Class: Net::SSH::Multi::Session

Inherits:
Object
  • Object
show all
Includes:
SessionActions
Defined in:
lib/net/ssh/multi/session.rb

Overview

Represents a collection of connections to various servers. It provides an interface for organizing the connections (#group), as well as a way to scope commands to a subset of all connections (#with). You can also provide a default gateway connection that servers should use when connecting (#via). It exposes an interface similar to Net::SSH::Connection::Session for opening SSH channels and executing commands, allowing for these operations to be done in parallel across multiple connections.

Net::SSH::Multi.start do |session|
  # access servers via a gateway
  session.via 'gateway', 'gateway-user'

  # define the servers we want to use
  session.use 'user1@host1'
  session.use 'user2@host2'

  # define servers in groups for more granular access
  session.group :app do
    session.use 'user@app1'
    session.use 'user@app2'
  end

  # execute commands on all servers
  session.exec "uptime"

  # execute commands on a subset of servers
  session.with(:app).exec "hostname"

  # run the aggregated event loop
  session.loop
end

Note that connections are established lazily, as soon as they are needed. You can force the connections to be opened immediately, though, using the #connect! method.

Concurrent Connection Limiting

Sometimes you may be dealing with a large number of servers, and if you try to have connections open to all of them simultaneously you’ll run into open file handle limitations and such. If this happens to you, you can set the #concurrent_connections property of the session. Net::SSH::Multi will then ensure that no more than this number of connections are ever open simultaneously.

Net::SSH::Multi.start(:concurrent_connections => 5) do |session|
  # ...
end

Opening channels and executing commands will still work exactly as before, but Net::SSH::Multi will transparently close finished connections and open pending ones.

Controlling Connection Errors

By default, Net::SSH::Multi will raise an exception if a connection error occurs when connecting to a server. This will typically bubble up and abort the entire connection process. Sometimes, however, you might wish to ignore connection errors, for instance when starting a daemon on a large number of boxes and you know that some of the boxes are going to be unavailable.

To do this, simply set the #on_error property of the session to :ignore (or to :warn, if you want a warning message when a connection attempt fails):

Net::SSH::Multi.start(:on_error => :ignore) do |session|
  # ...
end

The default is :fail, which causes the exception to bubble up. Additionally, you can specify a Proc object as the value for #on_error, which will be invoked with the server in question if the connection attempt fails. You can force the connection attempt to retry by throwing the :go symbol, with :retry as the payload, or force the exception to be reraised by throwing :go with :raise as the payload:

handler = Proc.new do |server|
  server[:connection_attempts] ||= 0
  if server[:connection_attempts] < 3
    server[:connection_attempts] += 1
    throw :go, :retry
  else
    throw :go, :raise
  end
end

Net::SSH::Multi.start(:on_error => handler) do |session|
  # ...
end

Any other thrown value (or no thrown value at all) will result in the failure being ignored.

Lazily Evaluated Server Definitions

Sometimes you might be dealing with an environment where you don’t know the names or addresses of the servers until runtime. You can certainly dynamically build server names and pass them to #use, but if the operation to determine the server names is expensive, you might want to defer it until the server is actually needed (especially if the logic of your program is such that you might not even need to connect to that server every time the program runs).

You can do this by passing a block to #use:

session.use do |opt|
  lookup_ip_address_of_remote_host
end

See #use for more information about this usage.

Instance Attribute Summary collapse

Instance Method Summary collapse

Methods included from SessionActions

#busy?, #connect!, #exec, #master, #open_channel, #send_global_request, #sessions

Constructor Details

#initialize(options = {}) ⇒ Session

Creates a new Net::SSH::Multi::Session instance. Initially, it contains no server definitions, no group definitions, and no default gateway.

You can set the #concurrent_connections property in the options. Setting it to nil (the default) will cause Net::SSH::Multi to ignore any concurrent connection limit and allow all defined sessions to be open simultaneously. Setting it to an integer will cause Net::SSH::Multi to allow no more than that number of concurrently open sessions, opening subsequent sessions only when other sessions finish and close.

