31 October 2006Own project · AdministrationOwn post

A server move in 19 hours: when my host Alturo shut down

Summary

In September 2006 United Internet announced the end of its discount hosting brand Alturo, where my server was located; all customers were given notice for 31 October. In a 19-hour session, 23 domains, a mail server, three forums and several gigabytes of data moved to a new server with a new Debian and much stricter security. This post from my old blog retells the move, shows what belonged to the toolkit at the time and what I would plan differently today.

The notice

On 19 September 2006 heise online reported that United Internet AG was closing its discount hosting subsidiary Alturo. The brand had been founded only two and a half years earlier to rent out older hardware of its sister provider 1&1 once more. Dedicated servers had since become cheaper, the target group was turning to brand-name products, and Alturo no longer paid off. All customers were given regular notice for 31 October 2006, and Alturo GmbH was to be dissolved on 30 November.

My server was there. So I had a good six weeks to move everything that had grown on it over the years.

19 hours in one go

In the end it happened in one stretch: 23 domains with their subdomains, a mail server, three forums, 70 MB of MySQL databases, 3.6 GB of Zope databases for the Plone blog, 8 GB of data served over HTTP and various home-made PHP scripts moved to the new server “Baldur”. MySQL, PHP, Apache and many other programs mostly jumped one major version, some two; all scripts quietly running in the background had to come along.

The old server still ran Debian Sarge, the new one the then testing branch Etch, which was only officially released in April 2007. The session lasted 19 hours. My ears, lungs and back complained loudly afterwards, and I still did not break my personal record.

What the new server could do

I used the move to make many things stricter. The firewall got a more effective brake against SSH brute-force attacks, behaved more RFC-compliant and now also regulated outgoing traffic. Instead of mod_security, a similar, leaner solution ran for performance reasons. The Teamspeak and Counter-Strike servers got rebuilt chroot environments.

In addition to the virus scanner and SpamAssassin, the mail server now fought spam with Razor, DCC and Pyzor, three services that collect and compare checksums of known spam. Logwatch 7 reported every error, an integrity check raised an alarm when files changed, and quotas prevented anyone from using their mailbox as a backup for their DVD collection. Everything sat on an AES-256 encrypted RAID 1 of 150 GB, with 1 GB of RAM and 3 GHz. Mail could be fetched encrypted, and there was no unencrypted FTP at all.

Barely online, and the spam starts

While the move was running, the mail server was down for a few hours, and anyone who wanted could watch a hair-tearing admin on a video stream. As soon as it was back, the spamming started. An excerpt from the log of the first minutes:

rejected EHLO from ...: syntactically invalid argument(s)
... rejected RCPT <...>: DNSBL listed at sbl-xbl.spamhaus.org
no host name found for IP address ...
... rejected EHLO or HELO ...: "Dropped IP-only or IP-starting helo"

Malformed EHLO greetings, senders on blocklists, hosts without names: the simple checks at the door filtered most of it before a message was even accepted.

And then the old server went off a day early

On 31 October came the punchline: according to a further heise report, Alturo disconnected all dedicated servers and web hosting packages at around 2 a.m., one day before the contracts ended. A spokesman for the sister company 1&1 confirmed that access had been switched off a day too early by mistake, restored it and extended access by two days until the night of 2 to 3 November. Anyone who had planned their move for the last day was briefly left without a server. My comment at the time: a typical case of oops.

What I would plan differently today

The move worked, but I would avoid 19 hours in one go today. Some things can be done beforehand: lower the TTL of the DNS records days in advance so that the switch takes effect quickly; catch mail via a second MX record instead of switching the server off; synchronise databases and data in advance and transfer only the difference on moving day.

The configuration belongs in a tool like Ansible, or at least in a Git repository, so that the new server is not built from memory. And the most important lesson from Alturo still applies: anyone renting from a provider needs a plan for the day that provider stops.

Numbers at a glance

heise reported the end of Alturo on 19 September 2006; all customers were given notice for 31 October 2006.
Moved: 23 domains, 1 mail server, 3 forums, 70 MB of MySQL, 3.6 GB of Zope databases, 8 GB of web data.
New server: 1 GB RAM, 3 GHz, encrypted RAID 1 with 150 GB.
According to heise, Alturo switched the servers off a day too early by mistake and then extended access by two days.

References

Remarks

This post was originally published on 31 October 2006 in my old Plone blog, together with a note on the mail server outage of the previous day. In 2026 I retrieved both from the Web Archive, merged and revised them.
The old post said “long live Debian Edge”. Etch was meant; the typo survived until now.

Original post in the Web Archive