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"
22 // A Keep "block" is 64MB.
23 const BLOCKSIZE = 64 * 1024 * 1024
25 var BlockNotFound = errors.New("Block not found")
26 var InsufficientReplicasError = errors.New("Could not write sufficient replicas")
27 var OversizeBlockError = errors.New("Exceeded maximum block size (" + strconv.Itoa(BLOCKSIZE) + ")")
28 var MissingArvadosApiHost = errors.New("Missing required environment variable ARVADOS_API_HOST")
29 var MissingArvadosApiToken = errors.New("Missing required environment variable ARVADOS_API_TOKEN")
30 var InvalidLocatorError = errors.New("Invalid locator")
32 // ErrorGetIndex is the generic GetIndex error
33 var ErrorGetIndex = errors.New("Error getting index")
35 // ErrorNoSuchKeepServer is returned when GetIndex is invoked with a UUID with no matching keep server
36 var ErrorNoSuchKeepServer = errors.New("No keep server matching the given UUID is found")
38 // ErrorIncompleteIndex is returned when the Index response does not end with a new empty line
39 var ErrorIncompleteIndex = errors.New("Got incomplete index")
41 const X_Keep_Desired_Replicas = "X-Keep-Desired-Replicas"
42 const X_Keep_Replicas_Stored = "X-Keep-Replicas-Stored"
44 // Information about Arvados and Keep servers.
45 type KeepClient struct {
46 Arvados *arvadosclient.ArvadosClient
49 localRoots *map[string]string
50 writableLocalRoots *map[string]string
51 gatewayRoots *map[string]string
55 // set to 1 if all writable services are of disk type, otherwise 0
56 replicasPerService int
59 // Create a new KeepClient. This will contact the API server to discover Keep
61 func MakeKeepClient(arv *arvadosclient.ArvadosClient) (*KeepClient, error) {
62 var matchTrue = regexp.MustCompile("^(?i:1|yes|true)$")
63 insecure := matchTrue.MatchString(os.Getenv("ARVADOS_API_HOST_INSECURE"))
68 Client: &http.Client{Transport: &http.Transport{
69 TLSClientConfig: &tls.Config{InsecureSkipVerify: insecure}}},
71 return kc, kc.DiscoverKeepServers()
74 // Put a block given the block hash, a reader, and the number of bytes
75 // to read from the reader (which must be between 0 and BLOCKSIZE).
77 // Returns the locator for the written block, the number of replicas
78 // written, and an error.
80 // Returns an InsufficientReplicas error if 0 <= replicas <
82 func (kc *KeepClient) PutHR(hash string, r io.Reader, dataBytes int64) (string, int, error) {
83 // Buffer for reads from 'r'
86 if dataBytes > BLOCKSIZE {
87 return "", 0, OversizeBlockError
89 bufsize = int(dataBytes)
94 t := streamer.AsyncStreamFromReader(bufsize, HashCheckingReader{r, md5.New(), hash})
97 return kc.putReplicas(hash, t, dataBytes)
100 // PutHB writes a block to Keep. The hash of the bytes is given in
101 // hash, and the data is given in buf.
103 // Return values are the same as for PutHR.
104 func (kc *KeepClient) PutHB(hash string, buf []byte) (string, int, error) {
105 t := streamer.AsyncStreamFromSlice(buf)
107 return kc.putReplicas(hash, t, int64(len(buf)))
110 // PutB writes a block to Keep. It computes the hash itself.
112 // Return values are the same as for PutHR.
113 func (kc *KeepClient) PutB(buffer []byte) (string, int, error) {
114 hash := fmt.Sprintf("%x", md5.Sum(buffer))
115 return kc.PutHB(hash, buffer)
118 // PutR writes a block to Keep. It first reads all data from r into a buffer
119 // in order to compute the hash.
121 // Return values are the same as for PutHR.
123 // If the block hash and data size are known, PutHR is more efficient.
124 func (kc *KeepClient) PutR(r io.Reader) (locator string, replicas int, err error) {
125 if buffer, err := ioutil.ReadAll(r); err != nil {
128 return kc.PutB(buffer)
132 // Get() retrieves a block, given a locator. Returns a reader, the
133 // expected data length, the URL the block is being fetched from, and
136 // If the block checksum does not match, the final Read() on the
137 // reader returned by this method will return a BadChecksum error
139 func (kc *KeepClient) Get(locator string) (io.ReadCloser, int64, string, error) {
141 for _, host := range kc.getSortedRoots(locator) {
142 url := host + "/" + locator
143 req, err := http.NewRequest("GET", url, nil)
147 req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", kc.Arvados.ApiToken))
148 resp, err := kc.Client.Do(req)
149 if err != nil || resp.StatusCode != http.StatusOK {
152 if resp.Body != nil {
153 respbody, _ = ioutil.ReadAll(&io.LimitedReader{resp.Body, 4096})
155 errs = append(errs, fmt.Sprintf("%s: %d %s",
156 url, resp.StatusCode, strings.TrimSpace(string(respbody))))
158 errs = append(errs, fmt.Sprintf("%s: %v", url, err))
162 return HashCheckingReader{
165 Check: locator[0:32],
166 }, resp.ContentLength, url, nil
168 log.Printf("DEBUG: GET %s failed: %v", locator, errs)
169 return nil, 0, "", BlockNotFound
172 // Ask() verifies that a block with the given hash is available and
173 // readable, according to at least one Keep service. Unlike Get, it
174 // does not retrieve the data or verify that the data content matches
175 // the hash specified by the locator.
177 // Returns the data size (content length) reported by the Keep service
178 // and the URI reporting the data size.