Net::SSH::Multi.start(:concurrent_connections => 10) do |session|
  session.use ...
end



171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
 
# File 'lib/net/ssh/multi/session.rb', line 171

def initialize(options={})
  @server_list = ServerList.new
  @groups = Hash.new { |h,k| h[k] = ServerList.new }
  @gateway = nil
  @open_groups = []
  @connect_threads = []
  @on_error = :fail
  @default_user = ENV['USER'] || ENV['USERNAME'] || "unknown"

  @open_connections = 0
  @pending_sessions = []
  @session_mutex = Mutex.new

  options.each { |opt, value| send("#{opt}=", value) }
end

Instance Attribute Details

#concurrent_connections ⇒ Object

The number of allowed concurrent connections. No more than this number of sessions will be open at any given time.



138
139
140
 
# File 'lib/net/ssh/multi/session.rb', line 138

def concurrent_connections
  @concurrent_connections
end

#default_gateway ⇒ Object (readonly)

The default Net::SSH::Gateway instance to use to connect to the servers. If nil, no default gateway will be used.



130
131
132
 
# File 'lib/net/ssh/multi/session.rb', line 130

def default_gateway
  @default_gateway
end

#default_user ⇒ Object

The default user name to use when connecting to a server. If a user name is not given for a particular server, this value will be used. It defaults to ENV || ENV, or “unknown” if neither of those are set.



149
150
151
 
# File 'lib/net/ssh/multi/session.rb', line 149

def default_user
  @default_user
end

#groups ⇒ Object (readonly)

The hash of group definitions, mapping each group name to a corresponding Net::SSH::Multi::ServerList.



134
135
136
 
# File 'lib/net/ssh/multi/session.rb', line 134

def groups
  @groups
end

#on_error ⇒ Object

How connection errors should be handled. This defaults to :fail, but may be set to :ignore if connection errors should be ignored, or :warn if connection errors should cause a warning.



143
144
145
 
# File 'lib/net/ssh/multi/session.rb', line 143

def on_error
  @on_error
end

#open_connections ⇒ Object (readonly)

The number of connections that are currently open.



152
153
154
 
# File 'lib/net/ssh/multi/session.rb', line 152

def open_connections
  @open_connections
end

#open_groups ⇒ Object (readonly)

The list of “open” groups, which will receive subsequent server definitions. See #use and #group.



156
157
158
 
# File 'lib/net/ssh/multi/session.rb', line 156

def open_groups
  @open_groups
end

#server_list ⇒ Object (readonly)

The Net::SSH::Multi::ServerList managed by this session.



126
127
128
 
# File 'lib/net/ssh/multi/session.rb', line 126

def server_list
  @server_list
end

Instance Method Details

#close ⇒ Object

Closes the multi-session by shutting down all open server sessions, and the default gateway (if one was specified using #via). Note that other gateway connections (e.g., those passed to #use directly) will not be closed by this method, and must be managed externally.



402
403
404
405
406
407
 
# File 'lib/net/ssh/multi/session.rb', line 402

def close
  server_list.each { |server| server.close_channels }
  loop(0) { busy?(true) }
  server_list.each { |server| server.close }
  default_gateway.shutdown! if default_gateway
end

#group(*args) ⇒ Object

At its simplest, this associates a named group with a server definition. It can be used in either of two ways:

First, you can use it to associate a group (or array of groups) with a server definition (or array of server definitions). The server definitions must already exist in the #server_list array (typically by calling #use):

server1 = session.use('host1', 'user1')
server2 = session.use('host2', 'user2')
session.group :app => server1, :web => server2
session.group :staging => [server1, server2]
session.group %w(xen linux) => server2
session.group %w(rackspace backup) => [server1, server2]

Secondly, instead of a mapping of groups to servers, you can just provide a list of group names, and then a block. Inside the block, any calls to #use will automatically associate the new server definition with those groups. You can nest #group calls, too, which will aggregate the group definitions.

