Class: Switchman::Shard

Inherits:
UnshardedRecord show all
Defined in:
lib/switchman/shard.rb

Constant Summary collapse

IDS_PER_SHARD =

ten trillion possible ids per shard. yup.

10_000_000_000_000
NIL_NIL_ID =

takes an id-ish, and returns a local id and the shard it’s local to. [nil, nil] if it can’t be interpreted. [id, nil] if it’s already a local ID. [nil, nil] if it’s a well formed id, but the shard it refers to does not exist

[nil, nil].freeze

Class Method Summary collapse

Instance Method Summary collapse

Class Method Details

._load(str) ⇒ Object 



489
490
491
 
# File 'lib/switchman/shard.rb', line 489

def self._load(str)
  lookup(str.to_i)
end

.activate(shards) ⇒ Object 



60
61
62
63
64
65
66
67
 
# File 'lib/switchman/shard.rb', line 60

def activate(shards)
  activated_classes = activate!(shards)
  yield
ensure
  activated_classes&.each do |klass|
    klass.connected_to_stack.pop
  end
end

.activate!(shards) ⇒ Object 



69
70
71
72
73
74
75
76
77
78
79
80
81
 
# File 'lib/switchman/shard.rb', line 69

def activate!(shards)
  activated_classes = nil
  shards.each do |klass, shard|
    next if klass == UnshardedRecord

    next unless klass.current_shard != shard.database_server.id.to_sym ||
                klass.current_switchman_shard != shard

    (activated_classes ||= []) << klass
    klass.connected_to_stack << { shard: shard.database_server.id.to_sym, klasses: [klass], switchman_shard: shard }
  end
  activated_classes
end

.active_shards ⇒ Object 



83
84
85
86
87
 
# File 'lib/switchman/shard.rb', line 83

def active_shards
  sharded_models.map do |klass|
    [klass, current(klass)]
  end.compact.to_h
end

.clear_cache ⇒ Object 



110
111
112
 
# File 'lib/switchman/shard.rb', line 110

def clear_cache
  cached_shards.clear
end

.current(klass = ::ActiveRecord::Base) ⇒ Object 



55
56
57
58
 
# File 'lib/switchman/shard.rb', line 55

def current(klass = ::ActiveRecord::Base)
  klass ||= ::ActiveRecord::Base
  klass.current_switchman_shard
end

.default(reload: false, with_fallback: false) ⇒ Object 



21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
 
# File 'lib/switchman/shard.rb', line 21

def default(reload: false, with_fallback: false)
  if !@default || reload
    # Have to create a dummy object so that several key methods still work
    # (it's easier to do this in one place here, and just assume that sharding
    # is up and running everywhere else).  This includes for looking up the
    # default shard itself. This also needs to be a local so that this method
    # can be re-entrant
    default = DefaultShard.instance

    # if we already have a default shard in place, and the caller wants
    # to use it as a fallback, use that instead of the dummy instance
    default = @default if with_fallback && @default

    # the first time we need a dummy dummy for re-entrancy to avoid looping on ourselves
    @default ||= default

    # Now find the actual record, if it exists
    @default = begin
      find_cached('default_shard') { Shard.where(default: true).take } || default
    rescue
      default
    end

    # make sure this is not erroneously cached
    @default.database_server.remove_instance_variable(:@primary_shard) if @default.database_server.instance_variable_defined?(:@primary_shard)

    # and finally, check for cached references to the default shard on the existing connection
    sharded_models.each do |klass|
      klass.connection.shard = @default if klass.connected? && klass.connection.shard.default?
    end
  end
  @default
end

.global_id_for(any_id, source_shard = nil) ⇒ Object

takes an id-ish, and returns an integral global id. returns nil if it can’t be interpreted



348
349
350
351
352
353
354
355
356
357
358
359
360
 
# File 'lib/switchman/shard.rb', line 348

def global_id_for(any_id, source_shard = nil)
  id = integral_id_for(any_id)
  return any_id unless id

  signed_id_operation(id) do |abs_id|
    if abs_id >= IDS_PER_SHARD
      abs_id
    else
      source_shard ||= Shard.current
      source_shard.global_id_for(abs_id)
    end
  end
end

.integral_id_for(any_id) ⇒ Object

converts an AR object, integral id, string id, or string short-global-id to a integral id. nil if it can’t be interpreted



