LIBVIRT_URI not being used, hardcoded `qemu:///system` instead.

[zakkg3]

Libvirt uri seems to be hardcoded and not being readed from the sourced env var at /var/lib/one/remotes/etc/vmm/kvm/kvmrc : export LIBVIRT_URI=qemu+tcp://localhost/system

this is not working then. http://docs.opennebula.org/5.8/deployment/open_cloud_host_setup/kvm_driver.html#opennebula-configuration

https://github.com/OpenNebula/one/blob/def9661b4aa7c708d51679eb051ce8bebce09d80/src/im_mad/remotes/kvm-probes.d/machines-models.rb#L33

https://github.com/OpenNebula/one/blob/def9661b4aa7c708d51679eb051ce8bebce09d80/src/im_mad/remotes/kvm-probes.d/machines-models.rb#L92

https://github.com/OpenNebula/one/blob/441cf1f7f9e726cb5f200d661d50e92a4042fff7/src/im_mad/remotes/kvm-probes.d/kvm.rb#L32

Progress Status

• [ ] Branch created
• [ ] Code committed to development branch
• [ ] Testing - QA
• [ ] Documentation
• [ ] Release notes - resolved issues, compatibility, known issues
• [ ] Code committed to upstream release/hotfix branches
• [ ] Documentation committed to upstream release/hotfix branches

[asteven]

FTR: I'm working on a PR. Seems to work so far, but want to test some more before submitting it.

[asteven]

So a bunch of problems here. And questions to upstream.

There's roughly 90 places where the virsh executable is used/called throughout the ONE codebase. Basically from within vnm_mad, tm_mad, im_mad, vmm_mad scripts. Some of the scripts are shell, some are ruby.

Handling all of these cases by replacing virsh -c qemu:///system with something like virsh -c $LIBVIRT_URI seems problematic. Especially given that the source of LIBVIRT_URI is src/vmm_mad/remotes/kvm/kvmrc. Sourcing, reading, including that from all the consumers in vnm_mad, tm_mad, im_mad, vmm_mad does not seem to be a good solution.

I instead propose the following.

• prepend a custom bin directory to the PATH (e.g. $SCRIPTS_REMOTE_DIR/bin (aka /var/tmp/one/bin)) for all script executions
• introduce a script in $SCRIPTS_REMOTE_DIR/bin/virsh that knows how to properly connect to libvirt
• replace all occurrences of virsh -c qemu:///system or virsh -c $LIBVIRT_URI with just plain virsh

This way we can handle all the implementation details in once place. Any new scripts added in the future don't have to worry about LIBVIRT_URI and such things.

Given that LIBVIRT_URI is used in many other places then just vmm_mad I propose to define the variable somewhere else then its current location (src/vmm_mad/remotes/kvm/kvmrc). If you agree, where would such a 'global' variable go? And how could it be exposed to the $SCRIPTS_REMOTE_DIR/bin/virsh wrapper script?

Is there some central place where all the remote scripts are called from?

[asteven]

Example script for $SCRIPTS_REMOTE_DIR/bin/virsh

#!/bin/sh
  
remove_from_path() {
   # Remove the given directory from the PATH.
   PATH="${PATH//":$1:"/":"}" # delete any instances in the middle
   PATH="${PATH/#"$1:"/}" # delete any instance at the beginning
   PATH="${PATH/%":$1"/}" # delete any instance at the end
   export PATH
}

mydir=$(cd "${0%/*}"; pwd -P)

# Remove mydir from path to unhide the real virsh executable.
remove_from_path "$mydir"

# TODO: get LIBVIRT_URI from some where.

exec virsh --connect "$LIBVIRT_URI" $@

[atodorov-storpool]

IMO it is better to patch the 4 places where "qemu://system" is hard-coded instead of hijacking the virsh binary in a custom path. Several reasons:

• If there are additional scripts they will need to be updated too this means all known and unknown addons
• It will become harder for the general users to debug issues because the virsh is hijack-ed but the user will try with default paths and experience different results.

The files in question are:

src/vnm_mad/remotes/lib/command.rb 
src/im_mad/remotes/kvm-probes.d/kvm.rb
src/im_mad/remotes/kvm-probes.d/machines-models.rb
src/im_mad/remotes/lxd-probes.d/lxd.rb

On a side note in why not adapting the ruby code from https://github.com/OpenNebula/one/blob/master/src/vmm_mad/remotes/kvm/poll in the other places?

[asteven]

Then call the script one-virsh or whatever. Why would you want to depend on a env variable that is defined in some shell script/rc file in one mad from all others. And read/parse that variable from every script/shell/ruby/executable again and again.

Anyway. Upstream does not seem to be interested. I'm not working on this. Feel free to PR with your preferred solution.