Research Computing Configuration

Jump to navigation Jump to search

This page describes the configuration of the new research computing machines: Father (Windows Server 2019) and Mother (Ubuntu Server 20.04). Note that the RDP Software Configuration describes the software installed on Father.

The hardware description and complete build notes and configuration information for Bastard, our blisteringly fast, multi-GPU, A.I. estimation platform, are on the DIGITS DevBox page. The hardware descriptions for Father and Mother are on the Research Computing Hardware page.

See also: Recovering Astarte for notes on moving content from an old mediawiki installation to a new one.


Both machines (Father and Mother)

Fan Control

Unless you want to go insane from the sound of fans cycling full on and off, you'll want to fix the IPMI fan settings. These are stored in the BIOS but not accessible through the BIOS screens. Instead, you should connect to the boxes by BMC over the network. The BMC IPv4 address is displayed during POST.

To read all about IPMI, see I also found these helpful:

Note the that default BMC username and password is ADMIN and ADMIN. You can download the SMCIPMITool (2.21.0_build.181029) and do the following, though I couldn't work out how to send manual configuration instructions using it:

.\SMCIPMITool.exe ADMIN ADMIN ipmi fan
.\SMCIPMITool.exe ADMIN ADMIN ipmi fan 0

The trick is to change the thresholds for the fans, especially the lower threshold. On a linux box:

sudo apt-get install ipmitool
  1. Reset the BMC if you've screwed it up (or if your fans are full on all the time)
  2. Check that the current mode is optimal (2)
  3. Take a look at the sensor multiple times to see that the fan is hitting the constraint (run multiple times to coincide with different sound levels)
  4. Reset the lower thresholds on the fans
  5. Enjoy a perfectly reasonable fan speed that doesn't fluctuate unduly
ipmitool -I lan -U ADMIN -P ADMIN -H raw 0x3c 0x40
ipmitool -I lan -U ADMIN -P ADMIN -H raw 0x30 0x45 0x00
ipmitool -I lan -U ADMIN -P ADMIN -H sensor
ipmitool -I lan -U ADMIN -P ADMIN -H sensor thresh FAN3 lower 120 220 320

Note that the thresholds are derived from -20% off 400 (min as per Noctua spec), then same -100, then same -200.

For the ARCTIC F8 PWM, the min is something like 250 at 5v (see I therefore used 100, 150, 200, which seemed to stabilize the fans nicely at a 300rpm idle on one box and 400rpm on the other.

Note that the BMC IP for mother often turns up on You can see the IP address of the localhost by running:

ipmitool lan print

Bios Settings

Because we want the NVMe drives to be bootable, we need to use (U)EFI for both machines:

  • Change CPU1 Slot 1 and 2 to EFI (from Legacy)
  • Change Onboard Video OPROM to EFi (from Legacy)
  • LAN device to EFI
  • Boot to EFI
  • Priority to onboard: auto
  • Boot install order CD above drive above Shell

This ultimately seems to be a problem for unsigned Nvidia drivers in Linux, which make putting a GPU into the dbase server box a major issue. I tried disabling validation and adding a key to secure boot (not sure that it happened) but nothing I could do would fix the resulting driver issue.

mokutil --disable-validation
sudo update-secureboot-policy --new-key
sudo update-secureboot-policy --enroll-key

RDP Server (Father)

The RDP server runs Windows 2019 Server. It installs directly off the media on to the NVMe drive. Don't worry about the RAID array during the installation, we do that later.

After installation:

  • Set computer name
  • Storage Pool
    • RAID 1 - Mirroring
  • Active Directory Controller
  • Remote Desktop Connection
    • RD Connection Broker
    • RD Licensing
    • RD Session Host
    • RD Vizualization Host

Change password complexity requirements:


The GPU was removed from the RDP and the chipset drivers from Supermicro were installed (files are in E:/installs/drivers). This addressed all the device issues.

The following software was uninstalled:

  • CUDA Development 10.1
  • CUDA Documentation 10.1
  • CUDA Samples 10.1
  • CUDA Visual Studio Integration 10.1
  • NVIDIA GeForce Experience
  • NVIDIA Graphics Driver 419.67
  • NVIDIA HD Audio Diver
  • NVIDIA Nsight Compute v2019.1
  • NVIDIA Nsight Systems v2018.3.3
  • NVIDIA Nsight Visual Studio Edition
  • NVIDIA PhysX System Software 9.12.0218
  • NVIDIA Toos Extension SDK
  • NVIDIA USBC Driver

Dbase Server (Mother)

The dbase server runs Ubuntu 18.04. You can mostly follow the instructions at

It is important to use the alternative installation iso, not the default cloud based one. However, you can set up the partitions using the standard live CD, which is more user friendly for partitioning.


Partition as:

  • 400G NVMe as ext4 and / (create new using whole drive, automatically sets aside 512m for /boot/efi)
  • 1.2T as ext4 and /data (do it manually)
  • RAID 10 for HDDs, all active then format at ext4 and mount at /bulk (DO NOT PARTITION! Just set up the software RAID and format.)

Standard Packages

Install the following straight from the media:

  • LAMP Server
  • Mail Server
    • Postfix - Internet Site
    • MailName:
  • PostgreSQL
  • Samba
  • OpenSSH


sudo apt-get install tasksel
 -> Ubuntu Desktop

A useful apt cheatsheet is:

Get the system up to date:

apt-get update
apt-get upgrade


This guide is helpful:

Check samba is running

samba --version

Then fix the conf file:

cp /etc/samba/smb.conf /etc/samba/smb.conf.bak
vi /etc/samba/smb.conf
 	usershare allow guests = no
	;comment [printers] and [print$]
	comment = Bulk RAID Array
	path = /bulk
	browseable = yes
	create mask= 0775
	directory mask = 0775
	read only = no
	guest ok = no

Test the parameters, change the permissions and ownership:

testparm /etc/samba/smb.conf
chmod 770 /bulk
groupadd smbusers
chown :smbusers /bulk

Now add the researcher account, and add it to the samba share

groupadd -g 1001 researcher
useradd -g researcher -G smbusers -s /bin/bash -p 1234 -d /home/researcher -m 
passwd researcher
	hint: littleamount
smbpasswd -a researcher

Finally restart samba:

systemctl restart smbd

Check it works:

smbclient -L localhost
(no root password)

And add users to the samba group:

useradd ed smbusers


This guide is helpful:

Test it!

psql --help
ss -nlt
	postgres is listening on 543

Back up the config file and try a manual launch:

cp /etc/postgresql/10/main/postgresql.conf /etc/postgresql/10/main/postgresql.conf.bak
mkdir /data/postgres
chown postgres:postgres postgres
su postgres
cd /usr/lib/postgresql/10/bin
./initdb -D /data/postgres

Now, tune the database server. See and

vi /etc/postgresql/10/main/postgresql.conf
 data_directory = '/data/postgres' #custom 1.2Tb NVME SSD
 listen_addresses = '*'
 max_connections = 10
 shared_buffers = 128GB 
 huge_pages = try                        # on, off, or try
 temp_buffers = 8GB                      # min 800kB
 work_mem = 4GB                          # min 64kB
 maintenance_work_mem = 64GB             # min 1MB
 max_stack_depth = 6MB
 max_wal_senders = 5   #MUST SET THIS TO BE LESS THAN max_connections
 effective_cache_size = 384GB

Now fix the hba.conf file for access rights:

cp /etc/postgresql/10/main/pg_hba.conf /etc/postgresql/10/main/pg_hba.conf.bak
vi /etc/postgresql/10/main/pg_hba.conf

 local all all peer
 local all all trust
Don't do this bit for now:
	 local   all             postgres                   peer 
	 local   all             postgres                   md5

And restart!

service postgresql restart
ss -nlt

If postgres isn't listening then it didn't start despite the message at

service postgresql status

Then we note that Postgres bins are in in:

cd /usr/lib/postgresql/10/bin


cat /etc/init.d/postgresql
cat /usr/share/postgresql-common/init.d-functions

To diagnose error do manual start as postgres from /usr/lib/postgresql/10/bin:

pg_ctl -w -D /data/postgres -o '--config-file=/etc/postgresql/10/main/' start

To check it is working:

ss -nlt
psql postgres
SHOW data_directory;
SHOW work_mem;

When it is all ok, you can go back to using

service postgresql restart
(it should take a few secs)

Make the researcher user!

createuser --interactive researcher

Add Extensions

Finally, add some extentions

apt-get install postgresql-plperl-10
apt-get install postgresql-plpython-10
apt-get install postgresql-10-plr
apt-get install postgresql-10-postgis-2.4 
apt-get install postgresql-10-postgis-scripts
apt-get install postgis
apt-get install postgis-gui

As postgres:

psql template1


We had a back off of an old MySQL mediawiki dbase and the contents of the mediawiki directory, and we wanted to restore the old wiki. This is what we did.

