1 /* Provides low-level Get/Put primitives for accessing Arvados Keep blocks. */
9 "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
10 "git.curoverse.com/arvados.git/sdk/go/streamer"
20 // A Keep "block" is 64MB.
21 const BLOCKSIZE = 64 * 1024 * 1024
23 // Error interface with an error and boolean indicating whether the error is temporary
24 type Error interface {
29 // multipleResponseError is of type Error
30 type multipleResponseError struct {
35 func (e *multipleResponseError) Temporary() bool {
39 // BlockNotFound is a multipleResponseError where isTemp is false
40 var BlockNotFound = &ErrNotFound{multipleResponseError{
41 error: errors.New("Block not found"),
45 // ErrNotFound is a multipleResponseError where isTemp can be true or false
46 type ErrNotFound struct {
50 var InsufficientReplicasError = errors.New("Could not write sufficient replicas")
51 var OversizeBlockError = errors.New("Exceeded maximum block size (" + strconv.Itoa(BLOCKSIZE) + ")")
52 var MissingArvadosApiHost = errors.New("Missing required environment variable ARVADOS_API_HOST")
53 var MissingArvadosApiToken = errors.New("Missing required environment variable ARVADOS_API_TOKEN")
54 var InvalidLocatorError = errors.New("Invalid locator")
56 // ErrNoSuchKeepServer is returned when GetIndex is invoked with a UUID with no matching keep server
57 var ErrNoSuchKeepServer = errors.New("No keep server matching the given UUID is found")
59 // ErrIncompleteIndex is returned when the Index response does not end with a new empty line
60 var ErrIncompleteIndex = errors.New("Got incomplete index")
62 const X_Keep_Desired_Replicas = "X-Keep-Desired-Replicas"
63 const X_Keep_Replicas_Stored = "X-Keep-Replicas-Stored"
65 // Information about Arvados and Keep servers.
66 type KeepClient struct {
67 Arvados *arvadosclient.ArvadosClient
69 localRoots *map[string]string
70 writableLocalRoots *map[string]string
71 gatewayRoots *map[string]string
76 // set to 1 if all writable services are of disk type, otherwise 0
77 replicasPerService int
79 // Any non-disk typed services found in the list of keepservers?
83 // MakeKeepClient creates a new KeepClient by contacting the API server to discover Keep servers.
84 func MakeKeepClient(arv *arvadosclient.ArvadosClient) (*KeepClient, error) {
86 return kc, kc.DiscoverKeepServers()
89 // New func creates a new KeepClient struct.
90 // This func does not discover keep servers. It is the caller's responsibility.
91 func New(arv *arvadosclient.ArvadosClient) *KeepClient {
92 defaultReplicationLevel := 2
93 value, err := arv.Discovery("defaultCollectionReplication")
95 v, ok := value.(float64)
97 defaultReplicationLevel = int(v)
103 Want_replicas: defaultReplicationLevel,
104 Client: &http.Client{Transport: &http.Transport{
105 TLSClientConfig: arvadosclient.MakeTLSConfig(arv.ApiInsecure)}},
111 // Put a block given the block hash, a reader, and the number of bytes
112 // to read from the reader (which must be between 0 and BLOCKSIZE).
114 // Returns the locator for the written block, the number of replicas
115 // written, and an error.
117 // Returns an InsufficientReplicas error if 0 <= replicas <
118 // kc.Wants_replicas.
119 func (kc *KeepClient) PutHR(hash string, r io.Reader, dataBytes int64) (string, int, error) {
120 // Buffer for reads from 'r'
123 if dataBytes > BLOCKSIZE {
124 return "", 0, OversizeBlockError
126 bufsize = int(dataBytes)
131 t := streamer.AsyncStreamFromReader(bufsize, HashCheckingReader{r, md5.New(), hash})
134 return kc.putReplicas(hash, t, dataBytes)
137 // PutHB writes a block to Keep. The hash of the bytes is given in
138 // hash, and the data is given in buf.
140 // Return values are the same as for PutHR.
141 func (kc *KeepClient) PutHB(hash string, buf []byte) (string, int, error) {
142 t := streamer.AsyncStreamFromSlice(buf)
144 return kc.putReplicas(hash, t, int64(len(buf)))
147 // PutB writes a block to Keep. It computes the hash itself.
