1 # -*- encoding: binary -*-
3 # \MogileFS file manipulation client.
5 # Create a new instance that will communicate with these trackers:
6 # hosts = %w[192.168.1.69:6001 192.168.1.70:6001]
7 # mg = MogileFS::MogileFS.new(:domain => 'test', :hosts => hosts)
9 # # Stores "A bunch of text to store" into 'some_key' with a class of 'text'.
10 # mg.store_content('some_key', 'text', "A bunch of text to store")
12 # # Retrieve data from 'some_key' as a string
13 # data = mg.get_file_data('some_key')
15 # # Store the contents of 'image.jpeg' into the key 'my_image' with a
17 # mg.store_file('my_image', 'image', 'image.jpeg')
19 # # Store the contents of 'image.jpeg' into the key 'my_image' with a
20 # # class of 'image' using an open IO object.
21 # File.open('image.jpeg') { |fp| mg.store_file('my_image', 'image', fp) }
23 # # Retrieve the contents of 'my_image' into '/path/to/huge_file'
24 # # without slurping the entire contents into memory:
25 # mg.get_file_data('my_image', '/path/to/huge_file')
27 # # Remove the key 'my_image' and 'some_key'.
28 # mg.delete('my_image')
29 # mg.delete('some_key')
31 class MogileFS::MogileFS < MogileFS::Client
32 include MogileFS::Bigfile
34 # The domain of keys for this MogileFS client.
37 # The timeout for get_file_data (per-read() system call).
38 # Defaults to five seconds.
39 attr_accessor :get_file_data_timeout
41 # The maximum allowed time for creating a new_file. Defaults to 1 hour.
42 attr_accessor :new_file_max_time
44 # Creates a new MogileFS::MogileFS instance. +args+ must include a key
45 # :domain specifying the domain of this client.
47 # Optional parameters for +args+:
49 # [:get_file_data_timeout => Numeric]
51 # See get_file_data_timeout
53 # [:new_file_max_time => Numeric]
55 # See new_file_max_time
57 # [:fail_timeout => Numeric]
59 # Delay before retrying a failed tracker backends.
60 # Defaults to 5 seconds.
62 # [:timeout => Numeric]
64 # Timeout for tracker backend responses.
65 # Defaults to 3 seconds.
67 def initialize(args = {})
68 @domain = args[:domain]
70 @get_file_data_timeout = args[:get_file_data_timeout] || 5
71 @new_file_max_time = args[:new_file_max_time] || 3600.0
72 @nhp_get = MogileFS::NHP.new('get')
73 @nhp_get.open_timeout = @nhp_get.read_timeout = @get_file_data_timeout
74 @nhp_put = MogileFS::NHP.new('put')
75 @nhp_put.open_timeout = @nhp_put.read_timeout = @new_file_max_time
77 raise ArgumentError, "you must specify a domain" unless @domain
79 if @backend = args[:db_backend]
86 # Enumerates keys, limited by optional +prefix+
87 # +args+ may also be specified for an optional +:limit+
88 # and +:after+ (default: nil)
89 def each_key(prefix = "", args = nil, &block)
96 keys, after = list_keys(prefix, after, limit || 1000, &block)
97 return unless keys && keys[0]
98 limit -= keys.size if limit
99 end while limit == nil || limit > 0
103 # Enumerates keys and yields a +file_info+ hash for each key matched by
105 def each_file_info(prefix = "", args = nil)
106 # FIXME: there's a lot of duplicate code from list_keys_verbose here...
107 raise ArgumentError, "need block" unless block_given?
108 ordered = ready = nil
109 on_file_info = lambda do |info|
110 Hash === info or raise info
111 file_info_cleanup(info)
113 # deal with trackers with multiple queryworkers responding out-of-order
114 ready[info["key"]] = info
115 while info = ready.delete(ordered[-1])
122 opts = { :domain => @domain }
123 opts[:devices] = 1 if args && args[:devices]
124 after = args ? args[:after] : nil
125 limit = args ? args[:limit] : nil
128 keys, after = list_keys(prefix, after, limit || 1000)
129 return nr unless keys && keys[0]
130 ordered = keys.reverse
133 limit -= keys.size if limit
137 @backend.pipeline_dispatch(:file_info, opts, &on_file_info)
139 @backend.pipeline_wait
140 rescue MogileFS::PipelineError, SystemCallError,
141 MogileFS::RequestTruncatedError,
142 MogileFS::UnreadableSocketError,
143 MogileFS::InvalidResponseError, # truncated response
146 keys = (ordered - ready.keys).reverse!