Restore the old dbase:

cd /bulk/mcnair/Web/mysqldump 
 dbase is mcnair
mysql -h localhost < web_mysqldump_backup_Fri_Aug_24_15_35_47_2018.sql

Connect to MySQL and check what we have:

show databases;
use mcnair;
show tables;

If you need to: systemctl stop mysql systemctl start mysql

Get the old install:

cd home/home/mcnair/Downloads/
tar -xvzf mediawiki-1.26.2.tar.gz 

Add php-xml

apt-get install php-xml
apachectl restart

We then had to fix the passwords in the dbase:

#Note: change passwords from hint before running
SELECT User, Host, Password FROM mysql.user;
UPDATE mysql.user SET Password = PASSWORD('hint') WHERE User = 'root';
UPDATE mysql.user SET Password = PASSWORD('hint') WHERE User = 'debian-sys-maint';
UPDATE mysql.user SET Password = PASSWORD('hint') WHERE User = 'mcnair_wp';

At this point, the basics are working, so go to and fill it out as per the old instructions (see Test Web Server Documentation and Web Server Documentation

Now overwrite LocalSettings.php with the old configuration:

cd /home/ed/Downloads/
mv LocalSettings.php /var/www/html/mediawiki/

Fix the requirements for mediawiki

apt-get install php-xml
apachectl restart

Allow short URLS, so enable mod-rewrite (if not already enabled)

a2enmod rewrite
systemctl restart apache2

Now fix the apache conf file

cp /etc/apache2/sites-available/000-default.conf /etc/apache2/sites-available/000-default.conf.bak
vi /etc/apache2/sites-available/000-default.conf 
       Alias /wiki /var/www/html/mediawiki/index.php
       #Enable the rewrite engine
       RewriteEngine On
       #Rewrite / to Main Page
       RewriteRule ^/*$ %{DOCUMENT_ROOT}/mediawiki/index.php [L]
service restart apache2

Now create phpinfo page for debugging

cd /var/www/html
vi phpinfo.php
	<?php echo phpinfo(); ?>
	#Browse to
	Shows Phar is installed and running
	Shows log is /var/log/apache2 

Check pcntl is enabled

php --ri pcntl
#Note that some pcntl functions are listed as disabled in phpinfo.php
dpkg -s snmp
apt-get install snmp

The Big Try

The process is as follows:

  1. Move the contents of /var/lib/mediawiki to somewhere else
  2. Move the contents of /bulk/mcnair/Web/www/var/www/html/mediawiki in
  3. Drop databases
  4. Restore databases
  5. Pray

Or more specifically:

mv /var/lib/mediawiki/ /var/lib/firstmediawikitry
mkdir /var/lib/mediawiki
cp -r /bulk/mcnair/Web/www/var/www/html/mediawiki /var/lib/
#Change password for dbase in LocalSettings.php
mysql -p
DROP DATABASE wordpress;
cd /bulk/mcnair/Web/mysqldump 
mysql -p -h localhost < web_mysqldump_backup_Fri_Aug_24_15_35_47_2018.sql
apachectl restart

When we did this, we got a blank page! Don't panic.

cd /var/log/apache2
cat error.log
apt-get install php7.2-mbstring
apachectl restart

Fix LocalSettings.conf again

change domain name, contact detail, etc.
Need to fix mail...
Also left GoogleAnalytics extension loaded for now...
Change enable_semantics IP address

Fix the Infoboxes:

chmod a+x /var/libr/mediawiki/extensions/Scribuntu...
chcon -t httpd_sys_script_exec_t /var/lib/mediawiki/Scribuntu...
mv /etc/apache2/sites-available/000-default.conf /etc/apache2/sites-available/
mv /etc/apache2/sites-available/000-default.conf.bak /etc/apache2/sites-available/000-default.conf
apachectl restart
mv /etc/apache2/sites-available/ /etc/apache2/sites-available/000-default.conf
apachectl restart

We were left with a problem were the page would never finish loading. This turned out to be a problem with the fonts in the Vector skin, which we had previously customized. We tried to fix the problem in the dbase as below but to no avail.

SELECT * FROM externallinks WHERE el_id=2599;
UPDATE externallinks SET el_to = '' WHERE el_id = 2720;
UPDATE externallinks SET el_index = '' WHERE el_id = 2720;
UPDATE externallinks SET el_to = '' WHERE el_id = 2721;
UPDATE externallinks SET el_index = ' WHERE el_id = 2721;
UPDATE externallinks SET el_to = ' ' WHERE el_id = 2722;
UPDATE externallinks SET el_index = ' ' WHERE el_id = 2722;
UPDATE externallinks SET el_to = ' WHERE el_id = 2723;
UPDATE externallinks SET el_index = '' WHERE el_id = 2723;
UPDATE externallinks SET el_to = '' WHERE el_id = 2724;
UPDATE externallinks SET el_index = '' WHERE el_id = 2724;
UPDATE externallinks SET el_to = '' WHERE el_id = 2739;
UPDATE externallinks SET el_index = '' WHERE el_id = 2739;
UPDATE externallinks SET el_to = '' WHERE el_id = 2599;
UPDATE externallinks SET el_index = '' WHERE el_id = 2599;

What did work was:

cd /var/www/html/mediawiki/skins
diff -r Vector VectorBackup
cp -r Vector/ VectorFromMcNair
vi /skins/Vector/variables.less

replace all font-family statements with "Linux Libertine", Georgia, Times, sans-serif, serif;

When I rebooted the MySQL database was inaccessible for reasons unknown... I followed this page:

service mysql stop
mysqld_safe --skip-grant-tables &
mkdir -p /var/run/mysqld
chown mysql:mysql /var/run/mysqld
mysqld_safe --skip-grant-tables &

And then all was good!

Confirm Account

We had some issues with some of the extensions, particularly confirm account. To debug the wiki add this line to LocalSettings.php

$wgShowExceptionDetails = true;

Fix the confirm account, see and add lines to LocalSettings.conf.

apt-get install php7.2-dev 
apt-get install php-pear
pear -version
pear config-get php_dir
phpinfo returns:

Then make check_pear.php as per It should return bool true


php -c /etc/php/7.2/apache2/php.ini -r 'echo get_include_path()."\n";'


; UNIX: "/path1:/path2"
include_path = ".:/usr/share/php"



From Test Web Server Documentation it doesn't look like we had to install anything before we installed Wordpress. The restoration plan is therefore:

  1. Restore the dbase (done already when we restored the mediawiki dbase)
  2. Copy over all of the wordpress files
  3. Create a /blog alias in apache
  4. Check the permissions and pray

So we did:

cd /bulk/mcnair/Web/www/var/www/html
cp -r ./blog/ /var/www/html/blog
cd /var/www/html/blog/

Check Apache2.conf

vi /etc/apache2/apache2.conf
 looks fine

Temporarily force some settings

vi wp-config.php
 define('DB_PASSWORD', 'tsn');

Make some changes to the dbase

connect wordpress;
SELECT ID, user_login, user_pass FROM wp_users;
UPDATE wp_users SET user_pass=MD5('newstrongpassword') WHERE ID = 4;
select * from wp_options where option_name='siteurl';
select * from wp_options where option_name='home';
UPDATE wp_options SET option_value='' WHERE option_name='siteurl';
UPDATE wp_options SET option_value='' WHERE option_name='home';

Now you can comment out the WP_HOME and WP_SITEURL settings in wp-config.php and change them (if you want) from the wp-admin interface:

The following plugin had to be disabled:

  • Social Share WordPress Plugin - AccessPress Social Share

Finally, fix the permalink issue by setting

vi /etc/apache2/apache2.conf
 AllowOverride All

Then yay!

The whole thing needs updating, a new skin (or at least clean up), and some of the plugins don't work. But the basics are now up and running.


Go in to the wp-admin interface and hit update. It seemed to work fine!

Run the site-health.php tool:

apt-get install php7.2-gd
apt-get install php7.2-bcmath 
Get and make imagemagick -- see 
 magick -help
apachectl restart

Also, update all of the plugins and remove the one inactive plugin that was causing problems earlier.

Other Web Server

For Google Analytics we linked the domain to on the Google Dashboard and added the key to LocalSettings.php. See

We also added write permissions to the images directory for www-data

chown -R www-data images/


The original intention was to install a GPU into the Dbase server, as GPU compute tasks wouldn't interfere (much) with the main operation of the server. The problem seems to be a combination of an unsigned Nvidia driver, Ubuntu 18.04, UEFI, and Secureboot (or not). See

Regardless here are some useful commands:

See what drivers are being used

apt-get install ubuntu-drivers-common
ubuntu-drivers devices
cat /proc/driver/nvidia/version

See the display hardware config

sudo lshw -c display
 If shows *-display UNCLAIMED and no driver associated with it

Just try to work out what is going on:

ubuntu-drivers devices
lshw -c display
sudo lspci -vk
ls -l /sys/firmware/efi/ 

Try installing CUDA and its driver: And then: And may:

Try installing the bundled cuda toolkit

apt-get install nvidia-cuda-toolkit
apt-get install cuda-samples-7-0 -y
cd /usr/local/cuda-7.0/samples

Try installing the bundled nvidia driver

sudo apt install nvidia-driver-415

Purge nvidia drivers and add the experimental repo

apt-get purge nvidia*
sudo add-apt-repository ppa:graphics-drivers

Get the latest driver from Nvidia and make it

apt-get install gcc
apt-get install make
cat /var/log/nvidia-installer.log

View system logs:

journalctl -xb

Remove nouveau if being used (

sudo ubuntu-drivers devices
sudo apt-get --purge remove xserver-xorg-video-nouveau

Add a secure boot key

sudo update-secureboot-policy --new-key
sudo update-secureboot-policy --enroll-key

Disable/enable secureboot:

mokutil --disable-validation
 create 8-12 char password (same as ed's)
 Reboot and disable secureboot
mokutil --enable-validation



I installed pdftk and configured it as follows:

snap install pdftk
ln -s /snap/pdftk/current/usr/bin/pdftk /usr/bin/pdftk

Its man page (kinda) is here:

But generally you want to do use it combine files:

pdftk *.pdf cat output newfile.pdf
pdftk a.pdf b.pdf cat output newfile.pdf

If you want to reduce a file that has large images in it, then the following sometimes works:

pdf2ps 1.pdf
ps2pdf -dPDFSETTINGS=/screen -dDownsampleColorImages=true -dColorImageResolution=144 -dColorImageDownsampleType=/Bicubic 1.pdf

To do

Mediawiki Redux

Mobile Front End

It seems that the Extension:MobileFrontend isn't working properly.

