Module: Gitlab::Database::Reindexing

Defined in:
lib/gitlab/database/reindexing.rb,
lib/gitlab/database/reindexing/coordinator.rb,
lib/gitlab/database/reindexing/queued_action.rb,
lib/gitlab/database/reindexing/reindex_action.rb,
lib/gitlab/database/reindexing/index_selection.rb,
lib/gitlab/database/reindexing/grafana_notifier.rb,
lib/gitlab/database/reindexing/reindex_concurrently.rb

Defined Under Namespace

Classes: Coordinator, GrafanaNotifier, IndexSelection, QueuedAction, ReindexAction, ReindexConcurrently

Constant Summary collapse

DEFAULT_INDEXES_PER_INVOCATION =

Number of indexes to reindex per invocation

2
SUPPORTED_TYPES = 
%w[btree gist].freeze
REMOVE_INDEX_RETRY_CONFIG =

When dropping an index, we acquire a SHARE UPDATE EXCLUSIVE lock, which only conflicts with DDL and vacuum. We therefore execute this with a rather high lock timeout and a long pause in between retries. This is an alternative to setting a high statement timeout, which would lead to a long running query with effects on e.g. vacuum.

[[1.minute, 9.minutes]] * 30

Class Method Summary collapse

Class Method Details

.automatic_reindexing(maximum_records: DEFAULT_INDEXES_PER_INVOCATION) ⇒ Object

Performs automatic reindexing for a limited number of indexes per call

  1. Consume from the explicit reindexing queue
  2. Apply bloat heuristic to find most bloated indexes and reindex those


51
52
53
54
55
56
57
58
59
60
61
62
 
# File 'lib/gitlab/database/reindexing.rb', line 51

def self.automatic_reindexing(maximum_records: DEFAULT_INDEXES_PER_INVOCATION)
  # Cleanup leftover temporary indexes from previous, possibly aborted runs (if any)
  cleanup_leftovers!

  # Consume from the explicit reindexing queue first
  done_counter = perform_from_queue(maximum_records: maximum_records)

  return if done_counter >= maximum_records

  # Execute reindexing based on bloat heuristic
  perform_with_heuristic(maximum_records: maximum_records - done_counter)
end

.cleanup_leftovers! ⇒ Object 



87
88
89
90
91
 
# File 'lib/gitlab/database/reindexing.rb', line 87

def self.cleanup_leftovers!
  PostgresIndex.reindexing_leftovers.each do |index|
    Coordinator.new(index).drop
  end
end

.enabled? ⇒ Boolean

Returns:

  • (Boolean) 


18
19
20
21
22
 
# File 'lib/gitlab/database/reindexing.rb', line 18

def self.enabled?
  return false if Feature.enabled?(:disallow_database_ddl_feature_flags, type: :ops)

  Feature.enabled?(:database_reindexing, type: :ops)
end

.invoke(database = nil) ⇒ Object 



24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
 
# File 'lib/gitlab/database/reindexing.rb', line 24

def self.invoke(database = nil)
  Gitlab::Database::EachDatabase.each_connection do |connection, connection_name|
    next if database && database.to_s != connection_name.to_s

    Gitlab::Database::SharedModel.logger = Logger.new($stdout) if Gitlab::Utils.to_boolean(ENV['LOG_QUERIES_TO_CONSOLE'], default: false)

    # Hack: Before we do actual reindexing work, create async indexes
    if Feature.disabled?(:disallow_database_ddl_feature_flags, type: :ops) && Feature.enabled?(:database_async_index_creation, type: :ops)
      Gitlab::Database::AsyncIndexes.create_pending_indexes!
    end

    Gitlab::Database::AsyncIndexes.drop_pending_indexes!

    if Feature.disabled?(:disallow_database_ddl_feature_flags, type: :ops) && Feature.enabled?(:database_async_foreign_key_validation, type: :ops)
      Gitlab::Database::AsyncConstraints.validate_pending_entries!
    end

    automatic_reindexing
  end
rescue StandardError => e
  Gitlab::AppLogger.error(e)
  raise
end

.minimum_index_size!(bytes) ⇒ Object 



93
94
95
 
# File 'lib/gitlab/database/reindexing.rb', line 93

def self.minimum_index_size!(bytes)
  update_reindexing_setting!(:reindexing_minimum_index_size, bytes)
end

.minimum_relative_bloat_size!(threshold) ⇒ Object 



97
98
99
 
# File 'lib/gitlab/database/reindexing.rb', line 97

def self.minimum_relative_bloat_size!(threshold)
  update_reindexing_setting!(:reindexing_minimum_relative_bloat_size, threshold)
end

.perform_from_queue(maximum_records: DEFAULT_INDEXES_PER_INVOCATION) ⇒ Object

Reindex indexes that have been explicitly enqueued (for a limited number of indexes per call)



75
76
77
78
79
80
81
82
83
84
85
 
# File 'lib/gitlab/database/reindexing.rb', line 75

def self.perform_from_queue(maximum_records: DEFAULT_INDEXES_PER_INVOCATION)
  QueuedAction.in_queue_order.limit(maximum_records).each do |queued_entry|
    Coordinator.new(queued_entry.index).perform

    queued_entry.done!
  rescue StandardError => e
    queued_entry.failed!

    Gitlab::AppLogger.error("Failed to perform reindexing action on queued entry #{queued_entry}: #{e}")
  end.size
end

.perform_with_heuristic(candidate_indexes = Gitlab::Database::PostgresIndex.reindexing_support, maximum_records: DEFAULT_INDEXES_PER_INVOCATION) ⇒ Object

Reindex based on bloat heuristic for a limited number of indexes per call

We use a bloat heuristic to estimate the index bloat and pick the most bloated indexes for reindexing.



68
69
70
71
72
 
# File 'lib/gitlab/database/reindexing.rb', line 68

def self.perform_with_heuristic(candidate_indexes = Gitlab::Database::PostgresIndex.reindexing_support, maximum_records: DEFAULT_INDEXES_PER_INVOCATION)
  IndexSelection.new(candidate_indexes).take(maximum_records).each do |index|
    Coordinator.new(index).perform
  end
end