179 func (kc *KeepClient) Ask(locator string) (int64, string, error) {
180 for _, host := range kc.getSortedRoots(locator) {
181 url := host + "/" + locator
182 req, err := http.NewRequest("HEAD", url, nil)
186 req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", kc.Arvados.ApiToken))
187 if resp, err := kc.Client.Do(req); err == nil && resp.StatusCode == http.StatusOK {
188 return resp.ContentLength, url, nil
191 return 0, "", BlockNotFound
194 // GetIndex retrieves a list of blocks stored on the given server whose hashes
195 // begin with the given prefix. The returned reader will return an error (other
196 // than EOF) if the complete index cannot be retrieved. This should only be
197 // expected to return useful results if the client is using a "data manager token"
198 // recognized by the Keep services.
199 func (kc *KeepClient) GetIndex(keepServiceUUID, prefix string) (io.Reader, error) {
200 url := kc.LocalRoots()[keepServiceUUID]
202 log.Printf("No such keep server found: %v", keepServiceUUID)
203 return nil, ErrorNoSuchKeepServer
211 req, err := http.NewRequest("GET", url, nil)
213 log.Printf("GET index error: %v", err)
214 return nil, ErrorGetIndex
217 req.Header.Add("Authorization", fmt.Sprintf("OAuth2 %s", kc.Arvados.ApiToken))
218 resp, err := kc.Client.Do(req)
219 if err != nil || resp.StatusCode != http.StatusOK {
220 log.Printf("GET index error: %v; status code: %v", err, resp.StatusCode)
221 return nil, ErrorGetIndex
225 if resp.Body != nil {
226 respBody, err = ioutil.ReadAll(resp.Body)
228 log.Printf("GET index error: %v", err)
229 return nil, ErrorGetIndex
232 // Got index; verify that it is complete
233 // The response should be "\n" if no locators matched the prefix
234 // Else, it should be a list of locators followed by a blank line
235 if (!strings.HasSuffix(string(respBody), "\n\n")) && (string(respBody) != "\n") {
236 log.Printf("Got incomplete index for %v", url)
237 return nil, ErrorIncompleteIndex
241 // Got complete index
242 return strings.NewReader(string(respBody)), nil
245 // LocalRoots() returns the map of local (i.e., disk and proxy) Keep
246 // services: uuid -> baseURI.
247 func (kc *KeepClient) LocalRoots() map[string]string {
249 defer kc.lock.RUnlock()
250 return *kc.localRoots
253 // GatewayRoots() returns the map of Keep remote gateway services:
255 func (kc *KeepClient) GatewayRoots() map[string]string {
257 defer kc.lock.RUnlock()
258 return *kc.gatewayRoots
261 // WritableLocalRoots() returns the map of writable local Keep services:
263 func (kc *KeepClient) WritableLocalRoots() map[string]string {
265 defer kc.lock.RUnlock()
266 return *kc.writableLocalRoots
269 // SetServiceRoots updates the localRoots and gatewayRoots maps,
270 // without risk of disrupting operations that are already in progress.
272 // The KeepClient makes its own copy of the supplied maps, so the
273 // caller can reuse/modify them after SetServiceRoots returns, but
274 // they should not be modified by any other goroutine while
275 // SetServiceRoots is running.
276 func (kc *KeepClient) SetServiceRoots(newLocals, newWritableLocals map[string]string, newGateways map[string]string) {
277 locals := make(map[string]string)
278 for uuid, root := range newLocals {
282 writables := make(map[string]string)
283 for uuid, root := range newWritableLocals {
284 writables[uuid] = root
287 gateways := make(map[string]string)
288 for uuid, root := range newGateways {
289 gateways[uuid] = root
293 defer kc.lock.Unlock()
294 kc.localRoots = &locals
295 kc.writableLocalRoots = &writables
296 kc.gatewayRoots = &gateways
299 // getSortedRoots returns a list of base URIs of Keep services, in the
300 // order they should be attempted in order to retrieve content for the
302 func (kc *KeepClient) getSortedRoots(locator string) []string {
304 for _, hint := range strings.Split(locator, "+") {
305 if len(hint) < 7 || hint[0:2] != "K@" {
306 // Not a service hint.
310 // +K@abcde means fetch from proxy at
311 // keep.abcde.arvadosapi.com
312 found = append(found, "https://keep."+hint[2:]+".arvadosapi.com")
313 } else if len(hint) == 29 {
314 // +K@abcde-abcde-abcdeabcdeabcde means fetch
315 // from gateway with given uuid
316 if gwURI, ok := kc.GatewayRoots()[hint[2:]]; ok {
317 found = append(found, gwURI)
319 // else this hint is no use to us; carry on.
322 // After trying all usable service hints, fall back to local roots.
323 found = append(found, NewRootSorter(kc.LocalRoots(), locator[0:32]).GetSortedRoots()...)
327 type Locator struct {
329 Size int // -1 if data size is not known
330 Hints []string // Including the size hint, if any
333 func (loc *Locator) String() string {
335 if len(loc.Hints) > 0 {
336 s = s + "+" + strings.Join(loc.Hints, "+")
341 var locatorMatcher = regexp.MustCompile("^([0-9a-f]{32})([+](.*))?$")
343 func MakeLocator(path string) (*Locator, error) {
344 sm := locatorMatcher.FindStringSubmatch(path)
346 return nil, InvalidLocatorError
348 loc := Locator{Hash: sm[1], Size: -1}
350 loc.Hints = strings.Split(sm[3], "+")
352 loc.Hints = []string{}
354 if len(loc.Hints) > 0 {
355 if size, err := strconv.Atoi(loc.Hints[0]); err == nil {