275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
 
# File 'lib/switchman/shard.rb', line 275

def integral_id_for(any_id)
  case any_id
  when ::Arel::Nodes::Casted
    any_id = any_id.value
  when ::Arel::Nodes::BindParam
    any_id = any_id.value.value_before_type_cast
  end

  case any_id
  when ::ActiveRecord::Base
    any_id.id
  when /^(\d+)~(-?\d+)$/
    local_id = $2.to_i
    signed_id_operation(local_id) do |id|
      return nil if id > IDS_PER_SHARD

      ($1.to_i * IDS_PER_SHARD) + id
    end
  when Integer, /^-?\d+$/
    any_id.to_i
  end
end

.local_id_for(any_id) ⇒ Object 



303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
 
# File 'lib/switchman/shard.rb', line 303

def local_id_for(any_id)
  id = integral_id_for(any_id)
  return NIL_NIL_ID unless id

  return_shard = nil
  local_id = signed_id_operation(id) do |abs_id|
    if abs_id < IDS_PER_SHARD
      abs_id
    elsif (return_shard = lookup(abs_id / IDS_PER_SHARD))
      abs_id % IDS_PER_SHARD
    else
      return NIL_NIL_ID
    end
  end
  [local_id, return_shard]
end

.lookup(id) ⇒ Object

Raises:

  • (ArgumentError) 


89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
 
# File 'lib/switchman/shard.rb', line 89

def lookup(id)
  id_i = id.to_i
  return current if id_i == current.id || id == 'self'
  return default if id_i == default.id || id.nil? || id == 'default'

  id = id_i
  raise ArgumentError if id.zero?

  unless cached_shards.key?(id)
    cached_shards[id] = Shard.default.activate do
      find_cached(['shard', id]) { find_by(id: id) }
    end
  end
  cached_shards[id]
end

.partition_by_shard(array, partition_proc = nil) ⇒ Object 



227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
 
# File 'lib/switchman/shard.rb', line 227

def partition_by_shard(array, partition_proc = nil)
  shard_arrays = {}
  array.each do |object|
    partition_object = partition_proc ? partition_proc.call(object) : object
    case partition_object
    when Shard
      shard = partition_object
    when ::ActiveRecord::Base
      if partition_object.respond_to?(:associated_shards)
        partition_object.associated_shards.each do |a_shard|
          shard_arrays[a_shard] ||= []
          shard_arrays[a_shard] << object
        end
        next
      else
        shard = partition_object.shard
      end
    when Integer, /^\d+$/, /^(\d+)~(\d+)$/
      local_id, shard = Shard.local_id_for(partition_object)
      local_id ||= partition_object
      object = local_id unless partition_proc
    end
    shard ||= Shard.current
    shard_arrays[shard] ||= []
    shard_arrays[shard] << object
  end
  # TODO: use with_each_shard (or vice versa) to get
  # connection management and parallelism benefits
  shard_arrays.inject([]) do |results, (shard, objects)|
    results.concat(shard.activate { Array.wrap(yield objects) })
  end
end

.preload_cache ⇒ Object 



105
106
107
108
 
# File 'lib/switchman/shard.rb', line 105

def preload_cache
  cached_shards.reverse_merge!(active_shards.values.index_by(&:id))
  cached_shards.reverse_merge!(all.index_by(&:id))
end

.relative_id_for(any_id, source_shard, target_shard) ⇒ Object

takes an id-ish, and returns an integral id relative to target_shard. returns nil if it can’t be interpreted, or the integral version of the id if it refers to a shard that does not exist



324
325
326
327
328
329
330
331
332
333
 
# File 'lib/switchman/shard.rb', line 324

def relative_id_for(any_id, source_shard, target_shard)
  integral_id = integral_id_for(any_id)
  local_id, shard = local_id_for(integral_id)
  return integral_id unless local_id

  shard ||= source_shard
  return local_id if shard == target_shard

  shard.global_id_for(local_id)
end

.shard_for(any_id, source_shard = nil) ⇒ Object 



362
363
364
365
366
367
 
# File 'lib/switchman/shard.rb', line 362

def shard_for(any_id, source_shard = nil)
  return any_id.shard if any_id.is_a?(::ActiveRecord::Base)

  _, shard = local_id_for(any_id)
  shard || source_shard || Shard.current
end

.sharded_models ⇒ Object 



17
18
19
 
# File 'lib/switchman/shard.rb', line 17

def sharded_models
  @sharded_models ||= [::ActiveRecord::Base, UnshardedRecord].freeze
end

.short_id_for(any_id) ⇒ Object

takes an id-ish, and returns a shortened global string id if global, and itself if local. returns any_id itself if it can’t be interpreted



338
339
340
341
342
343
344
 
# File 'lib/switchman/shard.rb', line 338

def short_id_for(any_id)
  local_id, shard = local_id_for(any_id)
  return any_id unless local_id
  return local_id unless shard

  "#{shard.id}~#{local_id}"
end

.signed_id_operation(input_id) ⇒ Object

it’s tedious to hold onto this same kind of sign state and transform the result in multiple places, so here we can operate on the absolute value in a provided block and trust the sign will stay as provided. This assumes no consumer will return a nil value from the block.