Does php have mbstring support?

First check the apache version:

apache2 -v
 Server version: Apache/2.4.29 (Ubuntu)
 Server built:   2018-10-10T18:59:25

And the php version:

php -v
 PHP 7.2.17-0ubuntu0.18.04.1 (cli) (built: Apr 18 2019 14:12:38) ( NTS )
 Copyright (c) 1997-2018 The PHP Group
 Zend Engine v3.2.0, Copyright (c) 1998-2018 Zend Technologies
   with Zend OPcache v7.2.17-0ubuntu0.18.04.1, Copyright (c) 1999-2018, by Zend Technologies

Try just installing it:

apt-get install php7.2-mbstring

This failed because of an issue with dpkg. I rebooted and then:

dpkg --configure -a
apt --fix-broken install

Then it installed but threw a notice about a modified configuration file. The differences were too large to show. I kept the old file (/etc/php/7.2/apache2/php.ini) for comparison to the new one (/usr/lib/php/7.2/php.ini-production).

In the old file, uncomment the mbstring extension and exif (after it) too, then reload the config and check it worked:

service apache2 reload
php -i | grep mbstring

So everything seems fine (the extension it is listed as loaded in Special:Version), but the interface still has obvious issues.

Check the skin

From Special:Version

MediaWiki	1.26.2
PHP	7.2.24-0ubuntu0.18.04.2 (apache2handler)
MySQL	5.7.25-0ubuntu0.18.04.2
Lua	5.1.5

Vector is the only installed skin.

And it looks like my version of mediawiki is too old to support Minerva Neue

Upgrade mediawiki

Essentially, follow instructions in Manual:Upgrading:


I just moved everything to a different directory, backed off the dbase, and started again.

cd /var/lib
mv mediawiki mediawikibackup26082020
mysqldump --user=root --password=password  > dbase.sql

New Install

Get a new version, put it in /var/lib/mediawiki (leaving the old shortcuts pointing there), then copy in the files.

tar -xvzf mediawiki-1.34.2.tar.gz 
mv mediawiki-1.34.2 mediawiki
rm mediawiki-1.34.2.tar.gz
cp mediawiki26082020/LocalSettings.php mediawiki/LocalSettings.php
cp -a mediawiki26082020/images/ mediawiki/
  Note: you don't need to change permissions because we used -a. This also copied the .htaccess file. Everything should be ok...
cp /var/lib/mediawiki26082020/resources/assets/EdEganDotCotWikiGreenTab.png /var/lib/mediawiki/resources/assets/
 Note: that's my wiki logo
cp /var/lib/mediawiki26082020/favicon.ico /var/lib/mediawiki/favicon.ico
 Note: that the favicon!
cd /var/lib/mediawiki/maintenance/
php update.php


Retrieve and include extensions

Get the bulk of them...

mkdir installs
cd installs

tar -xzf LabeledSectionTransclusion-REL1_34-4aa6bfa.tar.gz -C /var/lib/mediawiki/extensions

tar -xzf MobileFrontend-REL1_34-6a8ef84.tar.gz -C /var/lib/mediawiki/extensions

tar -xzf ImportUsers-REL1_34-2f1a670.tar.gz -C /var/lib/mediawiki/extensions

tar -xzf Scribunto-REL1_34-f7bc2e3.tar.gz -C /var/lib/mediawiki/extensions

tar -xzf googleAnalytics-REL1_34-6441403.tar.gz -C /var/lib/mediawiki/extensions

tar -xzf MultiUpload-REL1_34-e018c1d.tar.gz -C /var/lib/mediawiki/extensions

tar -xzf UserMerge-REL1_34-3517022.tar.gz -C /var/lib/mediawiki/extensions

tar -xzf MinervaNeue-REL1_34-ba11b7b.tar.gz -C /var/lib/mediawiki/skins

tar -xzf ConfirmAccount-REL1_34-3ffa446.tar.gz -C /var/lib/mediawiki/extensions

tar -xzf MassEditRegex-REL1_34-d3570f1.tar.gz -C /var/lib/mediawiki/extensions

tar -xzf DataTransfer-REL1_34-1fc1c61.tar.gz -C /var/lib/mediawiki/extensions

tar -xzf HTMLets-REL1_34-a8227c3.tar.gz -C /var/lib/mediawiki/extensions

tar -xzf MinervaNeue-REL1_34-ba11b7b.tar.gz -C /var/lib/mediawiki/skins

tar -xzf GeoData-REL1_34-8a52fa4.tar.gz -C /var/lib/mediawiki/extensions
tar -xzf NumberFormat-REL1_34-cf8a23e.tar.gz -C /var/lib/mediawiki/extensions

Do the semantic mediawiki install and config. Last time, we installed Semantic Mediawiki using composer, which is the preferred method. See Web_Server_Documentation.

cd installs
php -r "copy('', 'composer-setup.php');"
php -r "if (hash_file('sha384', 'composer-setup.php') === '8a6138e2a05a8c28539c9f0fb361159823655d7ad2deecb371b04a83966c61223adc522b0189079e3e9e277cd72b8897') { echo 'Installer verified'; } else { echo 'Installer corrupt'; unlink('composer-setup.php'); } echo PHP_EOL;"
php composer-setup.php --install-dir=/usr/local/bin --filename=composer
#php -r "unlink('composer-setup.php');"
vi composer.local.json
     "require": {
         "mediawiki/semantic-media-wiki": "~3.1",
         "mediawiki/semantic-result-formats": "~3.1"
composer update --no-dev
Add line to LocalSettings.php
php maintenance/update.php
cd ../extensions/
git clone
 add to LocalSettings.php

ConfirmAccount Bug

Installing ConfirmAccount gave an error whenever there was an account request.

Warning: file_exists(): Unable to find the wrapper "mwstore" - did you forget to enable it when you configured PHP? in /../w/extensions/ConfirmAccount/backend/ConfirmAccount.class.php on line 29

This seemed to be a registered bug - see It appears to happen as a consequence of the file_exists call, and is just a warning. It doesn't affect functionality. So I put an @ in front of the file_exists call, which is poor practice but it worked just fine.

Line 29:  if ( $path && @file_exists( $path ) ) {

Other Config

Check pre-reqs are good for Scribuntu.

php -r 'echo "pcre: " . ( extension_loaded( "pcre" ) ? PCRE_VERSION : "no" ) . "\n";'
php -r 'echo "mbstring: " . ( extension_loaded( "mbstring" ) ? "yes" : "no" ) . "\n";'

Fix some permissions:

chmod a+x extensions/Scribunto/includes/engines/LuaStandalone/binaries/lua5_1_5_linux_64_generic/lua
chmod a+x extensions/SyntaxHighlight_GeSHi/pygments/pygmentize

Tune up php and Apache2:

vi /etc/php/7.2/apache2/php.ini
 change upload_max_filesize to 4M
vi /etc/apache2/apache2.conf
  <Directory /var/www/wiki/images>
         Options -Indexes
service apache2 restart


I tried various methods to get MathML to work and always failed. It looks like the community bet on Mathoid working out, but there's been no development on it for 5 months now and it looks dead. The good news is that MathJax works just fine right out of the box:

git clone
wfLoadExtension( 'SimpleMathJax' );
#$wgSmjInlineMath = [ [ "$", "$" ], [ "\\(", "\\)" ] ];
 Note: the last line lets you demark math with LaTeX-like syntax. I disabled it, as I use $ signs way too much in other contexts.


PDFEmbed was the extension that I never knew that I always wanted!

git clone
wfLoadExtension( 'PDFEmbed' );


IntraACL ( doesn't work with Mediawiki 1.34, so it's out. I also wasn't wild about its patch-based approach.

This time around I went with SemanticACL for access control. It's in beta but its actively maintained and its approach is simple, sane, and harnesses the power of what's already there. No hooks, no patches, no blah... just an extension that does what you want out of the box.

The only other real contender was Extension:AccessControl, which is stable. However, its approach just isn't as clean, and the author's request for funds to translate his documentation from Czech to English didn't endear me.

Useful Links for SemanticACL:

tar -xzf SemanticACL-REL1_34-01ae8be.tar.gz -C /var/lib/mediawiki/extensions

In LocalSettings.php (for m1.34):

require_once "$IP/extensions/SemanticACL/SemanticACL.php";

To configure security on a page (

[[Visible to::whitelist]]
 [[Visible to group::team]]
 [[Editable by::whitelist]]
 [[Editable by user::User:whoever]]  

or equivalently (but silently):

{{#set: Visible to=whitelist|Visible to group=team}}

The BibTeX extension doesn't work anymore. Though it never really worked, so it's not much of a loss. I should probably build a replacement but I don't have the time right now.

Upload Multiple Files

Although the Upload multiple files extension installed fine, it is unmaintained and seems to have an issue. I removed its line from LocalSettings.php and deleted its extension directory.