149 // Return values are the same as for PutHR.
150 func (kc *KeepClient) PutB(buffer []byte) (string, int, error) {
151 hash := fmt.Sprintf("%x", md5.Sum(buffer))
152 return kc.PutHB(hash, buffer)
155 // PutR writes a block to Keep. It first reads all data from r into a buffer
156 // in order to compute the hash.
158 // Return values are the same as for PutHR.
160 // If the block hash and data size are known, PutHR is more efficient.
161 func (kc *KeepClient) PutR(r io.Reader) (locator string, replicas int, err error) {
162 if buffer, err := ioutil.ReadAll(r); err != nil {
165 return kc.PutB(buffer)
169 func (kc *KeepClient) getOrHead(method string, locator string) (io.ReadCloser, int64, string, error) {
170 if strings.HasPrefix(locator, "d41d8cd98f00b204e9800998ecf8427e+0") {
171 return ioutil.NopCloser(bytes.NewReader(nil)), 0, "", nil
176 tries_remaining := 1 + kc.Retries
178 serversToTry := kc.getSortedRoots(locator)
180 numServers := len(serversToTry)
183 var retryList []string
185 for tries_remaining > 0 {
189 for _, host := range serversToTry {
190 url := host + "/" + locator
192 req, err := http.NewRequest(method, url, nil)
194 errs = append(errs, fmt.Sprintf("%s: %v", url, err))
197 req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", kc.Arvados.ApiToken))
198 resp, err := kc.Client.Do(req)
200 // Probably a network error, may be transient,
202 errs = append(errs, fmt.Sprintf("%s: %v", url, err))
203 retryList = append(retryList, host)
204 } else if resp.StatusCode != http.StatusOK {
206 respbody, _ = ioutil.ReadAll(&io.LimitedReader{R: resp.Body, N: 4096})
208 errs = append(errs, fmt.Sprintf("%s: HTTP %d %q",
209 url, resp.StatusCode, bytes.TrimSpace(respbody)))
211 if resp.StatusCode == 408 ||
212 resp.StatusCode == 429 ||
213 resp.StatusCode >= 500 {
214 // Timeout, too many requests, or other
215 // server side failure, transient
216 // error, can try again.
217 retryList = append(retryList, host)
218 } else if resp.StatusCode == 404 {
224 return HashCheckingReader{
227 Check: locator[0:32],
228 }, resp.ContentLength, url, nil
231 return nil, resp.ContentLength, url, nil
236 serversToTry = retryList
238 DebugPrintf("DEBUG: %s %s failed: %v", method, locator, errs)
241 if count404 == numServers {
244 err = &ErrNotFound{multipleResponseError{
245 error: fmt.Errorf("%s %s failed: %v", method, locator, errs),
246 isTemp: len(serversToTry) > 0,
249 return nil, 0, "", err
252 // Get() retrieves a block, given a locator. Returns a reader, the
253 // expected data length, the URL the block is being fetched from, and
256 // If the block checksum does not match, the final Read() on the
257 // reader returned by this method will return a BadChecksum error
259 func (kc *KeepClient) Get(locator string) (io.ReadCloser, int64, string, error) {
260 return kc.getOrHead("GET", locator)
263 // Ask() verifies that a block with the given hash is available and
264 // readable, according to at least one Keep service. Unlike Get, it
265 // does not retrieve the data or verify that the data content matches
266 // the hash specified by the locator.
268 // Returns the data size (content length) reported by the Keep service
269 // and the URI reporting the data size.
270 func (kc *KeepClient) Ask(locator string) (int64, string, error) {
271 _, size, url, err := kc.getOrHead("HEAD", locator)
272 return size, url, err
275 // GetIndex retrieves a list of blocks stored on the given server whose hashes
276 // begin with the given prefix. The returned reader will return an error (other
277 // than EOF) if the complete index cannot be retrieved.
279 // This is meant to be used only by system components and admin tools.
280 // It will return an error unless the client is using a "data manager token"
281 // recognized by the Keep services.
