Project

General

Profile

Actions

Bug #14789

closed

Postgres fills up and can't be vaccumed

Added by Tobias Ell over 5 years ago. Updated over 5 years ago.

Status:
Released
Priority:
N/A
Category:
Web - Maintenance
Target version:
Severity:
Major - prevents use of part of Rudder | no simple workaround
UX impact:
User visibility:
Operational - other Techniques | Rudder settings | Plugins
Effort required:
Priority:
52
Name check:
Fix check:
Regression:

Description

Hi,
in the last 2 weeks I have had two occasions where the disk for my postgres db filled up.
I vaccum the rudder DB every night but lately that doesn't seem to be enough.
Checking the log of my housekeeping job I noticed that many tables have
"dead row versions cannot be removed yet" entries.
That seems to be a problem with transactions not committed.
I checked table pg_stat_activity and found an entry

16385 | rudder | 8071 | 16384 | rudder | PostgreSQL JDBC Driver | 127.0.0.1 | | 48393 | 2019-04-29 08:53:51.989498+02 |
2019-04-29 08:53:56.704253+02 | 2019-04-29 08:53:56.704567+02 | 2019-04-29 08:53:56.704589+02 | f | idle in transaction | SELECT 1

The client port 48393 belongs to a connection established by the jetty process.

My server is running on centOS7
all rudder components are version 5.0.9
postgres is 9.2.24

Has anyone encountered this problem before?
I am not too familiar with postgres, so it could be misconfiguration.
Any help is appreciated.


Files

vacuum_rudder_full.log (11.6 KB) vacuum_rudder_full.log Tobias Ell, 2019-04-30 15:29

Related issues 1 (0 open1 closed)

Related to Rudder - Bug #15413: Autovacuum may never finish, or take more than one day, on busy Rudder instance, because of resources attritionReleasedAlexis MoussetActions
Actions

Also available in: Atom PDF