267
268
269
270
271
 
# File 'lib/switchman/shard.rb', line 267

def signed_id_operation(input_id)
  sign = input_id.negative? ? -1 : 1
  output = yield input_id.abs
  output * sign
end

.with_each_shard(*args, parallel: false, exception: :raise, &block) ⇒ Object

Parameters

  • shards - an array or relation of Shards to iterate over

  • classes - an array of classes to activate

    parallel: - true/false to execute in parallel, or an integer of how many
            sub-processes. Note that parallel invocation currently uses
            forking.
exception: - :ignore, :raise, :defer (wait until the end and raise the first
            error), or a proc

Raises:

  • (ArgumentError) 


123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
 
# File 'lib/switchman/shard.rb', line 123

def with_each_shard(*args, parallel: false, exception: :raise, &block)
  raise ArgumentError, "wrong number of arguments (#{args.length} for 0...2)" if args.length > 2

  return Array.wrap(yield) unless default.is_a?(Shard)

  if args.length == 1
    if Array === args.first && args.first.first.is_a?(Class)
      classes = args.first
    else
      scope = args.first
    end
  else
    scope, classes = args
  end

  parallel = [Environment.cpu_count || 2, 2].min if parallel == true
  parallel = 0 if parallel == false || parallel.nil?

  scope ||= Shard.all
  scope = scope.order(::Arel.sql('database_server_id IS NOT NULL, database_server_id, id')) if ::ActiveRecord::Relation === scope && scope.order_values.empty?

  if parallel > 1
    if ::ActiveRecord::Relation === scope
      # still need a post-uniq, cause the default database server could be NULL or Rails.env in the db
      database_servers = scope.reorder('database_server_id').select(:database_server_id).distinct.
                         map(&:database_server).compact.uniq
      # nothing to do
      return if database_servers.count.zero?

      scopes = database_servers.to_h do |server|
        [server, scope.merge(server.shards)]
      end
    else
      scopes = scope.group_by(&:database_server)
    end

    # clear connections prior to forking (no more queries will be executed in the parent,
    # and we want them gone so that we don't accidentally use them post-fork doing something
    # silly like dealloc'ing prepared statements)
    ::ActiveRecord::Base.clear_all_connections!

    parent_process_name = `ps -ocommand= -p#{Process.pid}`.slice(/#{$0}.*/)
    ret = ::Parallel.map(scopes, in_processes: scopes.length > 1 ? parallel : 0) do |server, subscope|
      name = server.id
      # rubocop:disable Style/GlobalStdStream
      $stdout = Parallel::PrefixingIO.new(name, STDOUT)
      $stderr = Parallel::PrefixingIO.new(name, STDERR)
      # rubocop:enable Style/GlobalStdStream
      begin
        max_length = 128 - name.length - 3
        short_parent_name = parent_process_name[0..max_length] if max_length >= 0
        new_title = [short_parent_name, name].join(' ')
        Process.setproctitle(new_title)
        Switchman.config[:on_fork_proc]&.call
        with_each_shard(subscope, classes, exception: exception, &block).map { |result| Parallel::ResultWrapper.new(result) }
      rescue => e
        logger.error e.full_message
        Parallel::QuietExceptionWrapper.new(name, ::Parallel::ExceptionWrapper.new(e))
      end
    end.flatten

    errors = ret.select { |val| val.is_a?(Parallel::QuietExceptionWrapper) }
    unless errors.empty?
      raise errors.first.exception if errors.length == 1

      raise Errors::ParallelShardExecError,
            "The following database server(s) did not finish processing cleanly: #{errors.map(&:name).sort.join(', ')}",
            cause: errors.first.exception
    end

    return ret.map(&:result)
  end

  classes ||= []

  previous_shard = nil
  result = []
  ex = nil
  scope.each do |shard|
    # shard references a database server that isn't configured in this environment
    next unless shard.database_server

    shard.activate(*classes) do
      result.concat Array.wrap(yield)
    rescue
      case exception
      when :ignore
        # ignore
      when :defer
        ex ||= $!
      when Proc
        exception.call
      # when :raise
      else
        raise
      end
    end
    previous_shard = shard
  end
  raise ex if ex

  result
end

Instance Method Details

#_dump(_depth) ⇒ Object

custom serialization, since shard is self-referential



485
486
487
 
# File 'lib/switchman/shard.rb', line 485

def _dump(_depth)
  id.to_s
end

#activate(*classes, &block) ⇒ Object 



472
473
474
475
 
# File 'lib/switchman/shard.rb', line 472

def activate(*classes, &block)
  shards = hashify_classes(classes)
  Shard.activate(shards, &block)
end

