Possible bug/issue with AWS Aurora - sending too many queries to a single reader

[allexivan]

I am using latest proxy sql docker image in AWS ECS cluster.

I have an AWS Aurora cluster in autoscaling and I noticed that whenever there is a scale event where more rds readers are added, once the new reader comes only, it send most of the queries to that reader and overloads it, while the other readers have very few queries.
This happens up about when RDS cluster logs say "Cluster topology is updated." when it goes back to normal.

My Proxysql config:

mysql_variables=
{
	threads=10
	max_connections=2048
    	connect_timeout_client=90000
    	connect_timeout_server_max=90000
	default_query_delay=0
	default_query_timeout=36000000
	have_compress=true
	poll_timeout=2000
	interfaces="0.0.0.0:6033"
	default_schema="information_schema"
	stacksize=1048576
	server_version="8.0.34"
	connect_timeout_server=3000
	monitor_enabled=true
	monitor_username=""
	monitor_password=""
	monitor_history=600000
	monitor_connect_interval=60000
	monitor_ping_interval=10000
	monitor_read_only_interval=1500
	monitor_read_only_timeout=500
	ping_interval_server_msec=120000
	ping_timeout_server=500
	commands_stats=true
	sessions_sort=true
	connect_retries_on_failure=10
    	log_unhealthy_connections=false
}


mysql_users=
(
	{
        username=""
        password=""
        default_hostgroup=10
        frontend=1
        backend=0
	},

	{
        username=""
        password=""
        default_hostgroup=10
        frontend=0
        backend=1
	}
)

mysql_aws_aurora_hostgroups=
(
    {
        writer_hostgroup=10
        reader_hostgroup=20
        active=1
        aurora_port=3306
        domain_name=""
        max_lag_ms=600000
        check_interval_ms=1200
        check_timeout_ms=1100
        writer_is_also_reader=0
        new_reader_weight=1000
        add_lag_ms=30
        min_lag_ms=30
        lag_num_checks=1
        comment="prod-ar"
    }
)

mysql_servers =
(
	{
        address=""
        port=3306
        hostgroup_id=10
        status="ONLINE" 
        weight=1000
        max_connections=2000
        compression=0
        max_replication_lag=10
        use_ssl=0
        max_latency_ms=0
        comment="master server"
	}
)

[JavierJF]

Hi @allexivan,

issue description lack the minimal requested by the issue template, and aside that minimal information, for verifying a statement like this, it send most of the queries to that reader and overloads it, while the other readers have very few queries., metrics are required, if you could share the metrics used for that analysis, we could make more sense of the scenario.

Regards, Javier.