100

I installed the Bitnami Django stack which included PostgreSQL 8.4.

When I run psql -U postgres I get the following error:

psql: could not connect to server: No such file or directory
    Is the server running locally and accepting
    connections on Unix domain socket "/var/run/postgresql/.s.PGSQL.5432"?

PG is definitely running and the pg_hba.conf file looks like this:

# TYPE  DATABASE        USER            CIDR-ADDRESS            METHOD

# "local" is for Unix domain socket connections only
local   all             all                                     md5
# IPv4 local connections:
host    all             all             127.0.0.1/32            md5
# IPv6 local connections:
host    all             all             ::1/128                 md5

What gives?

"Proof" that pg is running:

root@assaf-desktop:/home/assaf# ps axf | grep postgres
14338 ?        S      0:00 /opt/djangostack-1.3-0/postgresql/bin/postgres -D /opt/djangostack-1.3-0/postgresql/data -p 5432
14347 ?        Ss     0:00  \_ postgres: writer process                                                                        
14348 ?        Ss     0:00  \_ postgres: wal writer process                                                                    
14349 ?        Ss     0:00  \_ postgres: autovacuum launcher process                                                           
14350 ?        Ss     0:00  \_ postgres: stats collector process                                                               
15139 pts/1    S+     0:00              \_ grep --color=auto postgres
root@assaf-desktop:/home/assaf# netstat -nltp | grep 5432
tcp        0      0 127.0.0.1:5432          0.0.0.0:*               LISTEN      14338/postgres  
tcp6       0      0 ::1:5432                :::*                    LISTEN      14338/postgres  
root@assaf-desktop:/home/assaf# 
2
  • I have no idea what you're asking and you never provided input. This is 100 people getting a generic error and reporting different things. It's totally out of format for the site. Jan 12 '17 at 4:15
  • I my case, I had previous versions installed and it caused a mess. I cleaned those files, purged everything and reinstalled Postgres. That solved. Oct 18 '20 at 16:56

25 Answers 25

100

This issue comes from installing the postgres package without a version number. Although postgres will be installed and it will be the correct version, the script to setup the cluster will not run correctly; it's a packaging issue.

If you're comfortable with postgres there is a script you can run to create this cluster and get postgres running. However, there's an easier way.

First purge the old postgres install, which will remove everything of the old installation, including databases, so back up your databases first.. The issue currently lies with 9.1 so I will assume that's what you have installed

sudo apt-get remove --purge postgresql-9.1

Now simply reinstall

sudo apt-get install postgresql-9.1

Note the package name with the version number. HTH.

3
  • 1
    It worked also for me using ubuntu 16.04 and postgresql 9.5. As @Evert mentioned all related package must be purged. I followed this post: johnmee.com/how-to-reinstall-postgresql-on-ubuntu suggesting to purge with the following commands: "apt-get --purge remove postgresql*; rm -r /etc/postgresql/; rm -r /etc/postgresql-common/; rm -r /var/lib/postgresql/; userdel -r postgres; groupdel postgres".
    – green69
    Nov 11 '16 at 9:47
  • Right now they're at 9.5. sudo apt-get install postgresql postgresql-contrib has helped.
    – Ismail
    Mar 31 '17 at 22:49
  • sudo pg_ctlcluster 12 main start Sep 7 at 16:46
32

The error message refers to a Unix-domain socket, so you need to tweak your netstat invocation to not exclude them. So try it without the option -t:

netstat -nlp | grep 5432

I would guess that the server is actually listening on the socket /tmp/.s.PGSQL.5432 rather than the /var/run/postgresql/.s.PGSQL.5432 that your client is attempting to connect to. This is a typical problem when using hand-compiled or third-party PostgreSQL packages on Debian or Ubuntu, because the source default for the Unix-domain socket directory is /tmp but the Debian packaging changes it to /var/run/postgresql.