#activate!(*classes) ⇒ Object

for use from console ONLY



478
479
480
481
482
 
# File 'lib/switchman/shard.rb', line 478

def activate!(*classes)
  shards = hashify_classes(classes)
  Shard.activate!(shards)
  nil
end

#database_server ⇒ Object 



446
447
448
 
# File 'lib/switchman/shard.rb', line 446

def database_server
  @database_server ||= DatabaseServer.find(database_server_id)
end

#database_server=(database_server) ⇒ Object 



450
451
452
453
 
# File 'lib/switchman/shard.rb', line 450

def database_server=(database_server)
  self.database_server_id = database_server.id
  @database_server = database_server
end

#description ⇒ Object 



459
460
461
 
# File 'lib/switchman/shard.rb', line 459

def description
  [database_server.id, name].compact.join(':')
end

#destroy ⇒ Object 



553
554
555
556
557
 
# File 'lib/switchman/shard.rb', line 553

def destroy
  raise('Cannot destroy the default shard') if default?

  super
end

#drop_database ⇒ Object 



493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
 
# File 'lib/switchman/shard.rb', line 493

def drop_database
  raise('Cannot drop the database of the default shard') if default?
  return unless read_attribute(:name)

  begin
    adapter = database_server.config[:adapter]
    sharding_config = Switchman.config || {}
    drop_statement = sharding_config[adapter]&.[](:drop_statement)
    drop_statement ||= sharding_config[:drop_statement]
    if drop_statement
      drop_statement = Array(drop_statement).dup.
                       map { |statement| statement.gsub('%{name}', name) }
    end

    case adapter
    when 'mysql', 'mysql2'
      activate do
        ::GuardRail.activate(:deploy) do
          drop_statement ||= "DROP DATABASE #{name}"
          Array(drop_statement).each do |stmt|
            ::ActiveRecord::Base.connection.execute(stmt)
          end
        end
      end
    when 'postgresql'
      activate do
        ::GuardRail.activate(:deploy) do
          # Shut up, Postgres!
          conn = ::ActiveRecord::Base.connection
          old_proc = conn.raw_connection.set_notice_processor {}
          begin
            drop_statement ||= "DROP SCHEMA #{name} CASCADE"
            Array(drop_statement).each do |stmt|
              ::ActiveRecord::Base.connection.execute(stmt)
            end
          ensure
            conn.raw_connection.set_notice_processor(&old_proc) if old_proc
          end
        end
      end
    end
  rescue
    logger.info "Drop failed: #{$!}"
  end
end

#global_id_for(local_id) ⇒ Object

takes an id local to this shard, and returns a global id



540
541
542
543
544
545
546
 
# File 'lib/switchman/shard.rb', line 540

def global_id_for(local_id)
  return nil unless local_id

  self.class.signed_id_operation(local_id) do |abs_id|
    abs_id + (id * IDS_PER_SHARD)
  end
end

#hash ⇒ Object

skip global_id.hash



549
550
551
 
# File 'lib/switchman/shard.rb', line 549

def hash
  id.hash
end

#name ⇒ Object 



432
433
434
435
436
437
438
439
 
# File 'lib/switchman/shard.rb', line 432

def name
  unless instance_variable_defined?(:@name)
    # protect against re-entrancy
    @name = nil
    @name = read_attribute(:name) || default_name
  end
  @name
end

#name=(name) ⇒ Object 



441
442
443
444
 
# File 'lib/switchman/shard.rb', line 441

def name=(name)
  write_attribute(:name, @name = name)
  remove_instance_variable(:@name) if name.nil?
end

#original_id ⇒ Object 



468
469
470
 
# File 'lib/switchman/shard.rb', line 468

def original_id
  id
end

#primary? ⇒ Boolean

Returns:

  • (Boolean) 


455
456
457
 
# File 'lib/switchman/shard.rb', line 455

def primary?
  self == database_server.primary_shard
end

#shard ⇒ Object

Shards are always on the default shard



464
465
466
 
# File 'lib/switchman/shard.rb', line 464

def shard
  Shard.default
end