Changelog

Below you can see the 5 most recent changes to Uberspace 7. For older changes, please refer to the Changelog Archive.

Note

Sometimes the version shown on your host may be higher than the newest version here. In this case we might have applied additional fixes shortly after a release or did internal changes without user impact. We deem updates like these hotfixes and they are not necessarily included in this changelog.


[7.10.0] - 2021-03-04

Added

  • updated Java to 15
  • each users CPU usage is now limited to 6 cores, improving stability.
  • pecl-mailparse
  • HEIC support for ImageMagick 6, and by extension PHP

Changed

  • recommended SPF record is now include:spf.uberspace.de so we can reoute mails more easily. The current records prevents us from relaying mails through another server temporarily.
  • to be consistent with our advice to use .uber.space domains for mail, user.host.uberspace.de is now no longer part of mail domain list.
  • uberspace mail domain add now explains that the trailing dot in MX records is correct, but not necessary or possible to enter in many DNS interfaces.
  • rspamd’s FORGED_RECIPIENTS test now adds fewer points to the spam score to counter many reported false-positive.

Fixed

  • RAM limits for users were not applying consistently, leading to outages in the recent past. We now apply the limits ourselves instead of relying on systemd, increasing stability in the future.
  • Sometimes systemd failed to reload nginx, leading to new domains not being available. We now use the nginx tooling directly instead of relying on systemd’s $MAINPID variable, hopefully fixing this.
  • MySQLs temporary files are now written to SSDs on all hosts, increasing performance for big queries that don’t fit into RAM.

Internal

  • there is a dummy uberspace-letsencrypt-renew script, which does nothing. Many U6 users leave their let’s encrypt cronjob in place, even though U7 does not need one. The resulting cron error mails confuse users, which increases our support volume. The dummy script automates those cases.
  • we rewrote the playbook, which updates MariaDB, enabling updates to 10.4 and 10.5 in the future.
  • some hosts have additional SSD devices for yum, rpm and the systemd journal. Since we are moving all hosts to SSDs, these are not necessary anymore. We wrote a playbook to remove them in the future, making all hosts consistent again.
  • we now detect and automatically ban more mining tools.
  • MySQL backups now only happen for databases, which changed since the last backup. This reduces the system load at night and further increases storage performance and stability.

[7.9.0] - 2021-01-25

Added

  • PHP 8.0
  • Ruby 3.0
  • InfluxDB and telegraf
  • PostgreSQL
  • CouchDB
  • MongoDB
  • .NET 5.0
  • gobject-introspection-devel, pango-devel, ripgrep, bat, asciidoc, ledger

Changed

  • legacy URLs like adminer., pma., and webmail.host.uberspace.de redirect to their global counterparts (e.g. https://webmail.uberspace.de)
  • removed PHP 7.2
  • removed .NET 2.0 and 2.2
  • httpd is now allowed to read files with user_home_t SELinux labels. This fixes usability issues because of files removed from home. It also enables CGI scripts to access libraries installied in $HOME/.local and similar. There is still no official support for CGI, though.

Fixed

  • tmux sessions no longer break after some time. We mistakenly removed them from /tmp automatically and now leave them be.
  • MySQL backups sometimes (1 or 3 databases in total on _all_ hosts) fail, so we now retry them once. This increases the reliability of the provided backups and silences our monitoring.
  • Sieve configuration files no longer show up as folders in mail clients.

Internal

  • We migrated additional hosts to SSD storage.
  • Add a test for redis.
  • Add monitoring check for failed user services. This way we will notice, if your supervisord or php-fpm fail.
  • Add monitoring check for individual MySQL backups. Monitoring for the backup process as a whole was already present.

[7.8.1] - 2020-12-22

Added

Changed

  • Increased the Rspamd reject score to 15 (up from 10).
  • Reduced the Rspamd score for INVALID_RCPT_8BIT to 3 (down from 6).
  • We limit the Spam Assassin rules we use with Rspamd to ZMI.
  • Deprecated PHP 7.2. It will be removed early next year.
  • We added a connection timeout on port 587 (hard capped at two hours, or one hour for idle connections). Our SMTP submit queue suffered from lingering connections, we hope this helps to mitigate it.
  • The output of uberspace mail domain add now ends domain names with a dot (.). We hope this helps avoiding situations, where it could otherwise be interpreted as relative to an origin (this mostly effects c/p to bind configurations, but also some web based GUIs).

Internal

  • We migrated a lot of hosts to SSD storage.
  • fstrim now runs about weekly on the hosts. Concrete times are distributed randomly, to minimize the impact on the cluster.
  • While creating SQL backup dumps, we now log and monitor MariaDB errors.
  • We moved the Rspamd logs out of the journal (for now). This allows us to have a longer retention policy for those, while still keeping them pretty verbose. We will fine tune our Spam filtering over the next releases, so this might come in handy.

[7.8.0] - 2020-12-01

Added

  • Support for managed Sieve
  • Mails from spam folder are now auto-expunged after 30 days.
  • Add special RFC 6154 folders to Dovecot config, so mail clients detect trash, spam, and other default folders automatically.

Changed

  • We now keep logs for incoming mails for 10 days instead of one day to aid debugging of missing mails in support.
  • System logs are now kept for one to two days, instead of just one.
  • The default values of max_execution_time and max_input_time are now 90 seconds and 60 seconds respectively, to free up stuck php-fpm workers more quickly. Higher values can be set using a config file in ~/etc/php.d. CLI invocations like cronjobs are not affected.

Internal

  • MySQL backups are now monitored, alerting us when the process stops working.
  • Optimized log output of our scripts, so we can keep the useful logs for longer.
  • Instead of reloading nginx after log rotation, we now call nginx -s reopen to reduce load spikes and thus increase reliability.
  • We now run fstrim regularly to free unused storage in our ceph cluster. This enables us to use it more efficiently.

[7.7.10] - 2020-11-17

Added

  • mg: a tiny Emacs-like editor
  • numactl command to please MongoDB
  • modern TLS settings for POP3/IMAP/SMTP-SUBMIT (reverted because they block connections from thunderbird)
  • opus, opus-tools, and opus-devel
  • many fonts to support non-western scripts

Fixed

  • MariaDB backups now includestored routines
  • table_definition_cache is now 20000 to meet friendica’s requirements
  • The SMTP connection limit introduced in v7.7.7 now actually works.
  • $user.uber.space is now correctly displayed in uberspace mail domain list
  • uberspace * domain list output is now sorted

Internal

  • Log rotation is now randomized to happen between 4 and 5 am. The time is constant for each host, so they are always rotated at the same time for a given host. This reduces the IO load on our storage and therefore improves performance and reliabilty at night.
  • Prometheus’ node_exporter can now be monitored by our icinga2 setup, leading to more complete graphs for us and better performance for you.
  • Sometimes our internal CI amassed a lot of temporary DNS records, which exceeded the quota of our DNS provider, griding our CI and development to a halt. The records are now purged reguarly.