Bart van Heukelom's questions

The past few days I've been setting up a used Dell PowerEdge 2950 server. It's been running fine until just now, when it suddenly stopped while running a fairly heavy task (svnadmin verify). The LCD was and is showing the following message:

PowerEdge 2950 E1422 CPU Machine Chk E2118 Fatal NB Mem CRC

Now, the memory error E2118 seems straightforward. One of the banks is probably broken. I'm running memtest right now. Update: Well, memtest didn't find any errors, so it's not that easy.

E1422, less so. Is this likely a separate problem, or just a result of E2118? Googling for this code says "update the BIOS" which isn't very specific as to the cause of the problem.

In this part of a Chef recipe, there are 4 execute resources. They should only run when a file does not exists. This is added as a condition to the first resource, which then triggers the entire chain to be run.

# Set up default SSL cert
execute "defaultcert1" do
    not_if {File.exist?("/vol/webserver/cert")}
    command "mkdir /vol/webserver/cert"
    notifies :run, "execute[defaultcert2]", :immediately
end
execute "defaultcert2" do
    action :nothing
    command "ln -s /etc/ssl/certs/ssl-cert-snakeoil.pem /vol/webserver/cert/server.crt"
    notifies :run, "execute[defaultcert3]", :immediately
end
execute "defaultcert3" do
    action :nothing
    command "ln -s /etc/ssl/private/ssl-cert-snakeoil.key /vol/webserver/cert/server.key"
    notifies :run, "execute[defaultcert4]", :immediately
end
execute "defaultcert4" do
    action :nothing
    command "chown -R ubuntu:ubuntu /vol/webserver/cert"
end

As you can see, it's a rather large bit of code to guard just 4 commands. Is there a better way to do this?

I use the following pseudo-script to create a TAR of my installed software

mkdir tmp
ln -s /path/to/app1/bin              tmp/app1
ln -s /and/path/going/to/the-app-2   tmp/app2
tar -c --dereference -f apps.tar tmp

I need the --dereference option here to follow the links I just made in tmp. The reason I make the links in the first place is to store the directories with a different name in the archive than they have on the filesystem.

Until now it has worked fine. However, I now have the situation that /path/to/app1 also contains links, and those I don't want to follow.

Is this possible with some changes to the tar command? Or do I need to completely switch around the way I build the archive?

Context:

I'm automatically installing postgresql-9.1 on an Ubuntu server with apt-get. This creates the required postgres user.

The Postgres data is on an external volume that survives reinstalls. This data is obviously owned by the postgres user.

The problem I'm having is that the ownership is not recorded under the name postgres, but under the UID that postgres had at creation time. When the server is reinstalled, postgres sometimes gets a different UID, and no longer owns the data directory, and thus does not work.

Question:

Can I force the UID of the user postgres created by apt-get to something fixed? Or is there another way to solve my problem?

(As you may have deduced, this is on Amazon EC2 with the data on an EBS volume)

Is it possible to seed ssh-keygen with a fixed string so that it consistently generates the same keypair?

root@mountain-lion:/opt/smartfox# ls -lha
total 180K
drwxr-xr-x 8 root root 4.0K 2012-06-01 14:09 .
drwxr-xr-x 4 root root 4.0K 2012-06-01 09:41 ..
drwxr-xr-x 8 root root 4.0K 2009-05-17 21:57 lib
lrwxrwxrwx 1 root root   22 2012-06-01 09:41 logs -> /var/opt/smartfox/logs
-rwxr-xr-x 1 root root 1.4K 2012-06-01 14:28 run.sh


root@mountain-lion:/opt/smartfox# cat run.sh 
#!/bin/bash    
java -cp "./:./sfsExtensions/:lib/activation.jar:lib/commons-beanutils.jar:lib/commons-collections-3.2.jar:lib/commons-dbcp-1.2.1.jar:lib/commons-lang-2.3.jar:lib/commons-logging-1.1.jar:lib/commons-pool-1.2.jar:lib/concurrent.jar:lib/ezmorph-1.0.3.jar:lib/h2.jar:lib/js.jar:lib/json-lib-2.1-jdk15.jar:lib/json.jar:lib/jsr173_1.0_api.jar:lib/jysfs.jar:lib/jython.jar:lib/nanoxml-2.2.1.jar:lib/wrapper.jar:lib/xbean.jar:lib/javamail/imap.jar:lib/javamail/mailapi.jar:lib/javamail/pop3.jar:lib/javamail/smtp.jar:lib/jetty/jetty.jar:lib/jetty/jetty-util.jar:lib/jetty/jstl.jar:lib/jetty/multipartrequest.jar:lib/jetty/servlet-api.jar:lib/jetty/standard.jar:lib/jsp-2.1/commons-el-1.0.jar:lib/jsp-2.1/core-3.1.0.jar:lib/jsp-2.1/jsp-2.1.jar:lib/jsp-2.1/jsp-api-2.1.jar:lib/jsp-2.1/jstl.jar:lib/jsp-2.1/standard.jar:lib/lsc.jar:lib/commons-io-1.4.jar" \
it.gotoandplay.smartfoxserver.SmartFoxServer > logs/smartfox.out 2>&1 &
JAVAPID=$!
echo "Started Smartfox. JVM PID = $JAVAPID"
trap "echo Stopping Smartfox.; kill $JAVAPID" INT TERM
wait
echo "Smartfox stopped."


