why this version's top event also XidInMVCCSnapshot

[digoal]

HI,
i had use this version test tpc-b.
but when i test 2000 connections, the hot spot the same as postgresql community version.

Samples: 18M of event 'cpu-clock', Event count (approx.): 609642457155                                                                                                                                                                       
Overhead  Shared Object         Symbol                                                                                                                                                                                                       
  59.00%  postgres              [.] XidInMVCCSnapshot                                                                                                                                                                                        
   4.52%  postgres              [.] TransactionIdIsInProgress  

this is test demo:

listen_addresses = '0.0.0.0'
port = 4001
max_connections = 21000
superuser_reserved_connections = 13
unix_socket_directories = '/tmp,.'
shared_buffers = 32GB
dynamic_shared_memory_type = posix
vacuum_cost_delay = 0
bgwriter_delay = 10ms
bgwriter_lru_maxpages = 900
bgwriter_lru_multiplier = 5.0
effective_io_concurrency = 0
wal_level = minimal 
synchronous_commit = off
full_page_writes = off
wal_buffers = 128MB
wal_writer_delay = 10ms
wal_writer_flush_after = 1MB
checkpoint_timeout = 25min
max_wal_size = 64GB
min_wal_size = 16GB
checkpoint_completion_target = 0.1
max_wal_senders = 0
seq_page_cost = 1.0
random_page_cost = 1.1
log_destination = 'csvlog'
logging_collector = on
log_truncate_on_rotation = on
log_checkpoints = on
log_connections = on
log_disconnections = on
log_error_verbosity = verbose  
log_timezone = 'PRC'
autovacuum = on
log_autovacuum_min_duration = 0
autovacuum_max_workers = 3
datestyle = 'iso, mdy'
timezone = 'PRC'
lc_messages = 'C'
lc_monetary = 'C'
lc_numeric = 'C'
lc_time = 'C'
default_text_search_config = 'pg_catalog.english'

first patch pgbench

https://commitfest.postgresql.org/18/1388/

and then

pgbench -i -s 20 --unlogged-tables

pgbench -M prepared -n -r -P 1 -c 2000 -j 1000

pgbench -M prepared -n -r -P 1 -c 20000 -j 1000

best regards,
digoal

[knizhnik]

Did you set session_pool_size? In the latest version by default it is set to 10. It means that there will be only 10 worker backends. Can you check number of backends spawned by postgres?

[digoal]

HI, my test host has 56 cores, what is a comfortable pool size?

[digoal]

it's my mistake, sorry.