1 # Copyright (C) The Arvados Authors. All rights reserved.
3 # SPDX-License-Identifier: Apache-2.0
6 The code in this module builds Arvados API client objects you can use to submit
7 Arvados API requests. This includes extending the underlying HTTP client with
8 niceties such as caching, X-Request-Id header for tracking, and more. The main
9 client constructors are `api` and `api_from_config`.
12 from __future__ import absolute_import
13 from future import standard_library
14 standard_library.install_aliases()
15 from builtins import range
33 from apiclient import discovery as apiclient_discovery
34 from apiclient import errors as apiclient_errors
40 from .logging import GoogleHTTPClientFilter, log_handler
42 _logger = logging.getLogger('arvados.api')
43 _googleapiclient_log_lock = threading.Lock()
45 MAX_IDLE_CONNECTION_DURATION = 30
47 Number of seconds that API client HTTP connections should be allowed to idle
48 in keepalive state before they are forced closed. Client code can adjust this
49 constant, and it will be used for all Arvados API clients constructed after
53 # An unused HTTP 5xx status code to request a retry internally.
54 # See _intercept_http_request. This should not be user-visible.
55 _RETRY_4XX_STATUS = 545
57 if sys.version_info >= (3,):
58 httplib2.SSLHandshakeError = None
60 _orig_retry_request = apiclient.http._retry_request
61 def _retry_request(http, num_retries, *args, **kwargs):
63 num_retries = max(num_retries, http.num_retries)
64 except AttributeError:
65 # `http` client object does not have a `num_retries` attribute.
66 # It apparently hasn't gone through _patch_http_request, possibly
67 # because this isn't an Arvados API client. Pass through to
68 # avoid interfering with other Google API clients.
69 return _orig_retry_request(http, num_retries, *args, **kwargs)
70 response, body = _orig_retry_request(http, num_retries, *args, **kwargs)
71 # If _intercept_http_request ran out of retries for a 4xx response,
72 # restore the original status code.
73 if response.status == _RETRY_4XX_STATUS:
74 response.status = int(response['status'])
75 return (response, body)
76 apiclient.http._retry_request = _retry_request
78 def _intercept_http_request(self, uri, method="GET", headers={}, **kwargs):
79 if not headers.get('X-Request-Id'):
80 headers['X-Request-Id'] = self._request_id()
82 if (self.max_request_size and
83 kwargs.get('body') and
84 self.max_request_size < len(kwargs['body'])):
85 raise apiclient_errors.MediaUploadSizeError("Request size %i bytes exceeds published limit of %i bytes" % (len(kwargs['body']), self.max_request_size))
87 headers['Authorization'] = 'OAuth2 %s' % self.arvados_api_token
89 if (time.time() - self._last_request_time) > self._max_keepalive_idle:
90 # High probability of failure due to connection atrophy. Make
91 # sure this request [re]opens a new connection by closing and
92 # forgetting all cached connections first.
93 for conn in self.connections.values():
95 self.connections.clear()
97 self._last_request_time = time.time()
99 response, body = self.orig_http_request(uri, method, headers=headers, **kwargs)
100 except ssl.SSLCertVerificationError as e:
101 raise ssl.SSLCertVerificationError(e.args[0], "Could not connect to %s\n%s\nPossible causes: remote SSL/TLS certificate expired, or was issued by an untrusted certificate authority." % (uri, e)) from None
102 # googleapiclient only retries 403, 429, and 5xx status codes.
103 # If we got another 4xx status that we want to retry, convert it into
104 # 5xx so googleapiclient handles it the way we want.
105 if response.status in retry._HTTP_CAN_RETRY and response.status < 500:
106 response.status = _RETRY_4XX_STATUS
107 return (response, body)
108 except Exception as e:
109 # Prepend "[request_id] " to the error message, which we
110 # assume is the first string argument passed to the exception
112 for i in range(len(e.args or ())):
113 if type(e.args[i]) == type(""):
114 e.args = e.args[:i] + ("[{}] {}".format(headers['X-Request-Id'], e.args[i]),) + e.args[i+1:]
115 raise type(e)(*e.args)
118 def _patch_http_request(http, api_token, num_retries):
119 http.arvados_api_token = api_token
120 http.max_request_size = 0
121 http.num_retries = num_retries
122 http.orig_http_request = http.request
123 http.request = types.MethodType(_intercept_http_request, http)
124 http._last_request_time = 0
125 http._max_keepalive_idle = MAX_IDLE_CONNECTION_DURATION
126 http._request_id = util.new_request_id
129 def _close_connections(self):
130 for conn in self._http.connections.values():
133 # Monkey patch discovery._cast() so objects and arrays get serialized
134 # with json.dumps() instead of str().