148 end while limit == nil || limit > 0
154 # Retrieves the contents of +key+. If +dst+ is specified, +dst+
155 # should be an IO-like object capable of receiving the +write+ method
156 # or a path name. +copy_length+ may be specified to limit the number of
157 # bytes to retrieve, and +src_offset+ can be specified to specified the
158 # start position of the copy.
159 def get_file_data(key, dst = nil, copy_length = nil, src_offset = nil)
160 paths = get_paths(key)
161 if src_offset || copy_length
163 range_end = copy_length ? src_offset + copy_length - 1 : nil
164 range = [ src_offset, range_end ]
168 sock = MogileFS::HTTPReader.first(paths, @get_file_data_timeout, range)
171 sock = MogileFS::HTTPReader.first(paths, @get_file_data_timeout, range)
176 uri = URI.parse(path)
177 get = Net::HTTP::Get.new(uri.path)
178 get["range"] = "bytes=#{range[0]}-#{range[1]}" if range
180 res = @nhp_get.request(uri, get)
185 (errors ||= []) << "#{path} - #{res.message} (#{res.class})"
187 (errors ||= []) << "#{path} - #{e.message} (#{e.class})"
190 raise MogileFS::Error,
191 "all paths failed with GET: #{errors.join(', ')}", []
194 sock.close if sock && ! sock.closed?
197 # Get the paths (URLs as strings) for +key+. If +args+ is specified,
199 # - :noverify -> boolean, whether or not the tracker checks (default: true)
200 # - :pathcount -> a positive integer of URLs to retrieve (default: 2)
201 # - :zone -> "alt" or nil (default: nil)
203 # :noverify defaults to true because this client library is capable of
204 # verifying paths for readability itself. It is also faster and more
205 # reliable to verify paths on the client.
206 def get_paths(key, *args)
210 :noverify => args[0],
215 opts[:noverify] = args[:noverify]
216 zone = args[:zone] and opts[:zone] = zone
217 pathcount = args[:pathcount] and opts[:pathcount] = pathcount.to_i
220 opts[:noverify] = false == opts[:noverify] ? 0 : 1
221 @backend.respond_to?(:_get_paths) and return @backend._get_paths(opts)
222 res = @backend.get_paths(opts)
223 (1..res['paths'].to_i).map { |i| res["path#{i}"] }
226 # Returns +true+ if +key+ exists, +false+ if not
228 args = { :key => key, :domain => @domain , :ruby_no_raise => true}
229 case rv = @backend.get_paths(args)
232 when MogileFS::Backend::UnknownKeyError
239 # Get the URIs for +key+ (paths) as URI::HTTP objects
240 def get_uris(key, *args)
241 get_paths(key, *args).map! { |path| URI.parse(path) }
244 # Creates a new file +key+ in the domain of this object.
246 # +bytes+ is the expected size of the file if known in advance
248 # It operates like File.open(..., "w") and may take an optional
249 # block, yielding an IO-like object with support for the methods
250 # documented in MogileFS::NewFile::Writer.
252 # This atomically replaces existing data stored as +key+
253 # when the block exits or when the returned object is closed.
255 # +args+ may contain the following options:
257 # [:content_length => Integer]
259 # This has the same effect as the (deprecated) +bytes+ parameter.
261 # [ :largefile => :stream, :content_range or :tempfile ]
263 # See MogileFS::NewFile for more information on this
265 # [ :class => String]
267 # The MogileFS storage class of the object.
269 # [:content_md5 => String, Proc, or :trailer]
271 # This can either be a Base64-encoded String, a Proc object, or
272 # the :trailer symbol. If given a String, it will be used as the
273 # Content-MD5 HTTP header. If given the :trailer symbol, this library
274 # will automatically generate an Content-MD5 HTTP trailer. If given
275 # a Proc object, this Proc object should give a Base64-encoded string
276 # which can be used as the Content-MD5 HTTP trailer when called at the
277 # end of the request.
