debian-mirror-gitlab/app/models/postgresql/replication_slot.rb

94 lines
3 KiB
Ruby
Raw Normal View History

2018-11-18 11:00:15 +05:30
# frozen_string_literal: true
module Postgresql
2022-01-26 12:08:38 +05:30
class ReplicationSlot < Gitlab::Database::SharedModel
2018-11-18 11:00:15 +05:30
self.table_name = 'pg_replication_slots'
2018-12-13 13:39:08 +05:30
# Returns true if there are any replication slots in use.
# PostgreSQL-compatible databases such as Aurora don't support
# replication slots, so this will return false as well.
def self.in_use?
transaction { exists? }
rescue ActiveRecord::StatementInvalid
false
end
2018-11-18 11:00:15 +05:30
# Returns true if the lag observed across all replication slots exceeds a
# given threshold.
#
# max - The maximum replication lag size, in bytes. Based on GitLab.com
# statistics it takes between 1 and 5 seconds to replicate around
# 100 MB of data.
def self.lag_too_great?(max = 100.megabytes)
2018-12-13 13:39:08 +05:30
return false unless in_use?
2021-01-03 14:25:43 +05:30
lag_function = "pg_wal_lsn_diff" \
"(pg_current_wal_insert_lsn(), restart_lsn)::bigint"
2018-11-18 11:00:15 +05:30
# We force the use of a transaction here so the query always goes to the
2021-09-04 01:27:46 +05:30
# primary, even when using the DB load balancer.
sizes = transaction { pluck(Arel.sql(lag_function)) }
2019-09-30 21:07:59 +05:30
too_great = sizes.compact.count { |size| size >= max }
2018-11-18 11:00:15 +05:30
# If too many replicas are falling behind too much, the availability of a
# GitLab instance might suffer. To prevent this from happening we require
# at least 1 replica to have data recent enough.
2020-10-24 23:57:45 +05:30
if sizes.any? && too_great > 0
2018-11-18 11:00:15 +05:30
(sizes.length - too_great) <= 1
else
false
end
end
2021-10-27 15:23:28 +05:30
def self.count
connection
.execute("SELECT COUNT(*) FROM pg_replication_slots;")
.first
.fetch('count')
.to_i
end
def self.unused_slots_count
connection
.execute("SELECT COUNT(*) FROM pg_replication_slots WHERE active = 'f';")
.first
.fetch('count')
.to_i
end
def self.used_slots_count
connection
.execute("SELECT COUNT(*) FROM pg_replication_slots WHERE active = 't';")
.first
.fetch('count')
.to_i
end
# array of slots and the retained_bytes
# https://www.skillslogic.com/blog/databases/checking-postgres-replication-lag
# http://bdr-project.org/docs/stable/monitoring-peers.html
def self.slots_retained_bytes
connection.execute(<<-SQL.squish).to_a
SELECT slot_name, database,
active, pg_wal_lsn_diff(pg_current_wal_insert_lsn(), restart_lsn)
AS retained_bytes
FROM pg_replication_slots;
SQL
end
# returns the max number WAL space (in bytes) being used across the replication slots
def self.max_retained_wal
connection.execute(<<-SQL.squish).first.fetch('coalesce').to_i
SELECT COALESCE(MAX(pg_wal_lsn_diff(pg_current_wal_insert_lsn(), restart_lsn)), 0)
FROM pg_replication_slots;
SQL
end
def self.max_replication_slots
connection.execute(<<-SQL.squish).first&.fetch('setting').to_i
SELECT setting FROM pg_settings WHERE name = 'max_replication_slots';
SQL
end
2018-11-18 11:00:15 +05:30
end
end