I then installed Simple Batch Upload using a tarball:

in mediawiki/installs
 tar -xzf 1.6.0.tar.gz -C /var/lib/mediawiki/extensions
 mv /var/lib/mediawiki/extensions/SimpleBatchUpload-1.6.0 /var/lib/mediawiki/extensions/SimpleBatchUpload

In LocalSettings.php:

wfLoadExtension( 'SimpleBatchUpload' );
$wgSimpleBatchUploadMaxFilesPerBatch = ['*' => 10,];

I had previously added Special:MultiUpload|Upload multiple files to I replaced it with a link to Special:BatchUpload.

Allow SVG images

See Essentially, add svg to $wgFileExtensions, then install and designate an image converter. I went with rsvg:

apt-get install librsvg2-bin
vi LocalSettings.php
 $wgSVGConverter = 'rsvg';
Add HitCounters
tar -xzf HitCounters-REL1_34-48dd6cb.tar.gz -C /var/lib/mediawiki/extensions
vi ../LocalSettings.php
 wfLoadExtension( 'HitCounters' );
cd ../maintenance
php update.php

Change the Dbase

I tried to 'rename' the dbase, creating a dedicated dbase user that has access to just the wiki's dbase, and resetting its password.

in /bulk/backups:

mysqldump --password olddbase > mediawiki.sql
mysql -u olduser -p 
mysql -u olduser -p wiki < mediawiki.sql
mysql -u olduser -p 
 CREATE USER 'wiki'@'localhost' IDENTIFIED BY 'password';
 GRANT ALL PRIVILEGES ON wiki.* TO 'wiki'@'localhost';

This led to two different errors. First, the cloned database didn't seem to support Semantic Mediawiki somehow. And second, the new user didn't seem to work. This was true even if I gave them rights on the olddbase. So I abandoned the attempt.

Clean up the attempt:

mysql -u olduser -p 
 DROP USER 'wiki'@'localhost';

I did put an .htaccess file in /var/lib/mediawiki to restrict access to LocalSettings.php, though I expect that this is redundant.

<files LocalSettings.php>
order allow,deny
deny from all

Setting up for advanced template(s) import


Importing the Robelbox, or other, templates on mediawiki is tricky [1], at least the first time that you do it. Once you've got everything up and running to support templates (see above):

  1. Find the template on another mediawiki installation
  2. Go to Special:Export on that installation and export the template (but not its revision history)
  3. Import the template on the wiki using Special:Import, set the interwiki prefix to something that designates the source, like en for English Wikipedias.

I did this for the Robelbox template, which I got from, however, it wasn't useable and I couldn't work out why. I ultimately deleted Robelbox, having found better boxes (see below) but I expect that my process for fixing the later issues would have sorted out the problems here too.

Fixing Template Issues

I got Template:Box-round from It required installation of TemplateStyles, which in turn might need JsonConfig:

In mediawiki/installs

tar -xzf TemplateStyles-REL1_34-c4d6f25.tar.gz -C /var/lib/mediawiki/extensions
tar -xzf JsonConfig-REL1_34-f877d87.tar.gz -C /var/lib/mediawiki/extensions

Add to LocalSettings.php

wfLoadExtension( 'TemplateStyles' );
wfLoadExtension( 'JsonConfig' );

Now Template:Box-round works fine but, like Template:Tl, has an error message on its page about JsonConfig being missing. (Note that Template:T1 previously said Module:TNT was missing, so I got it from www.mediawiki/w/Module:TNT). JsonConfig seems fine and shows in Special:Version.

I added:

$wgJsonConfigEnableLuaSupport = true;

which gave:

Lua error: bad argument #1 to "get" (not a valid title).

So instead, I put all the following into LocalSettings.php to configure JsonConfig (see [2]):

// Safety: before extension.json, these values were initialized by JsonConfig.php
if ( !isset( $wgJsonConfigModels ) ) {
	$wgJsonConfigModels = [];
if ( !isset( $wgJsonConfigs ) ) { 
	$wgJsonConfigs = [];

$wgJsonConfigEnableLuaSupport = true;


$wgJsonConfigModels['Tabular.JsonConfig'] = 'JsonConfig\JCTabularContent';
$wgJsonConfigs['Tabular.JsonConfig'] = [
	'namespace' => 486,
	'nsName' => 'Data',
	// page name must end in ".tab", and contain at least one symbol
	'pattern' => '/.\.tab$/',
	'license' => 'CC0-1.0',
	'isLocal' => false,

// Enable Tabular data namespace on Commons - T148745
$wgJsonConfigInterwikiPrefix = 'commons';
$wgJsonConfigs['Tabular.JsonConfig']['remote'] = [
	'url' => ''

Then I copied over Module:Documentation/styles.css from, commenting out the background image in line 168. And everything seems to work fine...

I got the following templates from mediawiki (via mediawiki's Special:Export to get dependencies):

  • Template:Colored box
  • Template:Navbox
  • Template:Help box
  • Template:Side box
  • Template:Note

Note that this overwrote Template:Tl, Template:TNT, and others that I resolved issues with previously.

Now a new set of issues has emerged. These include JsonConfig problems (again), and template loops (which I think are coming from Module:Template translation), and missing dependencies (e.g, Template:Mbox and Module:Color contrast). Moreover, most of the templates render the if and other conditional logic statements, rather then executing them. I installed ParserFunctions (which I should have done before) and it solved everything outstanding!

In mediawiki/installs

tar -xzf ParserFunctions-REL1_34-4de6f30.tar.gz -C /var/lib/mediawiki/extensions

Add to LocalSettings.php

wfLoadExtension( 'ParserFunctions' );
$wgPFEnableStringFunctions = true;

The Front Page

To do the front page, I copied the source of Template:Main page from mediawiki to a page (called Test) and created [Template:Main page/styles.css] using mediawiki's code. Then I rejigged the contents of the page!

The only minor but non-obvious change, was that I used h2 headings inside each mainpage_box, rather than h3's. As a consequence, I needed to add the following to Template:Main_page/styles.css:

.mainpage_box h2 {
	border-bottom: none;
.mainpage_box h2 .header_icon {
	margin-right: 5px;

Old instructions[3] suggest using Special:ExpandTemplate on mediawiki's wiki, with the input text {{:MediaWiki}}. But this isn't necessary as the template doesn't need expanding in its current (at the time of writing) incarnation. Naturally, the page works well on MobileFrontend.

Mass Edit

The Mass edit page contains several examples, two of which show "Undefined Control Sequence" errors. These examples use backslashed square brackets (I can't even write them on the wiki using nowiki tags), which have issues because of Extension:SimpleMathjax. Regardless, the extension seems to work just fine!

Update Linux

Get the system up to date:

apt-get update
apt-get upgrade

During the upgrade I chose:

  • keep the local smb.conf
  • keep the local grub (new version in /tmp/grub.l1gqsHmubw)

There were dependcy problem and other warnings during the process. It finished with:

Errors were encountered while processing:
E: Sub-process /usr/bin/dpkg returned an error code (1)
W: Operation was interrupted before it could finish

I rebooted the server. It came up ok and everything seemed fine (I'm doing this over terminal), but it is still claiming that there are updates. The issue might be related to a known Ubuntu bug: The solution(s) might be [4]:

Look at the keyboard config file:

cat /etc/default/keyboard
 # Consult the keyboard(5) manual page.

Remove and reinstall the keyboard-configuration:

apt-get remove keyboard-configuration
apt-get install keyboard-configuration 
 Note that this launches a screen where the only options are Afghani variants... so cancel it!
DEBIAN_FRONTEND=noninteractive apt-get install keyboard-configuration
 This ran and changed the keyboard layout file to the default:
  # Consult the keyboard(5) manual page.
vi /etc/default/keyboard
 I manually removed the af option and the commas, essentially reverting the file (I like a scroll light). 
cd /usr/share/X11/xkb/symbols
ln -s us en
 This was the other solution offered. I was missing an en option, so that might be it.
apt-get install keyboard-configuration
 Now it says that there is nothing to do, which is promising. 

Try the update again:

apt-get update && apt-get upgrade
 Nothing happened and everything seems fine...
shutdown -r now

On boot, the box reports:

Welcome to Ubuntu 18.04.5 LTS (GNU/Linux 4.15.0-45-generic x86_64)
36 packages can be updated.
30 updates are security updates.

Running apt-get upgrade gives:

The following packages have been kept back:
 fwupd fwupdate fwupdate-signed libfwup1 libgl1-mesa-dri libreoffice-avmedia-backend-gstreamer libreoffice-base-core
 libreoffice-calc libreoffice-core libreoffice-draw libreoffice-gnome libreoffice-gtk3 libreoffice-impress libreoffice-math
 libreoffice-ogltrans libreoffice-writer libxatracker2 linux-generic linux-headers-generic linux-image-generic
 python3-software-properties python3-uno software-properties-common software-properties-gtk
0 upgraded, 0 newly installed, 0 to remove and 25 not upgraded.

The latest LTS version (at the time of writing) is 20.04.1 (see So I could do an:

apt-get dist-upgrade

But I should really do a full backup and everything first, so that isn't going to happen today.



Rather than trying to update wordpress, I think it best to install the latest version and use the old dbase. This will likely cause problems with images... but we made several suboptimal choices when we built the last version, including using a non-standard theme and customizing it in a way that prevented updates.

There's a useful wordpress article on the basics of the approach, albeit from a hosted install perspective:

The main install instructions are:

For theme customization beyond that done in the interface, or through a plugin (like Code Snippets [5], wordpress says the best approach is to create a child theme [6].