282 func (kc *KeepClient) GetIndex(keepServiceUUID, prefix string) (io.Reader, error) {
283 url := kc.LocalRoots()[keepServiceUUID]
285 return nil, ErrNoSuchKeepServer
293 req, err := http.NewRequest("GET", url, nil)
298 req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", kc.Arvados.ApiToken))
299 resp, err := kc.Client.Do(req)
304 defer resp.Body.Close()
306 if resp.StatusCode != http.StatusOK {
307 return nil, fmt.Errorf("Got http status code: %d", resp.StatusCode)
311 respBody, err = ioutil.ReadAll(resp.Body)
316 // Got index; verify that it is complete
317 // The response should be "\n" if no locators matched the prefix
318 // Else, it should be a list of locators followed by a blank line
319 if !bytes.Equal(respBody, []byte("\n")) && !bytes.HasSuffix(respBody, []byte("\n\n")) {
320 return nil, ErrIncompleteIndex
323 // Got complete index; strip the trailing newline and send
324 return bytes.NewReader(respBody[0 : len(respBody)-1]), nil
327 // LocalRoots() returns the map of local (i.e., disk and proxy) Keep
328 // services: uuid -> baseURI.
329 func (kc *KeepClient) LocalRoots() map[string]string {
331 defer kc.lock.RUnlock()
332 return *kc.localRoots
335 // GatewayRoots() returns the map of Keep remote gateway services:
337 func (kc *KeepClient) GatewayRoots() map[string]string {
339 defer kc.lock.RUnlock()
340 return *kc.gatewayRoots
343 // WritableLocalRoots() returns the map of writable local Keep services:
345 func (kc *KeepClient) WritableLocalRoots() map[string]string {
347 defer kc.lock.RUnlock()
348 return *kc.writableLocalRoots
351 // SetServiceRoots updates the localRoots and gatewayRoots maps,
352 // without risk of disrupting operations that are already in progress.
354 // The KeepClient makes its own copy of the supplied maps, so the
355 // caller can reuse/modify them after SetServiceRoots returns, but
356 // they should not be modified by any other goroutine while
357 // SetServiceRoots is running.
358 func (kc *KeepClient) SetServiceRoots(newLocals, newWritableLocals, newGateways map[string]string) {
359 locals := make(map[string]string)
360 for uuid, root := range newLocals {
364 writables := make(map[string]string)
365 for uuid, root := range newWritableLocals {
366 writables[uuid] = root
369 gateways := make(map[string]string)
370 for uuid, root := range newGateways {
371 gateways[uuid] = root
375 defer kc.lock.Unlock()
376 kc.localRoots = &locals
377 kc.writableLocalRoots = &writables
378 kc.gatewayRoots = &gateways
381 // getSortedRoots returns a list of base URIs of Keep services, in the
382 // order they should be attempted in order to retrieve content for the
384 func (kc *KeepClient) getSortedRoots(locator string) []string {
386 for _, hint := range strings.Split(locator, "+") {
387 if len(hint) < 7 || hint[0:2] != "K@" {
388 // Not a service hint.
392 // +K@abcde means fetch from proxy at
393 // keep.abcde.arvadosapi.com
394 found = append(found, "https://keep."+hint[2:]+".arvadosapi.com")
395 } else if len(hint) == 29 {
396 // +K@abcde-abcde-abcdeabcdeabcde means fetch
397 // from gateway with given uuid
398 if gwURI, ok := kc.GatewayRoots()[hint[2:]]; ok {
399 found = append(found, gwURI)
401 // else this hint is no use to us; carry on.
404 // After trying all usable service hints, fall back to local roots.
405 found = append(found, NewRootSorter(kc.LocalRoots(), locator[0:32]).GetSortedRoots()...)
409 type Locator struct {
411 Size int // -1 if data size is not known
412 Hints []string // Including the size hint, if any
415 func (loc *Locator) String() string {
417 if len(loc.Hints) > 0 {
418 s = s + "+" + strings.Join(loc.Hints, "+")
423 var locatorMatcher = regexp.MustCompile("^([0-9a-f]{32})([+](.*))?$")
425 func MakeLocator(path string) (*Locator, error) {
426 sm := locatorMatcher.FindStringSubmatch(path)
428 return nil, InvalidLocatorError
430 loc := Locator{Hash: sm[1], Size: -1}
432 loc.Hints = strings.Split(sm[3], "+")
434 loc.Hints = []string{}
436 if len(loc.Hints) > 0 {
437 if size, err := strconv.Atoi(loc.Hints[0]); err == nil {