session.group :rackspace, :backup do
  session.use 'host1', 'user1'
  session.group :xen do
    session.use 'host2', 'user2'
  end
end



213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
 
# File 'lib/net/ssh/multi/session.rb', line 213

def group(*args)
  mapping = args.last.is_a?(Hash) ? args.pop : {}

  if mapping.any? && block_given?
    raise ArgumentError, "must provide group mapping OR block, not both"
  elsif block_given?
    begin
      saved_groups = open_groups.dup
      open_groups.concat(args.map { |a| a.to_sym }).uniq!
      yield self
    ensure
      open_groups.replace(saved_groups)
    end
  else
    mapping.each do |key, value|
      (open_groups + Array(key)).uniq.each do |grp|
        groups[grp.to_sym].concat(Array(value))
      end
    end
  end
end

#loop(wait = nil, &block) ⇒ Object

Run the aggregated event loop for all open server sessions, until the given block returns false. If no block is given, the loop will run for as long as #busy? returns true (in other words, for as long as there are any (non-invisible) channels open).



415
416
417
418
 
# File 'lib/net/ssh/multi/session.rb', line 415

def loop(wait=nil, &block)
  running = block || Proc.new { |c| busy? }
  loop_forever { break unless process(wait, &running) }
end

#loop_forever ⇒ Object 



409
 
# File 'lib/net/ssh/multi/session.rb', line 409

alias :loop_forever :loop

#next_session(server, force = false) ⇒ Object

Takes the #concurrent_connections property into account, and tries to return a new session for the given server. If the concurrent connections limit has been reached, then a Net::SSH::Multi::PendingConnection instance will be returned instead, which will be realized into an actual session as soon as a slot opens up.

If force is true, the concurrent_connections check is skipped and a real connection is always returned.



468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
 
# File 'lib/net/ssh/multi/session.rb', line 468

def next_session(server, force=false) #:nodoc:
  # don't retry a failed attempt
  return nil if server.failed?

  @session_mutex.synchronize do
    if !force && concurrent_connections && concurrent_connections <= open_connections
      connection = PendingConnection.new(server)
      @pending_sessions << connection
      return connection
    end

    @open_connections += 1
  end

  begin
    server.new_session

  # I don't understand why this should be necessary--StandardError is a
  # subclass of Exception, after all--but without explicitly rescuing
  # StandardError, things like Errno::* and SocketError don't get caught
  # here!
  rescue Exception, StandardError => e
    server.fail!
    @session_mutex.synchronize { @open_connections -= 1 }

    case on_error
    when :ignore then
      # do nothing
    when :warn then
      warn("error connecting to #{server}: #{e.class} (#{e.message})")
    when Proc then
      go = catch(:go) { on_error.call(server); nil }
      case go
      when nil, :ignore then # nothing
      when :retry then retry
      when :raise then raise
      else warn "unknown 'go' command: #{go.inspect}"
      end
    else
      raise
    end

    return nil
  end
end

#on(*servers) {|subsession| ... } ⇒ Object

Works as #with, but for specific servers rather than groups. It will return a new subsession (Net::SSH::Multi::Subsession) consisting of the given servers. (Note that it requires that the servers in question have been created via calls to #use on this session object, or things will not work quite right.) If a block is given, the new subsession will also be yielded to the block.

srv1 = session.use('host1', 'user')
srv2 = session.use('host2', 'user')
# ...
session.on(srv1, srv2).exec('hostname')

Yields:

  • (subsession) 


392
393
394
395
396
 
# File 'lib/net/ssh/multi/session.rb', line 392

def on(*servers)
  subsession = Subsession.new(self, servers)
  yield subsession if block_given?
  subsession
end

#postprocess(readers, writers) ⇒ Object

Runs the postprocess stage on all servers. Always returns true. This is called as part of the #process method.