Choosing the theme ( and is a major decision. This time I want a much more standard theme, that has better plugin and widget support, is responsive and gives a good mobile interface. I also don't want to pay but will have to trade that off against doing customization to make it look distinct.

It seems that Twenty Fifteen has the most active installs, but all the Twenty series, which are default themes made by wordpress are wildly popular. It might be worth using Twenty Twenty, as it is the most recent and takes advantage of the block editor (Twenty Nineteen does too but gets mediocre reviews), and I like the look of Twenty Fourteen.

Outside of the defaults, OceanWP is eCommerce oriented but looks good and is very popular. Neve sits between OceanWP and GeneratePress, which has a more magazine/news focus, and all three take advantage of the new block editor (Gutenberg), which was introduced in WPv5 (initial release in 2018).


Check PHP and MySQL. I need PHP >=7.3 and MySQL >=5.6 but:

php --version
 PHP 7.2.24-0ubuntu0.18.04.6 (cli) (built: May 26 2020 13:09:11) ( NTS )
mysql --version
 mysql  Ver 14.14 Distrib 5.7.31, for Linux (x86_64) using  EditLine wrapper

I might be able to upgrade my version of PhP without upgrading Ubuntu (see However, it is probably a good idea to just fix everything...

Upgrading Linux Distro

So it turns out that I shouldn't have done that last update... I do have an Xwindows Server on the box, running Gnome, and now I can't log in using the GUI on the box itself (it loops back to the login screen). This box doesn't contain the GPUs, just the database server, so the GUI isn't key, but it would be nice to have it working again. Hopefully, an upgrade will fix that, as well as other issues.

Backing off

First, mount the USB drive. Find what's mounted and what the dev is:

mount -t ext4
 (or just mount for everything)
ls -l /dev/disk/by-id/usb*
 (or fdisk -l or lsblk)
mkdir -p /media/usb
mount /dev/sda1 /media/usb

Back up the databases:

psql postgres
As researcher and in /bulk/backups/
 mv lbo_Fc.dump
 pg_dump -Fc allpatentsprocessed > allpatentsprocessed_Fc.dump
 pg_dump -Fc accelerators > accelerators_Fc.dump
 pg_dump -Fc grants > grants_Fc.dump
 pg_dump -Fc incubators > incubators_Fc.dump
 pg_dump -Fc lbo > lbo_Fc.dump
 pg_dump -Fc stockmarket > stockmarket_Fc.dump
 pg_dump -Fc crunchbase3 > crunchbase3_Fc.dump
 pg_dump -Fc vcdb20h1 > vcdb20h1_Fc.dump
 pg_dump -Fc vcdb4 > vcdb4_Fc.dump
mysql -u root -p
 mysqldump --databases --password mcnair > mcnair.sql
 mysqldump --databases --password wordpress > wordpress.sql
 mysqldump --password mcnair > mediawiki.sql

Do the file transfers

mkdir /media/usb/mother-2020-08-09
mkdir /media/usb/mother-2020-08-09/bulk
rsync -av --progress --exclude="mcnair" /bulk/ /media/usb/mother-2020-08-09/bulk/
mkdir /media/usb/mother-2020-08-09/html
rsync -av --progress /var/www/html/ /media/usb/mother-2020-08-09/html/


umount /media/usb

Do the upgrade


apt update 
apt upgrade
apt dist-upgrade
apt autoremove
 if no release found because you are too early, add the -d to allow development (it will still install LTS if that's available)
 do-release-upgrade -d

This failed on the first attempt. So I did:

grep ERROR /var/log/dist-upgrade/main.log
grep BROKEN /var/log/dist-upgrade/apt.log
apt-get remove postgresql-10-postgis-2.4


 do-release-upgrade -d
  I selected some choices (keep smd.conf, don't notify me of whatever, etc.)
  I let it replace postgres10 but it still gave me an "Obsolete Major Version" warning on postgres (I said ok).

Address the upgrade issues

The first casualty of the upgrade was the networking configuration. You'd think that developers would have figured that one out, as remote upgrades would leave boxes DOA until someone could get physical access. Nevertheless, the fix is straight forward.

The old ifup and down and eth0 etc. interface system is gone now, taking its config with it. To get the networking back:

 Outdated now, I think, but it still shows what's up...
ip -a
 This will get you the names of the interfaces etc.
 I already had a .yaml under a different interface name that set up DHCP, so I used it as a template for the interface that I wanted up that way
cp /etc/network/01-netcfg.yaml /etc/network/99_config.yaml
vi /etc/network/99_config.yaml
 change the interface name to eno0 
netplan apply

Now everything looks good for a foundation - apache2 is working, SSH is working, but I need to do a minor config fix for the wiki.

apt-get install php-mbstring
apachectl restart

And the wiki comes back up but with a error notice. The issue seems to be with PHP 7.4, and it looks like it affects both mediawiki and wordpress, though wordpress might have fixed it. Regardless, it is possible to install 7.3 as well, as use that with apache2.

add-apt-repository ppa:ondrej/php
apt-get update
apt-get install php7.3
apt-get install php7.3-cli php7.3-common php7.3-json php7.3-opcache php7.3-mysql php7.3-mbstring  php7.3-zip php7.3-fpm php7.3-intl php7.3-simplexml
 Note we may need to fix some config again as it said: Creating config file /etc/php/7.3/apache2/php.ini
a2dismod php7.4
a2enmod php7.3
 I ignored the following notices for now:
  NOTICE: To enable PHP 7.3 FPM in Apache2 do:
  NOTICE: a2enmod proxy_fcgi setenvif
  NOTICE: a2enconf php7.3-fpm
systemctl restart apache2
update-alternatives --set php /usr/bin/php7.3
update-alternatives --set phar /usr/bin/phar7.3
update-alternatives --set phar.phar /usr/bin/phar.phar7.3
update-alternatives --set phpize /usr/bin/phpize7.3
update-alternatives --set php-config /usr/bin/php-config7.3
error: no alternatives for php-config (ignored for now)

And the wiki now seems happy!

PostGIS Issues

I also checked postgres and everything seemed ok:

su researcher
psql vcdb4
 They are all there
 All my extensions report back.

Update: It seems something did go wrong. Just because the extensions report back doesn't mean they work! When I try to run queries that use PostGIS, I get:

SQL Error [58P01]: ERROR: could not access file "$libdir/postgis-2.4": No such file or directory

I tried updating the extension (I'm pretty sure that I'm running 2.4.3):


But that didn't fix anything. I checked the versions:

select version();
 PostgreSQL 10.14 (Ubuntu 10.14-0ubuntu0.18.04.1) on x86_64-pc-linux-gnu, compiled by gcc (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0, 64-bit
 --So somehow I'm still running version 10!
select PostGIS_full_version();
 ERROR:  could not access file "$libdir/postgis-2.4": No such file or directory
 CONTEXT:  SQL statement "SELECT public.postgis_lib_version()"
 PL/pgSQL function postgis_full_version() line 25 at SQL statement

So I backed off the data from the two dbases that I'd used since the upgrade:

pg_dump -Fc stockmarket > stockmarket_Fc_20201023.dump
pg_dump -Fc vcdb4 > vcdb4_Fc_20201023.dump

The second backoff failed:

pg_dump: [archiver (db)] query failed: ERROR:  could not access file "$libdir/postgis-2.4": No such file or directory
pg_dump: [archiver (db)] query was: SELECT a.attnum, a.attname, a.atttypmod, a.attstattarget, a.attstorage, t.typstorage, a.attnotnull, a.atthasdef, 
a.attisdropped, a.attlen, a.attalign, a.attislocal, pg_catalog.format_type(t.oid,a.atttypmod) AS atttypname, array_to_string(a.attoptions, ', ') AS attoptions, 
CASE WHEN a.attcollation <> t.typcollation THEN a.attcollation ELSE 0 END AS attcollation, a.attidentity, pg_catalog.array_to_string(ARRAY(SELECT 
pg_catalog.quote_ident(option_name) || ' ' || pg_catalog.quote_literal(option_value) FROM pg_catalog.pg_options_to_table(attfdwoptions) ORDER BY option_name), 
    ') AS attfdwoptions FROM pg_catalog.pg_attribute a LEFT JOIN pg_catalog.pg_type t ON a.atttypid = t.oid WHERE a.attrelid = '19998614'::pg_catalog.oid AND 
a.attnum > 0::pg_catalog.int2 ORDER BY a.attnum
Postgres Upgrade Attempt (Failed)

My changes weren't substantial, so I proceeded with an upgrade. First I checked to see if I had postgres12 installed and listening on another port or not:

locate postgres
ls /usr/bin/postgres
dpkg --get-selections | grep postgres
 Ver Cluster Port Status Owner    Data directory              Log file
 10  main    5432 online postgres /data/postgres              /var/log/postgresql/postgresql-10-main.log
 12  main    5433 online postgres /var/lib/postgresql/12/main /var/log/postgresql/postgresql-12-main.log
pg_upgradecluster 10 main
 #This failed: 
 pg_dump: error: query failed: ERROR:  could not access file "$libdir/postgis-2.4": No such file or directory

So... I can't automatically upgrade without first fixing the issue with v10 and postgis.

 But that put the following into /etc/apt/sources-list:
  deb focal main
vi it to (see
 deb focal-pgdg 
wget --quiet -O - | sudo apt-key add -
apt-get update
 #Throws a warning:
 N: Skipping acquire of configured file 'main/binary-i386/Packages' as repository ' focal-pgdg InRelease' doesn't support architecture 'i386'

But the wretched thing still doesn't seem to be available:

apt-get install postgresql-10-postgis-2.4
 Package postgresql-10-postgis-2.4 is not available, but is referred to by another package.