root@mountain-lion:/opt/smartfox# start-stop-daemon --start --make-pidfile --pidfile /var/opt/smartfox/smartfox.pid --exec ./run.sh
start-stop-daemon: unable to start ./run.sh (No such file or directory)

Why can't start-stop-daemon find the script?

When trying to start PostgreSQL, I get this 'empty' error message:

root@mountain-lion:/home/ubuntu# service postgresql start
 * Starting PostgreSQL 9.1 database server
 * Error: could not exec /usr/lib/postgresql/9.1/bin/pg_ctl /usr/lib/postgresql/9.1/bin/pg_ctl start -D /vol/postgres -l /var/log/postgresql/postgresql-9.1-main.log -s -o  -c config_file="/etc/postgresql/9.1/main/postgresql.conf" : 

This is a freshly installed Ubuntu 11.10 server (on EC2 / Scalarium), with postgresql-9.1 just installed via apt-get.

There's one change configuration value in postgresql.conf:

data_directory = '/vol/postgres'

/vol/postgres looks like this

drwx------ 5 postgres postgres   38 2012-05-18 15:13 base
drwx------ 2 postgres postgres 4096 2012-05-18 15:13 global
drwx------ 2 postgres postgres   17 2012-05-18 15:12 pg_clog
drwx------ 4 postgres postgres   34 2012-05-18 15:12 pg_multixact
drwx------ 2 postgres postgres   17 2012-05-18 15:13 pg_notify
drwx------ 2 postgres postgres    6 2012-05-18 15:12 pg_serial
drwx------ 2 postgres postgres   24 2012-05-18 15:13 pg_stat_tmp
drwx------ 2 postgres postgres   17 2012-05-18 15:12 pg_subtrans
drwx------ 2 postgres postgres    6 2012-05-18 15:12 pg_tblspc
drwx------ 2 postgres postgres    6 2012-05-18 15:12 pg_twophase
-rw------- 1 postgres postgres    4 2012-05-18 15:12 PG_VERSION
drwx------ 3 postgres postgres   58 2012-05-18 15:12 pg_xlog
-rw------- 1 postgres postgres  133 2012-05-18 15:13 postmaster.opts
lrwxrwxrwx 1 root     root       36 2012-05-18 15:13 server.crt -> /etc/ssl/certs/ssl-cert-snakeoil.pem
lrwxrwxrwx 1 root     root       38 2012-05-18 15:13 server.key -> /etc/ssl/private/ssl-cert-snakeoil.key

Any idea?

Update: When I set data_directory to the default (/var/lib/postgresql/9.1/main) it works fine, so there must be something wrong in that specific change.

I have this script line to install all packages for a new server.

apt-get install mysql-server openjdk-7-jre-headless tomcat7 tomcat7-admin jsvc apache2 ntp subversion

As you can see it installs Java 7, but Java 6 is also installed because tomcat7 (indirectly) depends on it. I'm pretty sure I can run Tomcat 7 with Java 7, so I don't want to install two JREs. How can I prevent Java 6 from being installed?

Update: Closer inspection shows that

• tomcat7, via tomcat-common, depends on default-jre-headless | java6-runtime-headless | java6-runtime | java-6-runtime (assuming the pipes mean "one of").
• openjdk-7-jre-headless provides java6-runtime-headless, so that dependency should be met.
• Instead, default-jre-headless is installed, which installs Java 6.

Why does that happen? Is it because I install them all at once. Should I split it up into 2 calls?

(Ubuntu 12.04 server, 64-bit)

I just discovered that an old MediaWiki of mine was infested with spam, and the database table named "text" (which contains the page content) is 3GB large. I've deleted all the spam pages manually, but:

• The table is still the same size.
• I wonder how it got to 3GB anyway. There wasn't that much spam (about a hundred medium-sized pages)

How can I get rid of this mess?

If you want to inspect the wiki, it's over here. The database is MySQL 5.0.75.