279 # Keep in mind most HTTP servers do not support HTTP trailers, so
280 # passing a String is usually the safest way to use this.
284 # This is an empty hash that will be filled the same information
285 # MogileFS::MogileFS#file_info.
287 # Additionally, it contains one additional key: :uris,
288 # an array of URI::HTTP objects to the stored destinations
289 def new_file(key, args = nil, bytes = nil) # :yields: file
290 raise MogileFS::ReadOnlyError if readonly?
291 opts = { :key => key, :multi_dest => 1 }
294 opts[:domain] = args[:domain]
295 open_args = args[:create_open_args]
296 klass = args[:class] and "default" != klass and opts[:class] = klass
298 opts[:class] = args if "default" != args
300 opts[:domain] ||= @domain
301 res = @backend.create_open(open_args ? open_args.merge(opts) : opts)
302 opts[:nhp_put] = @nhp_put
304 dests = if dev_count = res['dev_count'] # multi_dest succeeded
305 (1..dev_count.to_i).map { |i| [res["devid_#{i}"], res["path_#{i}"]] }
306 else # single destination returned
307 # 0x0040: d0e4 4f4b 2064 6576 6964 3d31 2666 6964 ..OK.devid=1&fid
308 # 0x0050: 3d33 2670 6174 683d 6874 7470 3a2f 2f31 =3&path=http://1
309 # 0x0060: 3932 2e31 3638 2e31 2e37 323a 3735 3030 92.168.1.72:7500
310 # 0x0070: 2f64 6576 312f 302f 3030 302f 3030 302f /dev1/0/000/000/
311 # 0x0080: 3030 3030 3030 3030 3033 2e66 6964 0d0a 0000000003.fid..
313 [[res['devid'], res['path']]]
316 opts.merge!(args) if Hash === args
317 opts[:backend] = @backend
318 opts[:fid] = res['fid']
319 opts[:content_length] ||= bytes if bytes
320 opts[:new_file_max_time] ||= @new_file_max_time
321 opts[:start_time] = Time.now
322 info = opts[:info] and info["class"] = klass || "default"
324 case (dests[0][1] rescue nil)
326 http_file = MogileFS::NewFile.new(dests, opts)
329 return http_file.commit # calls create_close
334 raise MogileFS::EmptyPathError,
335 "Empty path for mogile upload res=#{res.inspect}"
337 raise MogileFS::UnsupportedPathError,
338 "paths '#{dests.inspect}' returned by backend is not supported"
342 # Copies the contents of +file+ into +key+ in class +klass+. +file+ can be
343 # either a path name (String or Pathname object) or an IO-like object that
344 # responds to #read or #readpartial. Returns size of +file+ stored.
345 # This atomically replaces existing data stored as +key+
346 def store_file(key, klass, file, opts = nil)
347 raise MogileFS::ReadOnlyError if readonly?
348 (opts ||= {})[:class] = klass if String === klass
350 new_file(key, opts) { |mfp| mfp.big_io = file }
353 # Stores +content+ into +key+ in class +klass+, where +content+ is a String
354 # This atomically replaces existing data stored as +key+
355 def store_content(key, klass, content, opts = nil)
356 raise MogileFS::ReadOnlyError if readonly?
357 (opts ||= {})[:class] = klass if String === klass
359 new_file(key, opts) do |mfp|
360 if content.is_a?(MogileFS::Util::StoreContent)
361 mfp.streaming_io = content
370 raise MogileFS::ReadOnlyError if readonly?
372 @backend.delete :domain => @domain, :key => key
376 # Updates +key+ to +newclass+
377 def updateclass(key, newclass)
378 raise MogileFS::ReadOnlyError if readonly?
380 @backend.updateclass(:domain => @domain, :key => key, :class => newclass)
384 # Sleeps +duration+, only used for testing
385 def sleep(duration) # :nodoc:
386 @backend.sleep :duration => duration
389 # Renames a key +from+ to key +to+.
391 raise MogileFS::ReadOnlyError if readonly?