Trying a manual approach. Get the file, put it in /bulk/temp and cd there. Then:

apt-get install ./postgresql-10-postgis-2.4_2.4.3+dfsg-4_i386.deb
 #This failed too - there are unmet dependencies and they are 'not installable'.
Switching over the installations

So, I took the alternative approach of changing the data folders [7].

The plan:

  1. Take version 10 offline
  2. Move version 10's data to a new location (/var/lib/postgresql/10/main)
  3. Switch the ports of versions 10 and 12
  4. Move version 12's data to /data
  5. Put version 12 online
  6. Load up the data in version 12!
  7. Optionally wipe out the old installation

Shut it down:

pg_ctlcluster 12 master start
 #The cluster 
 Ver Cluster Port Status Owner    Data directory                Log file
 10  main    5432 online postgres /data/postgres                /var/log/postgresql/postgresql-10-main.log
 12  master  5433 online postgres /var/lib/postgresql/12/master /var/log/postgresql/postgresql-12-master.logg
systemctl stop postgresql
systemctl status postgresql

Edit the config files:

vi /etc/postgresql/10/main/postgresql.conf
 data_directory = '/var/lib/postgresql/10/main'
 port = 5433 
vi /etc/postgresql/12/master/postgresql.conf
 data_directory = '/data/postgres'
 port = 5432
 listen_addresses = '*'
 #While we are here do some performance tuning:
  shared_buffers = 512MB
  huge_pages = try                        
  temp_buffers = 8G
  work_mem = 4GB 
  maintenance_work_mem = 64
  effective_cache_size = 384GB 
  #Note that I didn't reduce the number of connections (and the max_wal_senders, which must be < max connections), or change max_stack_depth (which gives an error if you set it too high)
 vi /etc/postgresql/12/master/pg_hba.conf 
  Copy over the config to allow access from inside the network

Move the data:

 #to check diskspace
rm -R /var/lib/postgresql/10/main
 #Note that none of the config files in here were valid (though you should check this is true before you do it!)
rsync -av /data/postgres/ /var/lib/postgresql/10/main
 #Takes awhile, but make sure it is all done before the next step
rm -R /data/postgres
rsync -av /var/lib/postgresql/12/master/ /data/postgres
systemctl start postgresql
 Ver Cluster Port Status Owner    Data directory              Log file
 10  main    5433 online postgres /var/lib/postgresql/10/main /var/log/postgresql/postgresql-10-main.log
 12  master  5432 online postgres /data/postgres              /var/log/postgresql/postgresql-12-master.log

Do the installs for some extensions:

apt-get install postgresql-12-plr
apt-get install postgresql-plperl-12

Check it all works:

psql postgres
                                    List of installed extensions
 Name   | Version |   Schema   |                             Description


plpgsql | 1.0     | pg_catalog | PL/pgSQL procedural language
postgis | 3.0.0   | public     | PostGIS geometry, geography, and raster spatial types and functions
(2 rows)

Make the user:

createuser --interactive researcher

Then restore the databases (as researcher in /bulk/backup):

createdb stockmarket
pg_restore -Fc -d stockmarket stockmarket_Fc_20201023.dump
createdb vcdb4 
pg_restore -Fc -d vcdb4 vcdb4_Fc.dump

The restore threw some errors related to not having extension plpythonu, but otherwise seemed fine. The issue seems to be that pythonu is python2[8], and python2 is not available for postgres 12 (it might be here:

apt-cache search ".*plpython*.*"

Other Fixes

Remove redundant user accounts:

cat /etc/passwd
userdel -r username

I need to get Xwindows set up again. My best guess as to the cause of this issue is leftover Nvidia drivers from my attempts to install the GPUs on this box went bad in an earlier apt-get upgrade but I can't see them listed:

dpkg -l | grep nvidia-driver

There is a .Xauthority file, and an .ICEauthority file, in /home/ed and both are owned by ed:ed. The former is empty (0 bytes) and the latter has some non-UTF8 (I think?) characters in it. I'm not sure if either is an issue.

I didn't see xserver-xorg-video-nouvea in the package list or any video driver module, so I installed nouveau:

dpkg -l
lsmod | more
apt install xserver-xorg-video-nouveau
 I'm not sure if I should be fixing my boot image or not...
shutdown -r now
lsmod | more

After doing this the login would give a local desktop but neither the keyboard nor mouse worked. I tried uninstalling and reinstalled the keyboard-configuration again.

apt-get remove keyboard-configuration
apt-get install keyboard-configuration  
shutdown -r now

But that just put me back where I was: with a login loop problem. So I tried switching to lightdm:

apt-get install lightdm

And it worked even before a reboot. After a reboot, I had a different login screen but the actual desktop looked the same. The .Xauthority file is now 51 bytes big and I suddenly have a .xsession-errors, which contains a list of environment settings taking place... However, the machine then silently crashed that night and again the following morning. I couldn't find a specific cause in the logs but there did seem to be a number X and GNOME problems:

journalctl -b -1
journalctl --since "1 hour ago"

I ran an update from the GUI, which may have helped. However, there was a warning about an issue with a screensaver the first time that I loaded lightdm, and the crashes seemed to happen sometime after a clean boot. So I uninstalled lightdm, and installed gdm (which failed as installed already) and rebooted but got no GUI. Then I uninstalled and reinstalled gdm and everything seems fine now.

apt-get remove lightdm
apt-get remove gdm3
apt-get install gdm3

Incidentally, I left a clock running in a terminal so that I could see when the box went down if it crashed again. The clock code is:

while [ 1 ] ; do echo -en "$(date +%T)\r" ; sleep 1; done

Important Moves

I kept the old versions of mediawiki and wordpress and moved them to /bulk/retired (using yyymmdd dates)

mv /var/lib/mediawiki26082020 /bulk/retired/
mv /bulk/retired/mediawiki26082020 /bulk/retired/mediawiki20200826
mv /var/www/html/blog20200809 /bulk/retired/

Wordpress Redux


First, move the old folder to a new name, so that it is there for backup and then get the new install and unpack it.

cd /bulk/installs
mv /var/www/html/blog /var/www/html/blog20200809
tar -xzf latest.tar.gz -C /var/www/html/
cd /var/www/html/
mv wordpress/ blog/
chown -R www-data:www-data blog

Put an .htaccess file in that folder to restrict access while we work:

vi blog/.htaccess
 Require ip

Set up

Then set up the dbase by editing wp-config.php (it's easiest to modify the sample).

cp blog/wp-config-sample.php blog/wp-config.php
vi blog/wp-config.php
 Note get some keys from:

Then the backend works - go to! However the health check shows a missing required module and two missing recommended modules. Fix that:

apt-get install php7.3-gd
apt-get install php7.3-curl
apt-get install php7.3-imagick
apachectl restart

Ironically, it then recommends that I upgrade to PHP7.4... but that would just give issues for mediawiki. On the other hand, everything is now green and just 4 groups of recommendations remain.


See Wordpress Blog Site (Tool) for the McNair Center's build.

Using I configured the blog as follows:

  • Select Twenty Twenty as the theme
  • Add the permalink code to the .htaccess file, so that the URLs will work with postnames
  • Copy over images to wp-content/uploads (use cp -a to maintain permissions)
  • Change the site name to https (after fixing the https setup, see below)

Install plugins:

  • Yoast SEO
  • Wordfence Security
  • Disable Comments
  • Site Kit by Google (set up once live!)
  • Pixabay

I also added:

  • CoBlocks (free)
  • Advanced Gutenberg (free)
  • Otter

I didn't add Co-Authors Plus ( as it hasn't been tested on the latest version of wordpress. There are other plugins that offer equivalent functionality if I need one later.

Other plugins I might want are:

  • Revive Old Post (share with twitter)
  • Optimole (optimize images)
  • WP Rocket (implement cache)


Hardening Wordpress

I hardened the wordpress installation:

This included:

  • Fixing file ownership: For fully hardened, change ownership of everything to root, except wflogs, uploads and themes in wp-content, which should be owned by www-data. However, then you won't be able to install plugins etc. A compromise is -R root:root for blog and then www-data:www-data for wp-content.
  • Check file permissions: Everything is 644, except wp-content which is 755
  • Checking dbase rights and setting new passwords.
  • Changing passwords on old accounts (with posts, so the accounts shouldn't be deleted) to random strong strings.
  • Fixing up .htaccess file to impose restrictions
  • Install Sucuri
  • Enable more logging

Checking user rights in the dbase and changing their password:

mysql -user=root -p
use wordpress
SELECT User FROM mysql.user;
SHOW GRANTS FOR 'username'@'localhost';
SET PASSWORD FOR 'username'@'localhost'='newpassword';
 (Note that this shouldn't be logged in clear on the server, but might be on a client. Delete .mysql_history at the end of your session.)

.htaccess in wp-includes:

# Block the include-only files.
<IfModule mod_rewrite.c>
RewriteEngine On
RewriteBase /
RewriteRule ^wp-admin/includes/ - [F,L]
RewriteRule !^wp-includes/ - [S=3]
RewriteRule ^wp-includes/[^/]+\.php$ - [F,L]
RewriteRule ^wp-includes/js/tinymce/langs/.+\.php - [F,L]
RewriteRule ^wp-includes/theme-compat/ - [F,L]
# BEGIN WordPress

The #BEGIN WordPress tag is redundant as the file is 644 root.