455
456
457
458
 
# File 'lib/net/ssh/multi/session.rb', line 455

def postprocess(readers, writers) #:nodoc:
  server_list.each { |server| server.postprocess(readers, writers) }
  true
end

#preprocess(&block) ⇒ Object

Runs the preprocess stage on all servers. Returns false if the block returns false, and true if there either is no block, or it returns true. This is called as part of the #process method.



447
448
449
450
451
 
# File 'lib/net/ssh/multi/session.rb', line 447

def preprocess(&block) #:nodoc:
  return false if block && !block[self]
  server_list.each { |server| server.preprocess }
  block.nil? || block[self]
end

#process(wait = nil, &block) ⇒ Object

Run a single iteration of the aggregated event loop for all open server sessions. The wait parameter indicates how long to wait for an event to appear on any of the different sessions; nil (the default) means “wait forever”. If the block is given, then it will be used to determine whether #process returns true (the block did not return false), or false (the block returned false).



426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
 
# File 'lib/net/ssh/multi/session.rb', line 426

def process(wait=nil, &block)
  realize_pending_connections!
  wait = @connect_threads.any? ? 0 : wait

  return false unless preprocess(&block)

  readers = server_list.map { |s| s.readers }.flatten
  writers = server_list.map { |s| s.writers }.flatten

  readers, writers, = IO.select(readers, writers, nil, wait)

  if readers
    return postprocess(readers, writers)
  else
    return true
  end
end

#realize_pending_connections! ⇒ Object

Invoked by the event loop. If there is a concurrent_connections limit in effect, this will close any non-busy sessions and try to open as many new sessions as it can. It does this in threads, so that existing processing can continue.

If there is no concurrent_connections limit in effect, then this method does nothing.



532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
 
# File 'lib/net/ssh/multi/session.rb', line 532

def realize_pending_connections! #:nodoc:
  return unless concurrent_connections

  server_list.each do |server|
    server.close if !server.busy?(true)
    server.update_session!
  end

  @connect_threads.delete_if { |t| !t.alive? }

  count = concurrent_connections ? (concurrent_connections - open_connections) : @pending_sessions.length
  count.times do
    session = @pending_sessions.pop or break
    @connect_threads << Thread.new do
      session.replace_with(next_session(session.server, true))
    end
  end
end

#server_closed(server) ⇒ Object

Tells the session that the given server has closed its connection. The session indicates that a new connection slot is available, which may be filled by the next pending connection on the next event loop iteration.



517
518
519
520
521
522
523
 
# File 'lib/net/ssh/multi/session.rb', line 517

def server_closed(server) #:nodoc:
  @session_mutex.synchronize do
    unless @pending_sessions.delete(server.session)
      @open_connections -= 1
    end
  end
end

#servers ⇒ Object

Essentially an alias for #servers_for without any arguments. This is used primarily to satistfy the expectations of the Net::SSH::Multi::SessionActions module.



293
294
295
 
# File 'lib/net/ssh/multi/session.rb', line 293

def servers
  servers_for
end

#servers_for(*criteria) ⇒ Object

Returns the set of servers that match the given criteria. It can be used in any (or all) of three ways.

First, you can omit any arguments. In this case, the full list of servers will be returned.

all = session.servers_for

Second, you can simply specify a list of group names. All servers in all named groups will be returned. If a server belongs to multiple matching groups, then it will appear only once in the list (the resulting list will contain only unique servers).

servers = session.servers_for(:app, :db)

Last, you can specify a hash with group names as keys, and property constraints as the values. These property constraints are either “only” constraints (which restrict the set of servers to “only” those that match the given properties) or “except” constraints (which restrict the set of servers to those whose properties do not match). Properties are described when the server is defined (via the :properties key):

session.group :db do
  session.use 'dbmain', 'user', :properties => { :primary => true }
  session.use 'dbslave', 'user2'
  session.use 'dbslve2', 'user2'
end

# return ONLY on the servers in the :db group which have the :primary
# property set to true.
primary = session.servers_for(:db => { :only => { :primary => true } })