We have a server with an existing Linux installation (CentOS 6) + boot partition on disk 1, and an empty disk 2. What we'd like to setup is this:

• One disk (1 or 2, doesn't matter):
  • The main system in a snapshottable LVM (about 50%)
  • Backup space in an LVM (about 50%)
• Other disk
  • /home2 and MySQL data in a snapshottable LVM (about 50%)
  • Backup space in an LVM (about 50%)
• Either disk
  • /boot

The problem is, how can we move the main system onto an LVM without installing from scratch? The server is hosted so we don't have much influence over the OS installation process. We have:

• Regular SSH, of course
• A rescue mode, whereby a rescue OS is booted for us by tech support
• No remote KVM

Our idea was:

• In the running system, set up LVM on disk 2
• Reboot to rescue mode and copy the main system to disk 2 (leave /boot on disk1)
• Update grub and /etc/fstab
• Reboot into the main system, now on disk 2
• Wipe disk 1 and set up LVM for MySQL + /home2 on disk 1

But tips/verification on that plan are welcome.

Note: There is no important data on the server yet, though losing the configuration we already did would be a bit inconvenient.

I have configured my VirtualHost example.com to proxy all requests to a Tomcat server

ProxyPass / http://localhost:8088/app
ProxyPassReverse / http://localhost:8088/app

This works fine for URLs like example.com/page, but for example.com and example.com/ I get this redirection response, which obviously leads to nothing.

HTTP/1.1 302 Moved Temporarily
Date: Wed, 06 Jul 2011 21:13:37 GMT
Server: Apache-Coyote/1.1             <-- the redirect comes from tomcat
Location: http://example.com/app/     <-- nonsense
...

What can I do to fix it? Preferably in the Apache config.

I'm using Apache 2 and Tomcat 7

I'm writing a custom application that will be run as a service/daemon. According to convention, the binaries for it should be in /opt/appname. However, where do I put files such as:

• Config files
• Cache and storage
• Logs

We're running a (cPanel/WHM based) server which contains mainly PHP applications. They are executed with the permissions of their owner with the suExec component of Apache.

We're looking into moving some of our core applications to Java, running in Tomcat (with Apache proxying), which can apparently be easily installed from within WHM. However, is there a way to preserve the suExec functionality there, so Java applications also get executed with the correct permissions?

I'm trying to set up SSL, with SNI, in my apache and am getting the often-seen "ssl_error_rx_record_too_long" error in Firefox when accessing the site (https://test.me.dev.xxxx.net), from which I can conclude that the server is listening on port 443, but doesn't know to use SSL on it.

The server is Ubuntu 9.04 with Apache 2.2.11 I enabled SSL in the default way (a2enmod ssl).

Here is my relevant config:

NameVirtualHost *:*
Listen 80

<IfModule mod_ssl.c>
    Listen 443
</IfModule>

...

<VirtualHost *:*>

        DocumentRoot /home
        ServerAlias *.dev.xxxx.net
        UseCanonicalName Off
        # project.user.dev.xxxx.net
        VirtualDocumentRoot /home/%2/dev/%1/web

        SSLEngine On
        SSLCertificateFile /etc/apache2/certs/dev.crt
        SSLCertificateKeyFile /etc/apache2/certs/dev.key

</VirtualHost>

What is wrong?

I'm about to order a certificate and the order form asks me which server platform (e.g. Apache + mod_ssl) I'm using. It doesn't really matter what I choose, does it?

My host's backup server gives me my usage information in the following format:

Disk quotas for user vps**** (uid 1234):
     Filesystem  blocks   quota   limit   grace   files   quota   limit   grace
/dev/mapper/backup3-backup3
                6094452  2147483648 2147483648          365672       0       0

I've searched, including in man pages, but I can't figure out how to read the output. What do each of these columns mean? IIRC, my quota should be 10 GB.

I have a site with forums. The forums are currently at forums.example.com (a subdomain). I've heard that for SEO and usability in general it's better to have them at example.com/forums. Regarding security and maintainability, however, it would be better to host the forums in a seperate VirtualHost (or to be exact, a seperate cPanel account).

Is this even possible, should I want to do this?

I'm one of the leaders of revora.net, a creative gaming network that freely hosts lots of game (modding) projects. Lately, bandwidth has been an issue somewhat. To alleviate this we want to give our users the option to share their huge files with BitTorrent.

What we'd like to do is that people can upload their files via the FTP account they already have. They can then go to a custom web"site" where they can enable those files to be shared via BitTorrent.

We'd like our server to be the tracker (unless using another one is easier) and also act as a permanent seeder (so there's always at least one). We need to be able to control both (tracker and seeder-client) from PHP and other scripts.

What's the best way to set this up?

Quite simply, what is the easiest and most stable way to replace an apt-get installed PHP 5.2 with the new PHP 5.3 on an Ubuntu 8.10 server?

I'm running Dovecot to let people access their email on my server. Unfortunately, some are on networks where they can't connect on the normal IMAP port, so I'm trying to change it. I changed the config to contain:

protocol imap {
  listen = *.1433 *.143
  ...

Then dovecot fails to restart:

Fatal: listen: Can't resolve address *.1433: Name or service not known

What's the correct way of doing this? If possible, I'd prefer to have it listening on both ports.