135 _cast_orig = apiclient_discovery._cast
136 def _cast_objects_too(value, schema_type):
138 if (type(value) != type('') and
139 type(value) != type(b'') and
140 (schema_type == 'object' or schema_type == 'array')):
141 return json.dumps(value)
143 return _cast_orig(value, schema_type)
144 apiclient_discovery._cast = _cast_objects_too
146 # Convert apiclient's HttpErrors into our own API error subclass for better
148 # Reassigning apiclient_errors.HttpError is not sufficient because most of the
149 # apiclient submodules import the class into their own namespace.
150 def _new_http_error(cls, *args, **kwargs):
151 return super(apiclient_errors.HttpError, cls).__new__(
152 errors.ApiError, *args, **kwargs)
153 apiclient_errors.HttpError.__new__ = staticmethod(_new_http_error)
155 def http_cache(data_type):
156 """Set up an HTTP file cache
158 This function constructs and returns an `arvados.cache.SafeHTTPCache`
159 backed by the filesystem under `~/.cache/arvados/`, or `None` if the
160 directory cannot be set up. The return value can be passed to
161 `httplib2.Http` as the `cache` argument.
165 * data_type: str --- The name of the subdirectory under `~/.cache/arvados`
166 where data is cached.
169 homedir = pathlib.Path.home()
172 path = pathlib.Path(homedir, '.cache', 'arvados', data_type)
174 path.mkdir(parents=True, exist_ok=True)
177 return cache.SafeHTTPCache(str(path), max_age=60*60*24*2)
192 """Build an Arvados API client
194 This function returns a `googleapiclient.discovery.Resource` object
195 constructed from the given arguments. This is a relatively low-level
196 interface that requires all the necessary inputs as arguments. Most
197 users will prefer to use `api` which can accept more flexible inputs.
202 : A string naming the version of the Arvados API to use.
204 discoveryServiceUrl: str
205 : The URL used to discover APIs passed directly to
206 `googleapiclient.discovery.build`.
209 : The authentication token to send with each API call.
211 Keyword-only arguments:
214 : If true, loads the API discovery document from, or saves it to, a cache
215 on disk (located at `~/.cache/arvados/discovery`).
217 http: httplib2.Http | None
218 : The HTTP client object the API client object will use to make requests.
219 If not provided, this function will build its own to use. Either way, the
220 object will be patched as part of the build process.
223 : If true, ignore SSL certificate validation errors. Default `False`.
226 : The number of times to retry each API request if it encounters a
227 temporary failure. Default 10.
229 request_id: str | None
230 : Default `X-Request-Id` header value for outgoing requests that
231 don't already provide one. If `None` or omitted, generate a random
232 ID. When retrying failed requests, the same ID is used on all
236 : A timeout value for HTTP requests in seconds. Default 300 (5 minutes).
238 Additional keyword arguments will be passed directly to
239 `googleapiclient.discovery.build`.
242 http = httplib2.Http(
243 ca_certs=util.ca_certs_path(),
244 cache=http_cache('discovery') if cache else None,
245 disable_ssl_certificate_validation=bool(insecure),
247 if http.timeout is None:
248 http.timeout = timeout
249 http = _patch_http_request(http, token, num_retries)
251 # The first time a client is instantiated, temporarily route
252 # googleapiclient.http retry logs if they're not already. These are
253 # important because temporary problems fetching the discovery document
254 # can cause clients to appear to hang early. This can be removed after
255 # we have a more general story for handling googleapiclient logs (#20521).
256 client_logger = logging.getLogger('googleapiclient.http')
257 # "first time a client is instantiated" = thread that acquires this lock
258 # It is never released.
259 # googleapiclient sets up its own NullHandler so we detect if logging is
260 # configured by looking for a real handler anywhere in the hierarchy.