Possible workarounds:

  • Use the clients supplied by your third-party package (call /opt/djangostack-1.3-0/postgresql/bin/psql). Possibly uninstall the Ubuntu-supplied packages altogether (might be difficult because of other reverse dependencies).
  • Fix the socket directory of the third-party package to be compatible with Debian/Ubuntu.
  • Use -H localhost to connect via TCP/IP instead.
  • Use -h /tmp or equivalent PGHOST setting to point to the right directory.
  • Don't use third-party packages.
3
  • The -h localhost solves my issue.
    – azzamsa
    Mar 1 at 17:56
  • 1
    export PGHOST=localhost seems like a good workaround
    – Nux
    Mar 23 at 19:49
  • @azzamsa please mark that answer as chosen then. Aug 4 at 19:03
26

This works for me:

Edit: postgresql.conf

sudo nano /etc/postgresql/9.3/main/postgresql.conf

Enable or add:

listen_addresses = '*'

Restart the database engine:

sudo service postgresql restart

Also, you can check the file pg_hba.conf

sudo nano /etc/postgresql/9.3/main/pg_hba.conf

And add your network or host address:

host    all             all             192.168.1.0/24          md5
6
  • 1
    the listen_address = '*' made the trick. it was only listening on "localhost" and not on 127.0.0.1. thank you!
    – mwm
    Jun 25 '15 at 23:06
  • my god... finally something that worked - nothing else worked until I added the address and uncommented listen_address.
    – AntonB
    Aug 9 '16 at 18:09
  • Plus One! It worked for me. Oct 6 '16 at 20:18
  • 1
    this works on ubuntu windows bash. Apr 18 '18 at 8:21
  • I can confirm that it works on Ubuntu Server command bash on Windows 10.
    – Ronald
    Feb 26 '19 at 10:52
19

You can use psql -U postgres -h localhost to force the connection to happen over TCP instead of UNIX domain sockets; your netstat output shows that the PostgreSQL server is listening on localhost's port 5432.

You can find out which local UNIX socket is used by the PostgrSQL server by using a different invocavtion of netstat:

netstat -lp --protocol=unix | grep postgres

At any rate, the interfaces on which the PostgreSQL server listens to are configured in postgresql.conf.

0
19

Just create a softlink like this :

ln -s /tmp/.s.PGSQL.5432 /var/run/postgresql/.s.PGSQL.5432
4
  • 1
    This worked for me and seemed to be the easiest solution without having to change your postgresql configuration. Be sure that you are superuser when trying to make link.
    – brendan
    Mar 7 '13 at 20:11
  • 2
    ln: failed to create symbolic link '/var/run/postgresql/.s.PGSQL.5432': File exists Oct 27 '17 at 6:50
  • I've created "postgresql" folder in /var/run/ directory. It was not exist.
    – Ikrom
    Jun 10 '18 at 6:51
  • Works amazing, what is the reason behind this? Apr 11 '19 at 14:34
9

I make it work by doing this:

dpkg-reconfigure locales

Choose your preferred locales then run

pg_createcluster 9.5 main --start

(9.5 is my version of postgresql)

/etc/init.d/postgresql start

and then it works!

sudo su - postgres
psql
2
  • Hah,sorry, i think the commands have made it clear. for me,i meet this problem when i reinstall postgresql, i try to restart it by type service postgresql restart but it say i didn't have any postgresql cluster . Then I find the this way to help me out :)
    – mymusise
    Sep 13 '16 at 13:45
  • Well after three hours of Googling you finally fixed my issue. dpkg-reconfigure locales is pretty damn important. Feb 20 '17 at 22:35
6

I had to compile PostgreSQL 8.1 on Debian Squeeze because I am using Project Open, which is based on OpenACS and will not run on more recent versions of PostgreSQL.

The default compile configuration puts the unix_socket in /tmp, but Project Open, which relies on PostgreSQL, would not work because it looks for the unix_socket at /var/run/postgresql.

There is a setting in postgresql.conf to set the location of the socket. My problem was that either I could set for /tmp and psql worked, but not project open, or I could set it for /var/run/postgresql and psql would not work but project open did.

One resolution to the issue is to set the socket for /var/run/postgresql and then run psql, based on Peter's suggestion, as:

psql -h /var/run/postgresql

This runs locally using local permissions. The only drawback is that it is more typing than simply "psql".

The other suggestion that someone made was to create a symbolic link between the two locations. This also worked, but, the link disappeared upon reboot. It maybe easier to just use the -h argument, however, I created the symbolic link from within the PostgreSQL script in /etc/init.d. I placed the symbolic link create command in the "start" section. Of course, when I issue a stop and start or restart command, it will try to recreate an existing symbolic link, but other than warning message, there is probably no harm in that.

In my case, instead of:

ln -s /tmp/.s.PGSQL.5432 /var/run/postgresql/.s.PGSQL.5432

I have

ln -s /var/run/postgresql/.s.PGSQL.5432 /tmp/.s.PGSQL.5432

and have explicitly set the unix_socket to /var/run/postgresql/.s.PGSQL.5432 in postgresql.conf.

