Class: Purplelight::Snapshot

Inherits:
Object
  • Object
show all
Defined in:
lib/purplelight/snapshot.rb

Overview

Snapshot orchestrates partition planning, parallel reads, and writing.

Constant Summary collapse

DEFAULTS = 
{
  format: :jsonl,
  compression: :zstd,
  batch_size: 2_000,
  partitions: [Etc.respond_to?(:nprocessors) ? [Etc.nprocessors * 2, 4].max : 4, 32].min,
  queue_size_bytes: 256 * 1024 * 1024,
  rotate_bytes: 256 * 1024 * 1024,
  read_concern: { level: :majority },
  read_preference: :primary,
  no_cursor_timeout: true
}.freeze

Class Method Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(client:, collection:, output:, format: , compression: , partitions: , batch_size: , queue_size_bytes: , rotate_bytes: , query: {}, projection: nil, hint: nil, mapper: nil, resume: { enabled: true, state_path: nil, overwrite_incompatible: false }, sharding: { mode: :by_size, part_bytes: DEFAULTS[:rotate_bytes], prefix: nil }, logger: nil, on_progress: nil, read_concern: , read_preference: , no_cursor_timeout: , telemetry: nil, compression_level: nil, writer_threads: 1, write_chunk_bytes: nil, parquet_row_group: nil, parquet_max_rows: nil) ⇒ Snapshot

Returns a new instance of Snapshot.



34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
 
# File 'lib/purplelight/snapshot.rb', line 34

def initialize(client:, collection:, output:, format: DEFAULTS[:format], compression: DEFAULTS[:compression],
               partitions: DEFAULTS[:partitions], batch_size: DEFAULTS[:batch_size],
               queue_size_bytes: DEFAULTS[:queue_size_bytes], rotate_bytes: DEFAULTS[:rotate_bytes],
               query: {}, projection: nil, hint: nil, mapper: nil,
               resume: { enabled: true, state_path: nil, overwrite_incompatible: false },
               sharding: { mode: :by_size, part_bytes: DEFAULTS[:rotate_bytes], prefix: nil },
               logger: nil, on_progress: nil, read_concern: DEFAULTS[:read_concern], read_preference: DEFAULTS[:read_preference],
               no_cursor_timeout: DEFAULTS[:no_cursor_timeout], telemetry: nil,
               compression_level: nil, writer_threads: 1, write_chunk_bytes: nil, parquet_row_group: nil,
               parquet_max_rows: nil)
  @client = client
  @collection = client[collection]
  @output = output
  @format = format.to_sym
  @compression = compression.to_sym
  @partitions = partitions
  @batch_size = batch_size
  @queue_size_bytes = queue_size_bytes
  @rotate_bytes = rotate_bytes
  @query = query || {}
  @projection = projection
  @hint = hint
  @mapper = mapper
  @resume = resume || { enabled: true }
  @sharding = sharding || { mode: :by_size }
  @logger = logger
  @on_progress = on_progress
  @read_concern = read_concern
  @read_preference = read_preference
  @no_cursor_timeout = no_cursor_timeout
  @compression_level = compression_level
  @writer_threads = writer_threads || 1
  @write_chunk_bytes = write_chunk_bytes
  @parquet_row_group = parquet_row_group
  @parquet_max_rows = parquet_max_rows

  @running = true
  @telemetry_enabled = telemetry ? telemetry.enabled? : (ENV['PL_TELEMETRY'] == '1')
  @telemetry = telemetry || (
    @telemetry_enabled ? Telemetry.new(enabled: true) : Telemetry::NULL
  )
end

Class Method Details

.snapshot ⇒ Object 



30
31
32
 
# File 'lib/purplelight/snapshot.rb', line 30

def self.snapshot(...)
  new(...).run
end

Instance Method Details

#run ⇒ Object

rubocop:disable Naming/PredicateMethod



78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
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
 
# File 'lib/purplelight/snapshot.rb', line 78

def run
  dir, prefix = resolve_output(@output, @format)
  manifest_path = File.join(dir, "#{prefix}.manifest.json")
  query_digest = Manifest.query_digest(@query, @projection)

  manifest = if @resume && @resume[:enabled] && File.exist?(manifest_path)
               m = Manifest.load(manifest_path)
               unless m.compatible_with?(collection: @collection.name, format: @format, compression: @compression,
                                         query_digest: query_digest)
                 if @resume[:overwrite_incompatible]
                   m = Manifest.new(path: manifest_path)
                 else
                   raise IncompatibleResumeError,
                         'existing manifest incompatible with request; pass overwrite_incompatible: true to reset'
                 end
               end
               m
             else
               Manifest.new(path: manifest_path)
             end