Add the following to .htaccess in the wordpress dir:

<files wp-config.php>
order allow,deny
deny from all

If there are plugin installation issues then add to wp-config.php


Once I'm all done with the theme etc., I can uncomment the following from wp-config.php

define('DISALLOW_FILE_EDIT', true);
With hardened permissions, you won't be able to update Wordpress from the dashboard. To fix this, set ownership of the entire wordpress directory to www-data:www-data (i.e., chown -R www-data:www-data blog/), run the update, and then revert the ownership to root (or some other account).


I built a Branding palette, to standardize the colors. And I installed the Twentig plugin, to give extra configuration options.

I changed the site colors, added the logo and the tag line, and made other config changes.

Then, I added custom CSS as follows.

To reduce the header spacing:

.header-inner {
    padding: 1.5rem 0;

To remove the title from the landing page:

.page-id-2169 .entry-title{
    display:none !important;
.page-id-2169 .entry-header {
    padding: 0;

To do:

  • I need to add social media icons! That might be as easy as adding the social media menu [9].
  • Get a related posts widget? There's Yet Another Related Posts Plugin, Contextual Related Posts, and Inline Related Posts... I went with YARPP, as it is the most popular. It is apparently resource-heavy.

I tried the following blocks plugins:

  • Ultimate Addons for Guttenberg
  • It's free and adds some nice basic functionality
    • Post blocks include: Post Carousel, Post Grid, Post Masonry, Post Timeline, Advanced Columns -- but customization is limited and I can't do one post
  • Getwid:
    • Pretty highly customizable.
    • Can specify which posts to show in 3 blocks (Custom Post Type, Post Carousel, and Post Slider) and can build custom templates to arrange how they are displayed
    • Post blocks: Recent Posts, Custom Post Type, Post Carousel, and Post Slider
  • Redux -- It's a templates library. You get 5 for free and they upsell hard.
  • ZeGuten - Couldn't find it
  • Advanced Gutenberg - It's free and widely used...
  • CoBlocks -- Does the basics
    • Posts -- Can't specify specific posts. Can do category.
    • Post Carousel -- Likewise.
  • Stackable:
    • It requested opt-in, which I didn't like, and it wants you to 'Go Premium'.
    • It has settings for everything! By far the most detailed configuration.
    • Useful blocks:
      • Posts -- can't seem to specify a specific post
      • Advanced Columns and Grids -- for layout
      • Card -- could make posts links with buttons
      • Feature/Feature Grid -- likewise
      • Container? Might be helpful
  • Gutenberg Post Blocks
    • Untested with my version. Seems to work.
    • Has lots of options but does full-page things. Can limit to a post using include but has next page links...
    • Tried to push for an update to pro.
  • Magical Posts Display -- I dumped it for being too weird.
  • Otter Blocks
    • Google maps block and other useful things... I just don't need it right now.


  • Latest Posts (widget)

Chosen block plugins:

  • Getwid -- It's outstanding and embraces templates for serious bespoke customization
  • Stackable -- For its option-based customization
  • I might add back coblocks, Advanced Gutenberg and Ultimate Addons for Gutenberg


I installed WP Mail SMTP Lite.

I first set it up to use Google. Essentially you need to sign in to Google and set up an API in the console: However, this seemed to introduce a massive security hole unless you have G Suite, so I abandoned this approach.

I had previously set up SMTP through Google for the wiki (See Research_Computing_Configuration#Confirm_Account). So, I used the same approach with Wordpress. In WP Mail SMTP Lite choose 'Other' (see the second method). Then edit wp-config.php to hardcode the values (this ensures that the password, which is stored plain-text, is a little more secure.):

define( 'WPMS_ON', true ); //You MUST set this if you want hardcoded values to work!
define( 'WPMS_LICENSE_KEY',  ); 
define( 'WPMS_MAIL_FROM', '' );
define( 'WPMS_MAIL_FROM_FORCE', true ); 
define( 'WPMS_MAIL_FROM_NAME', 'The Blog at' );
define( 'WPMS_MAIL_FROM_NAME_FORCE', true ); 
define( 'WPMS_MAILER', 'smtp' ); // Possible values: 'mail', 'gmail', 'mailgun', 'sendgrid', 'smtp'.
define( 'WPMS_SET_RETURN_PATH', true ); 
define( 'WPMS_SMTP_HOST', 'ssl://' ); 
define( 'WPMS_SMTP_PORT', 465 ); 
define( 'WPMS_SSL', 'ssl' ); // Possible values , 'ssl', 'tls' - note TLS is not STARTTLS.
define( 'WPMS_SMTP_AUTH', true ); 
define( 'WPMS_SMTP_USER', '' ); // SMTP authentication username, only used if WPMS_SMTP_AUTH is true.
define( 'WPMS_SMTP_PASS', 'password generated by Google' );
define( 'WPMS_SMTP_AUTOTLS', true );
Author Comments

The blog supports multiple authors and by default, Wordpress emails an author whenever one of their posts gets a comment. If you'd like to disable author comment emails but keep the moderator emails, there's a simple fix:

Just go to wp-admin/options.php and set 'comments_notify' to 0. (See

More complicated methods involve writing your own plugin [10] to refine wp_new_comment_notify_postauthor[11] or changing the hooks[12] used in wp-includes/comment.php:

$maybe_notify = apply_filters( 'notify_post_author', $maybe_notify, $comment_ID );

Social Media Integration

Getting the social media icons on the menu and correctly linked up is very straight forward. Follow the guide for twenty-fifteen, which also works for 2020.

Getting some share buttons was more problematic, particular as my planned social media usage is somewhat atypical (Twitter, LinkedIn, and Reddit, really in reverse order), and because I don't want to pay anything.

The free version of Revive Old Posts lets you push content to Twitter and Facebook, but they want you pay to push to LinkedIn.

The best free options seem to be:

  • AddToAny Share Buttons - Integrates with Google Analytics
  • Simple Social Icons - The simplest option
  • Shared Counts -- Counts hits (but using a 3rd party for data?)
  • WordPress Social Login - if you want users to log in using their SM accounts (note: has a bimodal ratings distro)
  • JetPack -- The plugin used by for this functionality. The free version should suffice, but this thing is a monster. It also uses an account on the cloud, which is a pain for those who are self-hosting.

I went with AddToAny, as it had the most installations, is entirely open-source, and offers all the functionality I need.

Avoiding JetPack

I tried to add a profile picture, but by default, WordPress uses Gravitar, which, surprise, surprise, links to your account... and to add a self-hosted site, you have to install JetPack. At this point, I felt harassed and doubly so because I didn't install JetPack and yet, some how, the profile picture correctly updated from the one I'd posted on Gravitar. What the hecK?


I used Site Kit plugin for wordpess, and for mediawiki I made a sitemap to submit to Google.


In mediawiki:

mkdir sitemap
php maintenance/generateSitemap.php --memory-limit=50M --fspath=/var/www/html/mediawiki/sitemap/ --identifier=edegancom --urlpath=/sitemap/ --server= --compress=yes

Then submit it to Google... I did this by making an alias in apache2.conf from sitemap to /var/www/html/mediawiki/sitemap/, then submitting
#in retrospect, I wish I'd used an identifier with 'wiki' in it but what the hey.

And with that success behind you, install Google XML Sitemaps on Wordpress, chose some settings (on Settings -> XML-Sitemap), and then post the URL to Google:

It seems Yoast already builds a sitemap, you just need to submit to it Google... (I uninstalled XML Sitemaps):


To set up HTTPS using Let's Encrypt, see

Install it and make some directories...

apt update
apt install certbot
openssl dhparam -out /etc/ssl/certs/dhparam.pem 2048
 takes ~20 secs
mkdir -p /var/lib/letsencrypt/.well-known
chgrp www-data /var/lib/letsencrypt
chmod g+s /var/lib/letsencrypt

Set up the config files

vi /etc/apache2/conf-available/letsencrypt.conf
Alias /.well-known/acme-challenge/ "/var/lib/letsencrypt/.well-known/acme-challenge/"
<Directory "/var/lib/letsencrypt/">
    AllowOverride None
    Options MultiViews Indexes SymLinksIfOwnerMatch IncludesNoExec
    Require method GET POST OPTIONS
vi /etc/apache2/conf-available/ssl-params.conf
SSLProtocol             all -SSLv3 -TLSv1 -TLSv1.1
SSLHonorCipherOrder     off
SSLSessionTickets       off

SSLUseStapling On
SSLStaplingCache "shmcb:logs/ssl_stapling(32768)" 

SSLOpenSSLConfCmd DHParameters "/etc/ssl/certs/dhparam.pem"  

Header always set Strict-Transport-Security "max-age=63072000"

Enable some apache2 mods!

a2enmod ssl
a2enmod headers
a2enconf letsencrypt
a2enconf ssl-params
sudo a2enmod http2
systemctl reload apache2

Run certbot!

certbot certonly --agree-tos --email --webroot -w /var/lib/letsencrypt/ -d -d
Note that I needed an @ entry in my A record for pointed to my IP address to get the main challenge to succeed.