1
  • ln -s /var/run/postgresql/.s.PGSQL.5432 /tmp/.s.PGSQL.5432 this worked for me, I had installed postgresql using the apt package, but still I had to do this, else I had to login with sudo -su postgres everytime. Thanks for the answer. Oct 6 '20 at 7:04
5

If your Postgres service is up and running without any error or there is no error in starting the Postgres service and still you are getting the mentioned error, follow these steps

Step1: Running pg_lsclusters will list all the postgres clusters running on your device

eg:

Ver Cluster Port Status Owner    Data directory               Log file
9.6 main    5432 online postgres /var/lib/postgresql/9.6/main /var/log/postgresql/postgresql-9.6-main.log

most probably the status will be down in your case and postgres service

Step 2: Restart the pg_ctlcluster

#format is pg_ctlcluster <version> <cluster> <action>
sudo pg_ctlcluster 9.6 main start

#restart postgresql service
sudo service postgresql restart

Step 3: Step 2 failed and threw error

If this process is not successful it will throw an error. You can see the error log on /var/log/postgresql/postgresql-9.6-main.log

My error was:

FATAL: could not access private key file "/etc/ssl/private/ssl-cert-snakeoil.key": Permission denied
Try adding `postgres` user to the group `ssl-cert`

Step 4: check ownership of postgres

Make sure that postgres is the owner of /var/lib/postgresql/version_no/main

If not, run

sudo chown postgres -R /var/lib/postgresql/9.6/main/

Step 5: Check postgres user belongs to ssl-cert user group

It turned out that I had erroneously removed the Postgres user from the ssl-cert group. Run the below code to fix the user group issue and fix the permissions

#set user to group back with
sudo gpasswd -a postgres ssl-cert

# Fix ownership and mode
sudo chown root:ssl-cert  /etc/ssl/private/ssl-cert-snakeoil.key
sudo chmod 740 /etc/ssl/private/ssl-cert-snakeoil.key

# now postgresql starts! (and install command doesn't fail anymore)
sudo service postgresql restart
1
  • 1
    Having followed the instructions, I discovered the ownership of the /var/log/postgresql is me and not the postgres. Thanks for the full details.
    – mtoloo
    Sep 16 at 4:52
4

I found uninstalling Postgres sounds unconvincing. This helps to solve my problem:

  1. Start the postgres server:

    sudo systemctl start postgresql
    
  2. Make sure that the server starts on boot:

    sudo systemctl enable postgresql
    

Detail information can be found on DigitalOcean site Here.

3

Solution:

Do this

export LC_ALL="en_US.UTF-8"

and this. (9.3 is my current PostgreSQL version. Write your version!)

sudo pg_createcluster 9.3 main --start
1
  • woooow, it was the only solving my problem, thanks . Jul 17 '16 at 22:25
2

In my case it was caused by a typo I made while editing /etc/postgresql/9.5/main/pg_hba.conf

I changed:

# Database administrative login by Unix domain socket
local   all             postgres                                peer

to:

# Database administrative login by Unix domain socket
local   all             postgres                                MD5

But MD5 had to be lowercase md5:

# Database administrative login by Unix domain socket
local   all             postgres                                md5
1
  • 1
    This is the answer that fixed it for me :) I had previously changed mine to trusted instead of trust, and didn't restart the service, and it only broke the next day when I already forgot what I changed Mar 31 '18 at 7:25
2

I failed to solve this problem with my postgres-9.5 server. After 3 days of zero progress trying every permutation of fix on this and other sites I decided to re-install the server and lose 5 days worth of work. But, I did replicate the issue on the new instance. This might provide some perspective on how to fix it before you take the catastrophic approach I did.

First, disable all logging settings in postgresql.conf. This is the section:

# ERROR REPORTING AND LOGGING

Comment out everything in that section. Then restart the service.

When restarting, use /etc/init.d/postgresql start or restart I found it helpful to be in superuser mode while restarting. I had a x-window open just for that operation. You can establish that superuser mode with sudo -i.

Verify that the server can be reached with this simple command: psql -l -U postgres

If that doesn't fix it, then consider this:

I was changing the ownership on many folders while trying to find a solution. I knew that I'd probably be trying to revert those folder ownerships and chmods for 2 more days. If you have already messed with those folder ownerships and don't want to completely purge your server, then start tracking the settings for all impacted folders to bring them back to the original state. You might want to try to do a parallel install on another system and systematically check the ownership and settings of all folders. Tedious, but you may be able to get access to your data.

