Reputation: 391
I have a Rails application with a Postgres database under AWS RDS with multi-az architecture. The HA architecture used by RDS is master/slave and they provide the service with a single endpoint that points to the current master.
Whenever there's a database failover, Active Record will continue to try to connect to the same server, instead of retrying the connection in order to pick up the new IP for the master.
Is there a way to create a "global" rescue for the ActiveRecord::StatementInvalid: PG::ConnectionBad: PQsocket() can't get socket descriptor
error that simply runs ActiveRecord::Base.connection_pool.disconnect!
that will make the next query to work?
Upvotes: 10
Views: 2831
Reputation: 391
I was able to make Active Record reconnect after the failover event by applying a monkey patch to postgres_adapter
.
lib/core_ext/active_record/postgresql_adapter.rb
:
require 'active_record/connection_adapters/postgresql_adapter'
class ActiveRecord::ConnectionAdapters::PostgreSQLAdapter
private
def exec_no_cache(sql, name, binds)
log(sql, name, binds) { @connection.async_exec(sql, []) }
rescue ActiveRecord::StatementInvalid => e
if e.to_s.include?('PG::ConnectionBad')
ActiveRecord::Base.connection_pool.disconnect!
end
raise e
end
end
Upvotes: 6