+       for tries_remaining > 0 {
+               tries_remaining -= 1
+               retryList = nil
+
+               for _, host := range serversToTry {
+                       url := host + "/" + locator
+
+                       req, err := http.NewRequest(method, url, nil)
+                       if err != nil {
+                               errs = append(errs, fmt.Sprintf("%s: %v", url, err))
+                               continue
+                       }
+                       req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", kc.Arvados.ApiToken))
+                       resp, err := kc.Client.Do(req)
+                       if err != nil {
+                               // Probably a network error, may be transient,
+                               // can try again.
+                               errs = append(errs, fmt.Sprintf("%s: %v", url, err))
+                               retryList = append(retryList, host)
+                       } else if resp.StatusCode != http.StatusOK {
+                               var respbody []byte
+                               respbody, _ = ioutil.ReadAll(&io.LimitedReader{R: resp.Body, N: 4096})
+                               resp.Body.Close()
+                               errs = append(errs, fmt.Sprintf("%s: HTTP %d %q",
+                                       url, resp.StatusCode, bytes.TrimSpace(respbody)))
+
+                               if resp.StatusCode == 408 ||
+                                       resp.StatusCode == 429 ||
+                                       resp.StatusCode >= 500 {
+                                       // Timeout, too many requests, or other
+                                       // server side failure, transient
+                                       // error, can try again.
+                                       retryList = append(retryList, host)
+                               } else if resp.StatusCode == 404 {
+                                       count404++
+                               }
+                       } else {
+                               // Success.
+                               if method == "GET" {
+                                       return HashCheckingReader{
+                                               Reader: resp.Body,
+                                               Hash:   md5.New(),
+                                               Check:  locator[0:32],
+                                       }, resp.ContentLength, url, nil
+                               } else {
+                                       resp.Body.Close()
+                                       return nil, resp.ContentLength, url, nil
+                               }
+                       }
+
+               }
+               serversToTry = retryList
+       }
+       DebugPrintf("DEBUG: %s %s failed: %v", method, locator, errs)
+
+       var err error
+       if count404 == numServers {
+               err = BlockNotFound
+       } else {
+               err = &ErrNotFound{multipleResponseError{
+                       error:  fmt.Errorf("%s %s failed: %v", method, locator, errs),
+                       isTemp: len(serversToTry) > 0,
+               }}
+       }
+       return nil, 0, "", err
+}
+
 // Get() retrieves a block, given a locator. Returns a reader, the
 // expected data length, the URL the block is being fetched from, and
 // an error.
@@ -135,36 +254,7 @@ func (kc *KeepClient) PutR(r io.Reader) (locator string, replicas int, err error
 // reader returned by this method will return a BadChecksum error
 // instead of EOF.
 func (kc *KeepClient) Get(locator string) (io.ReadCloser, int64, string, error) {
-       var errs []string
-       for _, host := range kc.getSortedRoots(locator) {
-               url := host + "/" + locator
-               req, err := http.NewRequest("GET", url, nil)
-               if err != nil {
-                       continue
-               }
-               req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", kc.Arvados.ApiToken))
-               resp, err := kc.Client.Do(req)
-               if err != nil || resp.StatusCode != http.StatusOK {
-                       if resp != nil {
-                               var respbody []byte
-                               if resp.Body != nil {
-                                       respbody, _ = ioutil.ReadAll(&io.LimitedReader{resp.Body, 4096})
-                               }
-                               errs = append(errs, fmt.Sprintf("%s: %d %s",
-                                       url, resp.StatusCode, strings.TrimSpace(string(respbody))))
-                       } else {
-                               errs = append(errs, fmt.Sprintf("%s: %v", url, err))
-                       }
-                       continue
-               }
-               return HashCheckingReader{
-                       Reader: resp.Body,
-                       Hash:   md5.New(),
-                       Check:  locator[0:32],
-               }, resp.ContentLength, url, nil
-       }
-       log.Printf("DEBUG: GET %s failed: %v", locator, errs)
-       return nil, 0, "", BlockNotFound
+       return kc.getOrHead("GET", locator)
 }
 
 // Ask() verifies that a block with the given hash is available and
@@ -175,18 +265,8 @@ func (kc *KeepClient) Get(locator string) (io.ReadCloser, int64, string, error)
 // Returns the data size (content length) reported by the Keep service
 // and the URI reporting the data size.
 func (kc *KeepClient) Ask(locator string) (int64, string, error) {
-       for _, host := range kc.getSortedRoots(locator) {
-               url := host + "/" + locator
-               req, err := http.NewRequest("HEAD", url, nil)
-               if err != nil {
-                       continue
-               }
-               req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", kc.Arvados.ApiToken))
-               if resp, err := kc.Client.Do(req); err == nil && resp.StatusCode == http.StatusOK {
-                       return resp.ContentLength, url, nil
-               }
-       }
-       return 0, "", BlockNotFound
+       _, size, url, err := kc.getOrHead("HEAD", locator)
+       return size, url, err
 }
 
 // GetIndex retrieves a list of blocks stored on the given server whose hashes