[pgpool-general: 7687] Pgbench test: direct connection vs. pgpool connect

[Luca Maranzano]

Hi all,

we setup a 3 nodes cluster with PGPool-II version 4.2.3 on 3 physical
servers with CentOS 7 and PostgreSQL 12.7, like described here:

https://www.pgpool.net/docs/latest/en/html/example-cluster.html

Cluster status:
[postgres at server14 ~]$ psql -h PGPOOL-VIP -p 9999 -U pgpool postgres -c
"show pool_nodes"
 node_id |    hostname    | port | status | lb_weight |  role   |
select_cnt | load_balance_node | replication_delay | replication_state |
replication_sync_state | last_status_change
---------+----------------+------+--------+-----------+---------+------------+-------------------+-------------------+-------------------+------------------------+---------------------
 0       | server10       | 5432 | up     | 0.333333  | standby | 1
 | false             | 0                 | streaming         | async
           | 2021-09-07 17:01:29
 1       | server14       | 5432 | up     | 0.333333  | primary | 55832
 | false             | 0                 |                   |
           | 2021-09-07 16:43:37
 2       | server15       | 5432 | up     | 0.333333  | standby | 1
 | true              | 0                 | streaming         | async
           | 2021-09-07 16:43:37
(3 rows)


Performing some stress test with pgbench we noted a great difference
connecting directly to the Priamry DB versus connecting to the PGPool VIP:

Direct Connection on the Primary PostgreSQL:
[postgres at server14 ~]$ pgbench -h server14 -p5432  test -T60
Password:
starting vacuum...end.
transaction type: <builtin: TPC-B (sort of)>
scaling factor: 1
query mode: simple
number of clients: 1
number of threads: 1
duration: 60 s
number of transactions actually processed: 54441
*latency average = 1.102 ms*
tps = 907.336709 (including connections establishing)
tps = 907.584536 (excluding connections establishing)

Connection to the PGPool VIP:
[postgres at server14 ~]$ pgbench -h PGPOOL-VIP -p9999  test -T60
Password:
starting vacuum...end.
transaction type: <builtin: TPC-B (sort of)>
scaling factor: 1
query mode: simple
number of clients: 1
number of threads: 1
duration: 60 s
number of transactions actually processed: 21244
*latency average = 2.824 ms*
tps = 354.065350 (including connections establishing)
tps = 354.562738 (excluding connections establishing)

Yes, it's a simple test, but the difference is *huge* IMHO. Repeating
several times the same test get same results.

I'd like to know your experience and comments and what could be the problem.
Is there any specific tuning to do in PGPool-II ?

Thanks in advance.
Regards,
Luca
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.pgpool.net/pipermail/pgpool-general/attachments/20210908/79e4f3ab/attachment.htm>