You can, naturally, combine these methods:

# all servers in :app and :web, and all servers in :db with the
# :primary property set to true
servers = session.servers_for(:app, :web, :db => { :only => { :primary => true } })



334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
 
# File 'lib/net/ssh/multi/session.rb', line 334

def servers_for(*criteria)
  if criteria.empty?
    server_list.flatten
  else
    # normalize the criteria list, so that every entry is a key to a
    # criteria hash (possibly empty).
    criteria = criteria.inject({}) do |hash, entry|
      case entry
      when Hash then hash.merge(entry)
      else hash.merge(entry => {})
      end
    end

    list = criteria.inject([]) do |aggregator, (group, properties)|
      raise ArgumentError, "the value for any group must be a Hash, but got a #{properties.class} for #{group.inspect}" unless properties.is_a?(Hash)
      bad_keys = properties.keys - [:only, :except]
      raise ArgumentError, "unknown constraint(s) #{bad_keys.inspect} for #{group.inspect}" unless bad_keys.empty?

      servers = groups[group].select do |server|
        (properties[:only] || {}).all? { |prop, value| server[prop] == value } &&
        !(properties[:except] || {}).any? { |prop, value| server[prop] == value }
      end

      aggregator.concat(servers)
    end

    list.uniq
  end
end

#use(*hosts, &block) ⇒ Object

Defines a new server definition, to be managed by this session. The server is at the given host, and will be connected to as the given user. The other options are passed as-is to the Net::SSH session constructor.

If a default gateway has been specified previously (with #via) it will be passed to the new server definition. You can override this by passing a different Net::SSH::Gateway instance (or nil) with the :via key in the options.

session.use 'host'
session.use 'user@host2', :via => nil
session.use 'host3', :user => "user3", :via => Net::SSH::Gateway.new('gateway.host', 'user')

If only a single host is given, the new server instance is returned. You can give multiple hosts at a time, though, in which case an array of server instances will be returned.

server1, server2 = session.use "host1", "host2"

If given a block, this will save the block as a Net::SSH::Multi::DynamicServer definition, to be evaluated lazily the first time the server is needed. The block will recive any options hash given to #use, and should return nil (if no servers are to be added), a String or an array of Strings (to be interpreted as a connection specification), or a Server or an array of Servers.



274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
 
# File 'lib/net/ssh/multi/session.rb', line 274

def use(*hosts, &block)
  options = hosts.last.is_a?(Hash) ? hosts.pop : {}
  options = { :via => default_gateway }.merge(options)

  results = hosts.map do |host|
    server_list.add(Server.new(self, host, options))
  end

  if block
    results << server_list.add(DynamicServer.new(self, options, block))
  end

  group [] => results
  results.length > 1 ? results : results.first
end

#via(host, user, options = {}) ⇒ Object

Sets up a default gateway to use when establishing connections to servers. Note that any servers defined prior to this invocation will not use the default gateway; it only affects servers defined subsequently.

session.via 'gateway.host', 'user'

You may override the default gateway on a per-server basis by passing the :via key to the #use method; see #use for details.



243
244
245
246
 
# File 'lib/net/ssh/multi/session.rb', line 243

def via(host, user, options={})
  @default_gateway = Net::SSH::Gateway.new(host, user, options)
  self
end

#with(*groups) {|subsession| ... } ⇒ Object

Returns a new Net::SSH::Multi::Subsession instance consisting of the servers that meet the given criteria. If a block is given, the subsession will be yielded to it. See #servers_for for a discussion of how these criteria are interpreted.

session.with(:app).exec('hostname')

session.with(:app, :db => { :primary => true }) do |s|
  s.exec 'date'
  s.exec 'uptime'
end

Yields:

  • (subsession) 


375
376
377
378
379
 
# File 'lib/net/ssh/multi/session.rb', line 375

def with(*groups)
  subsession = Subsession.new(self, servers_for(*groups))
  yield subsession if block_given?
  subsession
end