261 client_logger_unconfigured = _googleapiclient_log_lock.acquire(blocking=False) and all(
262 isinstance(handler, logging.NullHandler)
263 for logger_name in ['', 'googleapiclient', 'googleapiclient.http']
264 for handler in logging.getLogger(logger_name).handlers
266 if client_logger_unconfigured:
267 client_level = client_logger.level
268 client_filter = GoogleHTTPClientFilter()
269 client_logger.addFilter(client_filter)
270 client_logger.addHandler(log_handler)
271 if logging.NOTSET < client_level < client_filter.retry_levelno:
272 client_logger.setLevel(client_level)
274 client_logger.setLevel(client_filter.retry_levelno)
276 svc = apiclient_discovery.build(
278 cache_discovery=False,
279 discoveryServiceUrl=discoveryServiceUrl,
281 num_retries=num_retries,
285 if client_logger_unconfigured:
286 client_logger.removeHandler(log_handler)
287 client_logger.removeFilter(client_filter)
288 client_logger.setLevel(client_level)
289 svc.api_token = token
290 svc.insecure = insecure
291 svc.request_id = request_id
292 svc.config = lambda: util.get_config_once(svc)
293 svc.vocabulary = lambda: util.get_vocabulary_once(svc)
294 svc.close_connections = types.MethodType(_close_connections, svc)
295 http.max_request_size = svc._rootDesc.get('maxRequestSize', 0)
297 http._request_id = lambda: svc.request_id or util.new_request_id()
300 def normalize_api_kwargs(
302 discoveryServiceUrl=None,
307 """Validate kwargs from `api` and build kwargs for `api_client`
309 This method takes high-level keyword arguments passed to the `api`
310 constructor and normalizes them into a new dictionary that can be passed
311 as keyword arguments to `api_client`. It raises `ValueError` if required
312 arguments are missing or conflict.
317 : A string naming the version of the Arvados API to use. If not specified,
318 the code will log a warning and fall back to 'v1'.
320 discoveryServiceUrl: str | None
321 : The URL used to discover APIs passed directly to
322 `googleapiclient.discovery.build`. It is an error to pass both
323 `discoveryServiceUrl` and `host`.
326 : The hostname and optional port number of the Arvados API server. Used to
327 build `discoveryServiceUrl`. It is an error to pass both
328 `discoveryServiceUrl` and `host`.
331 : The authentication token to send with each API call.
333 Additional keyword arguments will be included in the return value.
335 if discoveryServiceUrl and host:
336 raise ValueError("both discoveryServiceUrl and host provided")
337 elif discoveryServiceUrl:
338 url_src = "discoveryServiceUrl"
340 url_src = "host argument"
341 discoveryServiceUrl = 'https://%s/discovery/v1/apis/{api}/{apiVersion}/rest' % (host,)
343 # This specific error message gets priority for backwards compatibility.
344 raise ValueError("token argument provided, but host missing.")
346 raise ValueError("neither discoveryServiceUrl nor host provided")
348 raise ValueError("%s provided, but token missing" % (url_src,))
352 "Using default API version. Call arvados.api(%r) instead.",
356 'discoveryServiceUrl': discoveryServiceUrl,
362 def api_kwargs_from_config(version=None, apiconfig=None, **kwargs):
363 """Build `api_client` keyword arguments from configuration
365 This function accepts a mapping with Arvados configuration settings like
366 `ARVADOS_API_HOST` and converts them into a mapping of keyword arguments
367 that can be passed to `api_client`. If `ARVADOS_API_HOST` or
368 `ARVADOS_API_TOKEN` are not configured, it raises `ValueError`.
373 : A string naming the version of the Arvados API to use. If not specified,
374 the code will log a warning and fall back to 'v1'.
376 apiconfig: Mapping[str, str] | None
377 : A mapping with entries for `ARVADOS_API_HOST`, `ARVADOS_API_TOKEN`, and
378 optionally `ARVADOS_API_HOST_INSECURE`. If not provided, calls
379 `arvados.config.settings` to get these parameters from user configuration.
381 Additional keyword arguments will be included in the return value.
383 if apiconfig is None:
384 apiconfig = config.settings()
385 missing = " and ".join(
387 for key in ['ARVADOS_API_HOST', 'ARVADOS_API_TOKEN']
388 if key not in apiconfig
392 "%s not set.\nPlease set in %s or export environment variable." %
393 (missing, config.default_config_file),
395 return normalize_api_kwargs(
398 apiconfig['ARVADOS_API_HOST'],
399 apiconfig['ARVADOS_API_TOKEN'],
400 insecure=config.flag_is_true('ARVADOS_API_HOST_INSECURE', apiconfig),
404 def api(version=None, cache=True, host=None, token=None, insecure=False,
405 request_id=None, timeout=5*60, *,
406 discoveryServiceUrl=None, **kwargs):
407 """Dynamically build an Arvados API client
409 This function provides a high-level "do what I mean" interface to build an
410 Arvados API client object. You can call it with no arguments to build a
411 client from user configuration; pass `host` and `token` arguments just
412 like you would write in user configuration; or pass additional arguments
413 for lower-level control over the client.