Once you do gain access, systematically change each relevant line in the # ERROR REPORTING AND LOGGING section of the postgresql.conf file. Restart and test. I found that the default folder for the logs was causing a failure. I specifically commented out log_directory. The default folder the system drops the logs into is then /var/log/postgresql.

1

Possibly it could have happened because you changed the permissions of the /var/lib/postgresql/9.3/main folder.

Try changing it to 700 using the command below:

sudo chmod 700 main
1

This is not exactly related to the question since I'm using Flask, but this was the exact error I was getting and this was the most relevant thread to get ideas.

My setup: Windows Subsystem for Linux, Docker-compose w/ makefile w/ dockerfile, Flask, Postgresql (using a schema consisting of tables)

To connect to postgres, setup your connection string like this:

from flask import Flask
app = Flask(__name__)
app.config['SQLALCHEMY_DATABASE_URI'] = "postgresql+psycopg2://<user>:<password>@<container_name_in_docker-compose.yml>/<database_name>"

NOTE: I never got any IP (e.g. localhost, 127.0.0.1) to work using any method in this thread. Idea for using the container name instead of localhost came from here: https://github.com/docker-library/postgres/issues/297

Set your schema:

from sqlalchemy import MetaData
db = SQLAlchemy(app, metadata=MetaData(schema="<schema_name>"))

Set your search path for your functions when you setup your session:

db.session.execute("SET search_path TO <schema_name>")
1

The most upvoted answer isn't even remotely correct because you can see in the question the server is running on the expected port (he shows this with netstat).

While the OP did not mark the other answer as chosen, they commented that the other answer (which makes sense and works) was sufficient,

But for these reasons that solution is poor and insecure even if it the server wasn't running on port 5432:


What you're doing here when you say --purge is you're deleting the configuration file for PostgreSQL ((as well as all of the data with the database. You or may not even see a warning about this, but here is the warning just to show you now,

Removing the PostgreSQL server package will leave existing database clusters intact, i.e. their configuration, data, and log directories will not be removed. On purging the package, the directories can optionally be removed. Remove PostgreSQL directories when package is purged? [prompt for yes or no]

When you add it again PostgreSQL is reinstalling it to a port number that's not taken (which may be the port number you expect). Before you even try this solution, you need to answer a few questions along the same line:

  • Do I want multiple versions of PostgreSQL on my machine?
  • Do I want an older version of PostgreSQL?
  • What do I want to happen when I dist-upgrade and there is a newer version?

Currently when you dist-upgrade on Ubuntu (and any Debian variant), the new version of PostgreSQL is installed alongside the old copy and the port number on the new copy is the port number of the old copy + 1. So you can just start it up, increment the port number in your client and you've got a new install! And you have the old install to fall back on -- it's safe!

However, if you only one want version of PostgreSQL purging to change the configuration is still not the right option because it will destroy your database. The only time this could even be acceptable is you want to destroy everything related to PostgreSQL. You're better off ensuring your database is correct and then merely editing the configuration file so the new install runs on the old port number

#!/bin/bash

# We can find the version number of the newest PostgreSQL with this
VERSION=$(dpkg-query -W -f "\${Version}" 'postgresql' | sed -e's/+.*//')
PGCONF="/etc/postgresql/${VERSION}/main/postgresql.conf"

# Then we can update the port.
sudo sed -ie '/port = /s/.*/port = 5432/' "$PGCONF"

sudo systemctl restart postgresql

Do not install a specific version of PostgreSQL. Only ever install postgresql. If you install a specific version then when you dist-upgrade your version will simply remain on your computer forever without upgrades. The repo will no longer have the security patches for the old version (which they don't support). This must always be suboptimal to getting a newer version that they do support, running on a different port number.

0

I had the exact same problem Peter Eisentraut described. Using the netstat -nlp | grep 5432 command, I could see the server was listening on socket /tmp/.s.PGSQL.5432.

To fix this, just edit your postgresql.conf file and change the following lines:

listen_addresses = '*'
unix_socket_directories = '/var/run/postgresql'

Now run service postgresql-9.4 restart (Replace 9-4 with your version), and remote connections should be working now.

Now to allow local connections, simply create a symbolic link to the /var/run/postgresql directory.

ln -s /var/run/postgresql/.s.PGSQL.5432 /tmp/.s.PGSQL.5432

Don't forget to make sure your pg_hba.conf is correctly configured too.

0

In my case, all i had to do was this:

sudo service postgresql restart

and then

sudo -u postgres psql

This worked just fine. Hope it helps. Cheers :) .

