Weekly Changes – 15/03/2021
The changes for this week include an important fix for glibc on SL7, anyone with machines using CUDA will want to apply it as soon as possible. Otherwise the main changes are a range of improvements to Ubuntu support. Here’s the full summary…
SL7 glibc update
The 2.17-322.el7_9 security update for glibc on SL7 contains a flaw – bug #1925204 – which results in printf of long double with a value of 0 producing the string “nan”, this is known to break nvidia CUDA. An update has been provided in the SL7.9 fastbugs updates repository – 2.17-323.el7_9 – but nothing has been provided for SL7.8 so we have imported the fixed packages into our lcfg repository. It is somewhat surprising that a flaw in a security update for SL7.8 and SL7.9 has only been fixed in the fastbugs repository for SL7.9, it’s not clear if it will ever be backported officially. As this could break other software we decided to override the default glibc version on SL7.8.
Ubuntu kernel
The INF_TEST kernel for Ubuntu has been updated to 5.4.0-66.74.
remctl server
The schema for the remctld component has been updated to version 3 to add a cfgfile resource. The new resource is set to the correct path for both SL7 and Ubuntu. See bug#1271 for details, thanks to Barry O’Rourke in the School of Physics for providing the patch.
wallet server
The wallet server header has been updated to add software dependencies for Ubuntu. This work to support Ubuntu isn’t yet complete but will be in the stable release for next week. See bug#1273 for details.
network nagios monitoring
The network component nagios passive monitoring check script has gained support for checking for an IPv6 default route. As not all sites/systems will have IPv6 enabled this check is optional. There is a new nagios_network_checks resource, to enable the check for the default route add ipv6route to the list (note that this is enabled by default for DICE machines). In the same way, bonding checks are now enabled by adding bonding to that resource, it is enabled by default at the LCFG-level to match with previous behaviour.
LCFG minimal mode
Very basic support has been added for a minimal mode which means that LCFG can be used with just a small number of components. Only the client, file and inifile components are included by default. This mode is primarily intended for learning and testing LCFG, we do not plan to provide complete support for the mode in all headers. The mode can be enabled by defining the LCFG_MINIMAL macro at the top of a profile.
LCFG service
The LCFG server (often known as the compiler) code has been updated to support Ubuntu. This fixes various file paths and permissions and also adds support for starting the daemon (mkxprof) from a systemd service. The server component, which we use to configure and manage the service, has not yet been updated to handle these changes. It looks like the service works fine on Ubuntu but we currently recommend running your LCFG service on SL7 whilst we do more testing.
LCFG preseed web service
The configuration for the LCFG preseed web service has been improved so that it integrates better with the standard apacheconf way of doing things. Various software dependencies have also been added so that it works on Ubuntu. Note that the apacheconf component has not yet been ported to Ubuntu so this service is not yet ready for deployment on that platform.
New Perl package options
Ubuntu has gained various new Perl package options:
LCFG_OPTIONS_PERL_CGI– CGI supportLCFG_OPTIONS_PERL_SQLITE– SQLite database supportLCFG_OPTIONS_PERL_POSTGRES– PostgreSQL database supportLCFG_OPTIONS_PERL_DATETIME– DateTime modules
These can be enabled by adding the option name to the profile.pkgcppopts resource, for example:
!profile.pkgcppopts mADD(LCFG_OPTIONS_PERL_SQLITE)
dkms
On Ubuntu the latest version of the dkms package has a new dependency on dctrl-tools. This has been added to the postship package list for the LCFG_OPTIONS_DKMS package option. Due to a deficiency in the way the postship and options interact it also had to be added for a few other options which depend on dkms. This issue is likely to affect other options in a similar way so at some point we will have to tweak how the postship package list is included in the stack of package lists.
Recent comments