415 This function returns a `arvados.safeapi.ThreadSafeApiCache`, an
416 API-compatible wrapper around `googleapiclient.discovery.Resource`. If
417 you're handling concurrency yourself and/or your application is very
418 performance-sensitive, consider calling `api_client` directly.
423 : A string naming the version of the Arvados API to use. If not specified,
424 the code will log a warning and fall back to 'v1'.
427 : The hostname and optional port number of the Arvados API server.
430 : The authentication token to send with each API call.
432 discoveryServiceUrl: str | None
433 : The URL used to discover APIs passed directly to
434 `googleapiclient.discovery.build`.
436 If `host`, `token`, and `discoveryServiceUrl` are all omitted, `host` and
437 `token` will be loaded from the user's configuration. Otherwise, you must
438 pass `token` and one of `host` or `discoveryServiceUrl`. It is an error to
439 pass both `host` and `discoveryServiceUrl`.
441 Other arguments are passed directly to `api_client`. See that function's
442 docstring for more information about their meaning.
447 request_id=request_id,
450 if discoveryServiceUrl or host or token:
451 kwargs.update(normalize_api_kwargs(version, discoveryServiceUrl, host, token))
453 kwargs.update(api_kwargs_from_config(version))
454 version = kwargs.pop('version')
455 # We do the import here to avoid a circular import at the top level.
456 from .safeapi import ThreadSafeApiCache
457 return ThreadSafeApiCache({}, {}, kwargs, version)
459 def api_from_config(version=None, apiconfig=None, **kwargs):
460 """Build an Arvados API client from a configuration mapping
462 This function builds an Arvados API client from a mapping with user
463 configuration. It accepts that mapping as an argument, so you can use a
464 configuration that's different from what the user has set up.
466 This function returns a `arvados.safeapi.ThreadSafeApiCache`, an
467 API-compatible wrapper around `googleapiclient.discovery.Resource`. If
468 you're handling concurrency yourself and/or your application is very
469 performance-sensitive, consider calling `api_client` directly.
474 : A string naming the version of the Arvados API to use. If not specified,
475 the code will log a warning and fall back to 'v1'.
477 apiconfig: Mapping[str, str] | None
478 : A mapping with entries for `ARVADOS_API_HOST`, `ARVADOS_API_TOKEN`, and
479 optionally `ARVADOS_API_HOST_INSECURE`. If not provided, calls
480 `arvados.config.settings` to get these parameters from user configuration.
482 Other arguments are passed directly to `api_client`. See that function's
483 docstring for more information about their meaning.
485 return api(**api_kwargs_from_config(version, apiconfig, **kwargs))
487 class OrderedJsonModel(apiclient.model.JsonModel):
488 """Model class for JSON that preserves the contents' order
490 .. WARNING:: Deprecated
491 This model is redundant now that Python dictionaries preserve insertion
492 ordering. Code that passes this model to API constructors can remove it.
494 In Python versions before 3.6, API clients that cared about preserving the
495 order of fields in API server responses could use this model to do so.
496 Typical usage looked like:
498 from arvados.api import OrderedJsonModel
499 client = arvados.api('v1', ..., model=OrderedJsonModel())
501 @util._deprecated(preferred="the default model and rely on Python's built-in dictionary ordering")
502 def __init__(self, data_wrapper=False):
503 return super().__init__(data_wrapper)
506 RETRY_DELAY_INITIAL = 0
508 .. WARNING:: Deprecated
509 This constant was used by retry code in previous versions of the Arvados SDK.
510 Changing the value has no effect anymore.
511 Prefer passing `num_retries` to an API client constructor instead.
512 Refer to the constructor docstrings for details.
515 RETRY_DELAY_BACKOFF = 0
517 .. WARNING:: Deprecated
518 This constant was used by retry code in previous versions of the Arvados SDK.
519 Changing the value has no effect anymore.
520 Prefer passing `num_retries` to an API client constructor instead.
521 Refer to the constructor docstrings for details.
526 .. WARNING:: Deprecated
527 This constant was used by retry code in previous versions of the Arvados SDK.
528 Changing the value has no effect anymore.
529 Prefer passing `num_retries` to an API client constructor instead.
530 Refer to the constructor docstrings for details.