393 @backend.rename :domain => @domain, :from_key => from, :to_key => to
397 # Returns the size of +key+.
399 @backend.respond_to?(:_size) and return @backend._size(domain, key)
401 file_info(key)["length"].to_i
402 rescue MogileFS::Backend::UnknownCommandError
403 paths_size(get_paths(key))
407 def paths_size(paths) # :nodoc:
408 require "mogilefs/paths_size"
409 MogileFS::PathsSize.call(paths)
412 # Lists keys starting with +prefix+ following +after+ up to +limit+. If
413 # +after+ is nil the list starts at the beginning.
414 def list_keys(prefix = "", after = nil, limit = 1000, &block)
415 @backend.respond_to?(:_list_keys) and
416 return @backend._list_keys(domain, prefix, after, limit, &block)
418 res = @backend.list_keys(:domain => domain, :prefix => prefix,
419 :after => after, :limit => limit,
420 :ruby_no_raise => true)
421 MogileFS::Backend::NoneMatchError === res and return
422 raise res if MogileFS::Error === res
424 keys = (1..res['key_count'].to_i).map { |i| res["key_#{i}"] }
427 keys.each { |key| block.call(key) }
429 list_keys_verbose(keys, block)
433 [ keys, res['next_after'] ]
436 def list_keys_verbose(keys, block) # :nodoc:
437 # emulate the MogileFS::Mysql interface, slowly...
438 ordered = keys.reverse
440 on_file_info = lambda do |info|
441 Hash === info or raise info
442 file_info_cleanup(info)
444 # deal with trackers with multiple queryworkers responding out-of-order
445 ready[info["key"]] = info
446 while info = ready.delete(ordered[-1])
447 block.call(ordered.pop, info["length"], info["devcount"])
450 opts = { :domain => @domain }
454 @backend.pipeline_dispatch(:file_info, opts, &on_file_info)
456 @backend.pipeline_wait
457 rescue MogileFS::Backend::UnknownCommandError # MogileFS < 2.45
458 @backend.shutdown # reset the socket
459 args = { :pathcount => 0x7fffffff }
461 paths = get_paths(key, args)
462 block.call(key, paths_size(paths), paths.size)
464 rescue MogileFS::PipelineError, SystemCallError,
465 MogileFS::RequestTruncatedError,
466 MogileFS::UnreadableSocketError,
467 MogileFS::InvalidResponseError, # truncated response
470 keys = (ordered - ready.keys).reverse!
478 # Return metadata about a file as a hash.
479 # Returns the domain, class, length, devcount, etc. as keys.
480 # Optionally, device ids (not paths) can be returned as
481 # well if :devices is specified and +true+.
483 # This should only be used for informational purposes, and not usually
484 # for dynamically serving files.
486 # mg.file_info("bar")
493 # "class" => "default",
497 def file_info(key, args = nil)
498 opts = { :domain => @domain, :key => key }
499 args and devices = args[:devices] and opts[:devices] = devices ? 1 : 0
500 file_info_cleanup(@backend.file_info(opts))
503 def file_info_cleanup(rv) # :nodoc:
504 %w(fid length devcount).each { |f| rv[f] = rv[f].to_i }
505 devids = rv["devids"] and
506 rv["devids"] = devids.split(/,/).map! { |x| x.to_i }
510 # Given an Integer +fid+ or String +key+ and domain, thorougly search
511 # the database for all occurences of a particular fid.
513 # Use this sparingly, this command hits the master database numerous
514 # times and is very expensive. This is not for production use, only
515 # troubleshooting and debugging.
517 # Searches for fid=666:
519 # client.file_debug(666)
521 # Search for key=foo using the default domain for this object:
523 # client.file_debug("foo")
525 # Search for key=foo in domain="bar":
527 # client.file_debug(:key => "foo", :domain => "bar")
531 when Integer then args = { "fid" => args }
532 when String then args = { "key" => args }
534 opts = { :domain => args[:domain] || @domain }.merge!(args)
536 rv = @backend.file_debug(opts)
539 when /_(?:classid|devcount|dmid|fid|length|
540 nexttry|fromdevid|failcount|flags|devid|type)\z/x
543 rv[k] = v.split(/,/).map! { |x| x.to_i }