Log in

Previous Entry | Next Entry

vzctl 4.4

A shiny new vzctl 4.4 was released just today. Let's take a look at its new features.

As you know, vzctl was able to download OS templates automatically for quite some time now, when vzctl create --ostemplate was used with a template which is not available locally. Now, we have just moved this script to a standard /usr/sbin place and added a corresponding vztmpl-dl(8) man page. Note you can use the script to update your existing templates as well.

Next few features are targeted to make OpenVZ more hassle-free. Specifically, this release adds a post-install script to configure some system aspects, such as changing some parameters in /etc/sysctl.conf and disabling SELinux. This is something that has to be done manually before, so it was described in OpenVZ installation guide. Now, it's just one less manual step, and one less paragraph from the Quick installation guide.

Another "make it easier" feature is automatic namespace propagation from the host to the container. Before vzctl 4.4 there was a need to set a nameserver for each container, in order for DNS to work inside a container. So, the usual case was to check your host's /etc/resolv.conf, find out what are nameservers, and set those using something like vzctl set $CTID --nameserver x.x.x.x --nameserver y.y.y.y --save. Now, a special value of "inherit" can be used instead of a real nameserver IP address to instruct vzctl get IPs from host's /etc/resolv.conf and apply them to a container. Same applies to --searchdomain option / SEARCHDOMAIN config parameter.

Now, since defaults for most container parameters can be set in global OpenVZ configuration file /etc/vz/vz.conf, if it contains a line like NAMESERVER=inherit, this becomes a default for all containers not having nameserver set explicitly. Yes, we added this line to /etc/vz/vz.conf with this release, meaning all containers with non-configured nameservers will automatically get those from the host. If you don't like this feature, remove the NAMESERVER= line from /etc/vz/vz.conf.

Another small new feature is ploop-related. When you start (or mount) a ploop-based container, fsck for its inner filesystem is executed. This mimics the way a real server works -- it runs fsck on boot. Now, there is a 1/30 or so probability that fsck will actually do filesystem check (it does that every Nth mount, where N is about 30 and can be edited with tune2fs). For a large container, fsck could be a long operation, so when we start containers on boot from the /etc/init.d/vz initscript, we skip such check to not delay containers start-up. This is implemented as a new --skip-fsck option to vzctl start.

Thanks to our user and contributor Mario Kleinsasser, vzmigrate is now able to migrate containers between boxes with different VE_ROOT/VE_PRIVATE values. Such as, if one server runs Debian with /var/lib/vz and another is CentOS with /vz, vzmigrate is smart enough to note that and do proper conversion. Thank you, Mario!

Another vzmigrate enhancement is option -f/--nodeps which can be used to disable some pre-migration checks. For example, in case of live migration destination CPU capabilities (such as SSE3) are cross-checked against the ones of the source server, and if some caps are missing, migration is not performed. In fact, not too many applications are optimized to use all CPU capabilities, therefore there are moderate chances that live migration can be done. This --nodeps option is exactly for such cases -- i.e. you can use it if you know what you do.

This is more or less it regarding new features. Oh, it makes sense to note that default OS template is now centos-6-x86, and NEIGHBOR_DEVS parameter is commented out by default, because this increases the chances container networking will work "as is".

Fixes? There are a few -- to vzmigrate, vzlist, vzctl convert, vzctl working on top of upstream kernel (including some fixes for CRIU-based checkpointing), and build system. Documentation (those man pages is updated to reflect all the new options and changes.

A list of contributors to this vzctl release is quite impressive, too -- more than 10 people.

As always, if you find a bug in vzctl, please report it to bugzilla.openvz.org.



( 2 comments — Leave a comment )
Aug. 4th, 2013 06:27 pm (UTC)
On a slightly different tangent: what is your opinion on the relevance of Docker to OpenVZ?
Aug. 7th, 2013 03:29 pm (UTC)
Docker is nice, they pushed that idea of application containers. Although, using LXC as a base is weak (at least for now).
( 2 comments — Leave a comment )

Latest Month

July 2016
Powered by LiveJournal.com
Designed by Tiffany Chow