Then set up a new apache2 config file (in /etc/apache):

mv sites-available/000-default.conf sites-available/000-default.conf.bak
vi sites-available/
 <VirtualHost *:80>
       Redirect permanent /
<VirtualHost *:443>
       Protocols h2 http/1.1
       DocumentRoot /var/www/html
       ErrorLog ${APACHE_LOG_DIR}/error.log
       CustomLog ${APACHE_LOG_DIR}/access.log combined
       SSLEngine On
       SSLCertificateFile /etc/letsencrypt/live/
       SSLCertificateKeyFile /etc/letsencrypt/live/
       # Other Apache Configuration
       Alias /wiki /var/www/html/mediawiki/index.php
       RewriteEngine On
       RewriteRule ^/*$ %{DOCUMENT_ROOT}/mediawiki/index.php [L]
ln -s sites-available/ sites-enabled/
systemctl reload apache2

Test it by going to

Finally, edit /etc/cron.d/certbot and append the following to the last line (after -renew):

--renew-hook "systemctl reload apache2"
certbot renew --dry-run
 Tests the renewal!

PDFEmbed Issue

Enabling and requiring HTTPS causes an issue with PDFEmbed on mediawiki, where you get a blank frame. The PDF is still there, other images load fine, but the PDF frame won't render the PDF. The problem is actually that the PDF is served with HTTP and the rest of the page is served with HTTPS, and Chrome (and perhaps other browsers) don't render the insecure content as a consequence (see [13] for a description of the symptoms, but not the solution.).

The solution is to edit mediawiki/extensions/PDFEmbed/PDFEmbed.hooks.php. For me it was line 103 that previously said:

'src' => $file->getFullUrl().'#page='.$page,

I changed this line to:

'src' => preg_replace("/^http:/i", "https:", $file->getFullUrl()).'#page='.$page,

This is mentioned in a comment on a topic page, though presumably for an earlier version:

Another Issue

Interestingly, I started getting a message from Google Chrome whenever I went to post wiki entries saying: "The information you’re about to submit is not secure". There's an option to "Proceed anyway" or "Go back".

This started after I had MultiTail running viewing apache's logs, but I couldn't see, beyond some kind of file lock examination, how it could work. I figured that it was a coincidence and something else might have happened.

My first thought was that my SSL certificate might have expired. However, the certificate looks valid and good, and the issue survived a reboot.

By inspecting the webpages (in Chrome) and then reviewing the Console, I could see that it was caused by a mixed content problem:

Mixed Content: The page at '<URL>' was loaded over HTTPS, but requested an insecure font '<URL>'. 
This request has been blocked; the content must be served over HTTPS.

It seemed that I somehow have some font addresses hardcoded somewhere:

Mixed Content: The page at ... was loaded over HTTPS, but requested an insecure font 
This request has been blocked; the content must be served over HTTPS.

The copy of Chrome on my desktop must somehow have been upgraded? Or something else changed to cause a change in behavior...

The IP is from the old web server at the McNair Center, suggesting that when I migrated the McNair database into the new wiki, I migrated this issue. (Note that it doesn't appear to be something hardcoded into a .css file, or similar -- I can't find any trace on the filesystem and besides, this wiki was built from a fresh install.)

I found the URLs hardcoded in MediaWiki:Common.css (it must have been moved with the last big batch of pages and I somehow didn't notice!) but then couldn't edit it! It seems that following Mediawiki 1.32, the rights to edit the interface were separated out, and users now need the editinterface right to change anything in the Mediawiki namespace. So, I went to Special:UserRights and gave myself permission. Then I edited the page, which changed the look-and-feel of my editor (I have no idea why), removed the consol messages, but left the problem (even after ctrl-shift-r cache flush on Chrome).

Install VSFTPD

With the security restrictions on wordpress, I now need an FTP server to get files for themes, plugins, etc. I like VSFTPD, as its simple, secure, and has a nice standalone config. Old documentation on an earlier install on the old Wordpress Blog Site (Tool) page. Instructions are here:

apt-get install vsftpd
cp /etc/vsftpd.conf /etc/vsftpd.conf_orig
vi /etc/vsftpd.conf
 #Change the following
 #Add the following (forces ssl)

/etc/init.d/vsftpd restart

Then add a user and set it up:

useradd -m blog
passwd blog
usermod -a -G www-data blog
usermod -d /var/www/html/blog blog

Test it:


See also:

To address some of the issues with the FTP server's file permissions in wordpress add to wp-config.php:

define( 'WP_CONTENT_DIR', 'wp-content' );
define( 'FTP_BASE', '/var/www/html/blog/' );

If I chmod blog:blog /var/www/html/blog then everything seems to work find when I sftp but wordpress is unable to create a directory... I can't work out why this is happening. I expect it has to do with the need for another wordpress specific define() statement, but I'm spending too much time on it. So I'm going to use direct installation of plugins instead, and remove the FTP server as it is a point of vulnerability.

apt-get remove vsftpd
userdel blog

Final Configuration Changes to Apache

Lock down apache somewhat further (as now there are directories that shouldn't be listable, etc.)

cd /etc/apache2
vi apache2.conf
 #Change the directory definitions. Notes that if -SomeOption is used then other options must have + or - in front of them:
 <Directory /var/www/html>
       Options -Indexes +FollowSymLinks
       AllowOverride All
       Require all granted

systemctl reload apache2 
#To debug:
 systemctl status apache2.service

Remove the debug setup

In the wiki (LocalSettings.php), comment the debug lines (I can't see when I added them from the documentation, but if you want to see error messages during the config, you'd want them uncommented):

#error_reporting( -1 );
#ini_set( 'display_errors',1 );
#$wgShowExceptionDetails = true;
#$wgShowDBErrorBacktrace = true;
#$wgShowSQLErrors = true;

Check the permissions set using $wgGroupPermissions - see

Run all the updates to the blog, etc., from the consol before locking it down. Then in wp-config.php, lock down the ability to install plugins, etc., by commenting:


Edit the .htaccess files in blog and mediawiki to allow access but with appropriate restrictions.

Note that the rewrite rules for the blog are in its .htaccess file

<IfModule mod_rewrite.c>
RewriteEngine On
RewriteBase /blog/
RewriteRule ^index\.php$ - [L]
RewriteCond %{REQUEST_FILENAME} !-f
RewriteCond %{REQUEST_FILENAME} !-d
RewriteRule . /blog/index.php [L]

To make the blog the default, edit /etc/apache/, add an alias (don't alias to index.php as it will cause design issues, the rewrite rule for that is already in the .htaccess file!):

Alias /blog /var/www/html/blog/

And change:

RewriteRule ^/*$ %{DOCUMENT_ROOT}/mediawiki/index.php [L]


RewriteRule ^/*$ %{DOCUMENT_ROOT}/blog/index.php [L]


systemctl reload apache2

Note: Don't change the DocumentRoot to the blog, as this will destroy the design of the wiki. The last rewrite rule will decide the default site!

Changing the Wordpress url

It seems likely that some Reddit bots are automatically blocking postings with the word blog in their URL. So I decided to move my Wordpress installation from /blog to /journal.

Note that I briefly tried 'article' (singular to save letters). I also considered 'paper', which is shorter but not quite right. Likewise study, etc. Also digest (which was a close second), review, bulletin, and pub (which was just too ambiguous). The problem with article is that although it looks good for article URLS, the landing URL is then A collection word is better. And journal appeals to the academic in me.

To do this takes about 10 minutes (see You have to:

  1. Make the change in Wordpress through wp-admin (you have to do this first!) - Set both the Wordpress Address (URL) and Site Address (URL) fields
  2. Move the directory to the new name (at this point I could access
  3. Fix aliases in apache and the default subdomain for landing (see below)
  4. Fix the permalinks (update the .htaccess file)
  5. Fix the links hardcoded in menus
  6. (Create and) load up a new icon image (it is set under Appearance -> Customize -> Site Identity)
  7. Run any updates etc.
  8. Reconnect Google site kit
  9. Rerun Yoast SEO optimizer

To fix the main alias in apache:

vi /etc/apache2/sites-available/
  Alias /blog /var/www/html/blog/
  RewriteRule ^/*$ %{DOCUMENT_ROOT}/blog/index.php [L]
systemctl reload apache2

I also needed to:

  1. Perform a backup (I really should have done this before moving the site but hey...)
  2. Fix the images etc. (they initially fine... but I came to suspect that they were cached and an HTML inspected revealed the wrong URLs)

Do the backup (with a user that has process privileges):

cd /bulk/backups
mysqldump --add-drop-table -h localhost -u user -p wordpress | gzip > wordpress-20210209.sql.gz
mysqldump --add-drop-table -h localhost -u user -p wordpress > wordpress-20210209.sql

Then install and use Velvet Blues (see and use it to change the URLs (use https:// as a page inspection shows that this is correct)

Useful tools


I installed Multitail:

apt-get multitail

The manual is pretty weak, but the examples are good and the feature list is excellent. Here's some useful commands to review log files:

 multitail -cS apache -ev "Bot" /var/log/apache2/access.log -ci white -e "Bot" -I /var/log/apache2/access.log
 multitail -cS apache -ev "Bot" -ev "bot" -ev "internal dummy connection" /var/log/apache2/access.log


apt install traceroute

Note: Zmap seems popular nowadays, based on traffic logs.


Connect with smb from Mother to Father:

smbclient // -U Domainname/username
Note that you need to specify the domain.

Mount a thru connection:

mount -t cifs -o user=username // /mnt/father
ln -s /mnt/father/whatever/ /bulk/whatever

Old machines

For the configuration of the servers built for the McNair Center, see the old Center IT page or the pages below:

Some of this information is still useful!

In addition, at UC Berkeley, Ed designed and built three machines - two postgresql database servers and a wiki server. The documentation is here: