[pgpool-general: 2890] pgpool-3.1.7 causes high load when DB on different machine is experiencing high load

[Vincent Lasmarias]

Thanks in advance for your answer to the following issue.

I have the following setup:

- pgpool-3.1.7 running on 2 machines
- connected to a master-slave Postgres setup that are running on separate
machines
- the pgpool client is an Apache Php process running on the same machine as
the pgpool processes

When either the master or slave experiences high load, the pgpool returns
the following ERRORs and also causes extremely high load (via 'uptime' on
the servers they are running on):

2014-05-29 17:08:11 ERROR: pid 31018: read_startup_packet: incorrect packet
length (438329760)
2014-05-29 17:08:20 ERROR: pid 31840: read_startup_packet: incorrect packet
length (438269056)
2014-05-29 17:09:37 ERROR: pid 32730: read_startup_packet: incorrect packet
length (438308800)
2014-05-29 17:09:37 ERROR: pid 31862: pool_read: read failed (Connection
reset by peer)
2014-05-29 17:09:37 ERROR: pid 373: pool_read: read failed (Connection reset
by peer)

Why would the high load on either the master or slave Postgres server cause
the pgpool running to also get into high load?

Our pertinent pgpool configuration is as follows:

# - Backend Connection Settings -

backend_weight0 = 1
backend_data_directory0 = '/usr/local/pgsql/data'
backend_flag0 = 'ALLOW_TO_FAILOVER'

backend_weight1 = 8
backend_data_directory1 = '/usr/local/pgsql/data'
backend_flag1 = 'ALLOW_TO_FAILOVER'

# - Authentication -

enable_pool_hba = off
authentication_timeout = 60

#---------------------------------------------------------------------------
---
# POOLS
#---------------------------------------------------------------------------
---

num_init_children = 50
max_pool = 4
child_life_time = 300
connection_life_time = 0
child_max_connections = 1000
client_idle_limit = 900

#---------------------------------------------------------------------------
---
# CONNECTION POOLING
#---------------------------------------------------------------------------
---

connection_cache = on
reset_query_list = 'ABORT; DISCARD ALL'

#---------------------------------------------------------------------------
---
# REPLICATION MODE
#---------------------------------------------------------------------------
---

replication_mode = off
replicate_select = off
insert_lock = off
lobj_lock_table = ''
replication_stop_on_mismatch = off
failover_if_affected_tuples_mismatch = off

# - Degenerate handling -

replication_stop_on_mismatch = off
failover_if_affected_tuples_mismatch = off

#---------------------------------------------------------------------------
---
# LOAD BALANCING MODE
#---------------------------------------------------------------------------
---

load_balance_mode = on
ignore_leading_white_space = on
white_function_list = ''
black_function_list =
'never_registered_before_eod_sim_list,unregistered_beacon_eod_sim,currval,la
stval,nextval,setval,add_.*,assign_.*,audit_.*,beacon_.*,change_.*,clear_.*,
consolidate_.*,delete_.*,del_.*,disable_.*,enable_.*,increment_.*,insert_.*,
move_.*,quick_.*,reactivate_.*,rebuild_.*,remove_.*,repopulate.*,retire_.*,s
etup_.*,toggle_.*,unassign_.*,update.*,upd_.*,spw_.*'

#---------------------------------------------------------------------------
---
# MASTER/SLAVE MODE
#---------------------------------------------------------------------------
---

master_slave_mode = on
master_slave_sub_mode = 'stream'
# - Streaming -

sr_check_period = 2
# delay_threshold is in bytes, set here to ~ 10MB
delay_threshold = 100
log_standby_delay = 'if_over_threshold'

#---------------------------------------------------------------------------
---
# HEALTH CHECK
#---------------------------------------------------------------------------
---

health_check_timeout = 20
health_check_period = 5

#---------------------------------------------------------------------------
---
# FAILOVER AND FAILBACK
#---------------------------------------------------------------------------
---

failover_command = ''
failback_command = ''
fail_over_on_backend_error = on

#---------------------------------------------------------------------------
---
# PARALLEL MODE AND QUERY CACHE
#---------------------------------------------------------------------------
---

parallel_mode = false
enable_query_cache = false
pgpool2_hostname = ''
system_db_dbname = 'pgpool'

#---------------------------------------------------------------------------
---
# ONLINE RECOVERY
#---------------------------------------------------------------------------
---
recovery_1st_stage_command = ''
recovery_2nd_stage_command = ''
recovery_timeout = 90
client_idle_limit_in_recovery = 0


#---------------------------------------------------------------------------
---
# OTHERS
#---------------------------------------------------------------------------
---

ssl = off 
relcache_expire = 3600