View Issue Details

Jump to NotesJump to History
IDProjectCategoryView StatusDate SubmittedLast Update
0000556Pgpool-IIBugpublic2019-11-08 15:422019-12-13 07:19
Reportergsl23 Assigned Tot-ishii  
PrioritynormalSeverityminorReproducibilityalways
Status closedResolutionopen 
OS VersionRHEL 7.5 
Product Version3.7.3 
Summary0000556: Leaking sessions on pg backend.
DescriptionWe have leaking session on database. My suggestion is that pgpool not closing session on db correctly.
 
Leaking sessions look like this :
select query, t.client_addr
 ,t.backend_start
 ,t.query_start
 ,t.wait_event
 ,t.state
from pg_stat_activity t order by query_start ;
query client_addr backend_start query_start wait_event state
DISCARD ALL 10.x.x.100 2019-11-08 08:05:58.161193+03 2019-11-08 08:34:04.004066+03 idle
DISCARD ALL 10.x.x.100 2019-11-08 08:28:39.133505+03 2019-11-08 08:34:06.832851+03 idle
DISCARD ALL 10.x.x.100 2019-11-08 08:15:14.533253+03 2019-11-08 08:35:48.405851+03 idle
DISCARD ALL 10.x.x.100 2019-11-08 08:48:14.914966+03 2019-11-08 08:48:15.145783+03 idle
DISCARD ALL 10.x.x.100 2019-11-08 08:48:16.06388+03 2019-11-08 08:48:16.172367+03 idle
DISCARD ALL 10.x.x.100 2019-11-08 08:28:58.24892+03 2019-11-08 08:50:11.85173+03 idle
DISCARD ALL 10.x.x.100 2019-11-08 08:41:17.423939+03 2019-11-08 08:50:39.68808+03 idle
DISCARD ALL 10.x.x.100 2019-11-08 08:54:55.598009+03 2019-11-08 08:54:55.697679+03 idle
DISCARD ALL 10.x.x.100 2019-11-08 08:58:02.941944+03 2019-11-08 08:58:02.967815+03 idle
... etc
all with "DISCARD ALL" query

10.x.x.100 - pgpool master IP

Its strange or I not understand something with pgpool parameters
We have next connections lifetime parameters

connection_life_time = 900

child_life_time = 3600
client_idle_limit = 1200

connection_life_time - 900 sec. Does pgpool not must close idle connection on backend after 900 sec ?

pgpool -v
pgpool-II version 3.7.3 (amefuriboshi)

Full config in attachments.
TagsNo tags attached.

Activities

gsl23

2019-11-08 15:42

reporter  

pgpool.conf (2,352 bytes)

gsl23

2019-11-08 16:30

reporter   ~0002958

Ps. We have script, it's killing sessions on db older than one hour, else them does not close a days long, and reside all connections (max_connections setting) on db.

t-ishii

2019-11-12 11:50

developer   ~0002959

Pgpool-II is waiting for reply from PostgreSQL after sending DISCARD ALL.
All connection_life_time, child_life_time,client_idle_limit parameters will not be activated for session in use.

You said that the issue can reproduced. So can you share debug log of pgpool for a session disconnected and pg_stat_avtivity log?

gsl23

2019-11-13 21:22

reporter   ~0002960

Thx t-ishii.
I did some analyze pgpool processes and pg_stat_avtivity. I find answers and tune some pool size parameters.
Pls close this issue.

t-ishii

2019-11-14 09:49

developer   ~0002961

Can you share how to solve your problem so that other users having same issue as you?

gsl23

2019-11-29 16:22

reporter   ~0002984

I tune main pool parameters : max_pool and num_init_children. They was just too big.

t-ishii

2019-12-05 18:13

developer   ~0003002

Thanks. I am going to close this issue.

Issue History

Date Modified Username Field Change
2019-11-08 15:42 gsl23 New Issue
2019-11-08 15:42 gsl23 File Added: pgpool.conf
2019-11-08 16:30 gsl23 Note Added: 0002958
2019-11-12 11:50 t-ishii Note Added: 0002959
2019-11-12 11:51 t-ishii Assigned To => t-ishii
2019-11-12 11:51 t-ishii Status new => feedback
2019-11-13 21:22 gsl23 Note Added: 0002960
2019-11-13 21:22 gsl23 Status feedback => assigned
2019-11-14 09:49 t-ishii Note Added: 0002961
2019-11-20 21:48 t-ishii Status assigned => feedback
2019-11-29 16:22 gsl23 Note Added: 0002984
2019-11-29 16:22 gsl23 Status feedback => assigned
2019-12-05 18:13 t-ishii Note Added: 0003002
2019-12-13 07:19 t-ishii Status assigned => closed