0

Find your file:

sudo find /tmp/ -name .s.PGSQL.5432

Result:

/tmp/.s.PGSQL.5432

Login as postgres user:

su postgres
psql -h /tmp/ yourdatabase
0

I had the same problem (on Ubuntu 15.10 (wily)). sudo find / -name 'pg_hba.conf' -print or sudo find / -name 'postgresql.conf' -print turned up empty. Before that it seemed that multiple instances of postgresql were installed.

You might have similar when you see as installed, or dependency problems listing

.../postgresql
.../postgresql-9.x 

and so on.

In that case you must sudo apt-get autoremove each package 1 by 1.

Then following this to the letter and you will be fine. Especially when it comes to key importing and adding to source list FIRST

sudo apt-get update && sudo apt-get -y install python-software-properties && wget --quiet -O - https://www.postgresql.org/media/keys/ACCC4CF8.asc | sudo apt-key add -

If not using wily, replace wily with your release, i.e with the output of lsb_release -cs

sudo sh -c 'echo "deb http://apt.postgresql.org/pub/repos/apt/ wily-pgdg main" >> /etc/apt/sources.list.d/postgresql.list'
sudo apt-get update && sudo apt-get install postgresql-9.3 pgadmin3

And then you should be fine and be able to connect and create users.

Expected output:

Creating new cluster 9.3/main ...
config /etc/postgresql/9.3/main
data   /var/lib/postgresql/9.3/main
locale en_US.UTF-8
socket /var/run/postgresql
port   5432

Source of my solutions (credits)

0

While having the same issue I tried something different:

Starting the postgresql daemon manually I got:

FATAL:  could not create shared memory segment ...
   To reduce the request size (currently 57237504 bytes), reduce PostgreSQL's
   shared memory usage, perhaps by reducing shared_buffers or max_connections.

So what I did was to set a lower limit for shared_buffers and max_connections into postgresql.conf and restart the service.

This fixed the problem!

Here's the full error log:

$ sudo service postgresql start
 * Starting PostgreSQL 9.1 database server                                                                                                                                                               * The PostgreSQL server failed to start. Please check the log output:
2013-06-26 15:05:11 CEST FATAL:  could not create shared memory segment: Invalid argument
2013-06-26 15:05:11 CEST DETAIL:  Failed system call was shmget(key=5432001, size=57237504, 03600).
2013-06-26 15:05:11 CEST HINT:  This error usually means that PostgreSQL's request for a shared memory segment exceeded your kernel's SHMMAX parameter.  You can either reduce the request size or reconfigure the kernel with larger SHMMAX.  To reduce the request size (currently 57237504 bytes), reduce PostgreSQL's shared memory usage, perhaps by reducing shared_buffers or max_connections.
    If the request size is already small, it's possible that it is less than your kernel's SHMMIN parameter, in which case raising the request size or reconfiguring SHMMIN is called for.
    The PostgreSQL documentation contains more information about shared memory configuration.
0

After many exhausting attempts, I found the solution based on other posts!

dpkg -l | grep postgres
apt-get --purge remove <package-founded-1> <package-founded-2>
whereis postgres
whereis postgresql
sudo rm -rf <paths-founded>
sudo userdel -f postgres
1
0
  1. Check the status of postgresql:

    service postgresql status
    

    If it shows online, proceed to step no 3 else execute step no 2.

  2. To make postgresql online, execute the following command:

    sudo service postgresql start
    

    Now check the status by running the command of the previous step. It should show online.

  3. To start psql session, execute the following command:

    sudo su postreg
    
  4. Finally, check if it's working or not by executing:

    psql
    
-1

Create postgresql directory inside run and then run the following command.

ln -s /tmp/.s.PGSQL.5432 /var/run/postgresql/.s.PGSQL.5432
1
  • 1
    Putting the socket in /tmp is a horrible idea and insecure. When you restart anyone can recreate that file with regular root perms and confuse your app into talking to them instead. You wouldn't know if you had a man-in-the-middle. Aug 4 at 21:11
-1

Simply add /tmp unix_socket_directories

postgresql.conf

unix_socket_directories = '/var/run/postgresql,/tmp'
2
-1

I had this problem with another port. The problem was, that I had a system variable in /etc/environments with the following value:

PGPORT=54420

As I removed it (and restarted), psql was able to connect.

2
  • You added it there. No one else is going to have done that... Aug 4 at 21:09
  • @EvanCarroll First of all, it's a possible solution for this problem. Secondly, why do you care at all if it does not concern you. And why do you think you know what settings the users have.
    – Bevor
    Aug 5 at 4:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.