  manifest.configure!(collection: @collection.name, format: @format, compression: @compression, query_digest: query_digest, options: {
                        partitions: @partitions,
                        batch_size: @batch_size,
                        queue_size_bytes: @queue_size_bytes,
                        rotate_bytes: @rotate_bytes,
                        hint: @hint,
                        read_concern: (@read_concern.is_a?(Hash) ? @read_concern : { level: @read_concern }),
                        no_cursor_timeout: @no_cursor_timeout,
                        writer_threads: @writer_threads,
                        compression_level: @compression_level || (ENV['PL_ZSTD_LEVEL']&.to_i if @compression.to_s == 'zstd') || ENV['PL_ZSTD_LEVEL']&.to_i,
                        write_chunk_bytes: @write_chunk_bytes || ENV['PL_WRITE_CHUNK_BYTES']&.to_i,
                        parquet_row_group: @parquet_row_group || ENV['PL_PARQUET_ROW_GROUP']&.to_i,
                        parquet_max_rows: @parquet_max_rows,
                        sharding: @sharding,
                        resume_overwrite_incompatible: @resume && @resume[:overwrite_incompatible] ? true : false,
                        telemetry: @telemetry_enabled
                      })
  manifest.ensure_partitions!(@partitions)

  # Plan partitions
  t_plan = @telemetry.start(:partition_plan_time)
  partition_filters = Partitioner.object_id_partitions(collection: @collection, query: @query,
                                                       partitions: @partitions, telemetry: @telemetry)
  @telemetry.finish(:partition_plan_time, t_plan)

  # Reader queue
  queue = ByteQueue.new(max_bytes: @queue_size_bytes)

  # Writer
  writer = case @format
           when :jsonl
             WriterJSONL.new(directory: dir, prefix: prefix, compression: @compression,
                             rotate_bytes: @rotate_bytes, logger: @logger, manifest: manifest)
           when :csv
             single_file = @sharding && @sharding[:mode].to_s == 'single_file'
             WriterCSV.new(directory: dir, prefix: prefix, compression: @compression, rotate_bytes: @rotate_bytes,
                           logger: @logger, manifest: manifest, single_file: single_file)
           when :parquet
             single_file = @sharding && @sharding[:mode].to_s == 'single_file'
             row_group = @parquet_row_group || ENV['PL_PARQUET_ROW_GROUP']&.to_i || WriterParquet::DEFAULT_ROW_GROUP_SIZE
             WriterParquet.new(directory: dir, prefix: prefix, compression: @compression, logger: @logger,
                               manifest: manifest, single_file: single_file, row_group_size: row_group,
                               rotate_rows: @parquet_max_rows)
           else
             raise ArgumentError, "format not implemented: #{@format}"
           end

  # Start reader threads
  readers = partition_filters.each_with_index.map do |pf, idx|
    Thread.new do
      local_telemetry = @telemetry_enabled ? Telemetry.new(enabled: true) : Telemetry::NULL
      read_partition(idx: idx, filter_spec: pf, queue: queue, batch_size: @batch_size, manifest: manifest, telemetry: local_telemetry)
      # Merge per-thread telemetry
      @telemetry.merge!(local_telemetry) if @telemetry_enabled
    end
  end

  # Writer loop
  writer_telemetry = @telemetry_enabled ? Telemetry.new(enabled: true) : Telemetry::NULL
  writer_thread = Thread.new do
    Thread.current[:pl_telemetry] = writer_telemetry if @telemetry_enabled
    loop do
      batch = queue.pop
      break if batch.nil?

      writer.write_many(batch)
    end
  ensure
    writer.close
  end

  progress_thread = Thread.new do
    Time.now
    loop do
      sleep 2
      break unless @running

      @on_progress&.call({ queue_bytes: queue.size_bytes })
    end
  end

  # Join readers
  readers.each(&:join)
  queue.close
  writer_thread.join
  @telemetry.merge!(writer_telemetry) if @telemetry_enabled
  @running = false
  progress_thread.join
  if @telemetry_enabled
    total = @telemetry.timers.values.sum
    breakdown = @telemetry.timers
                          .sort_by { |_k, v| -v }
                          .map { |k, v| [k, v, total.zero? ? 0 : ((v / total) * 100.0)] }
    if @logger
      @logger.info('Telemetry (seconds and % of timed work):')
      breakdown.each { |k, v, pct| @logger.info("  #{k}: #{v.round(3)}s (#{pct.round(1)}%)") }
    else
      puts 'Telemetry (seconds and % of timed work):'
      breakdown.each { |k, v, pct| puts "  #{k}: #{v.round(3)}s (#{pct.round(1)}%)" }
    end
  end
  true
end