Vagrant with chef solo fails to install in guest OS

[rferguson10]

Vagrant version

Vagrant 2.2.15

Host operating system

Windows 10 Home

Guest operating system

Centos 8 and 8.3

Vagrantfile

Vagrant.configure(2) do |config|

  config.vm.box = "bento/centos-8.3"
  config.vm.hostname = 'wordpress'

  config.vm.network :private_network, ip: '172.16.231.130'
  config.vm.network :forwarded_port, guest: 80, host: 8080
  config.vm.network :forwarded_port, guest: 9200, host: 9201

  if Vagrant.has_plugin?("vagrant-omnibus")
    config.omnibus.chef_version = 'latest'
  end

  config.ssh.insert_key = false
  config.vm.synced_folder ".", "/vagrant", type: "nfs", mount_options:['nolock,vers=3,udp,noatime,actimeo=1']

  config.vm.provider "virtualbox" do |v|
    v.memory = 2048
  end

  config.vbguest.auto_update = false

  config.vm.provision :shell, :inline => "ulimit -n 4048"

  config.vm.provision "chef_solo" do |chef|
    chef.arguments = "--chef-license accept"
    chef.run_list = [
      "recipe[wordpress]"
    ]
  end

end

Debug output

https://gist.github.com/rferguson10/a130b436c1106c1e8a170d88d3d65b88

Expected behavior

Chef should install into the guest OS and continue to provision using the provided recipe

Actual behavior

I'm presented with the following output:

Vagrant could not detect Chef on the guest! Even after Vagrant attempted to install Chef, it could still not find Chef on the system. Please make sure you are connected to the Internet and can access Chef's package distribution servers. If you already have Chef installed on this guest, you can disable the automatic Chef detection by setting the 'install' option in the Chef configuration section of your Vagrantfile:

chef.install = false

Steps to reproduce

1. vagrant up, watch for output

[ralfbetazi]

I find myself experiencing same problem as @rferguson10 but in a different environment as of this morning (~7:30 AM Pacific time).

Vagrant Version

2.2.15

Host operating system

Mac OS Big Sur 11.3

Guest operating system

Ubuntu 20.04

Vagrantfile

Vagrant.configure("2") do |config|
  config.vm.box = "bento/ubuntu-20.04"
  config.vm.define('demo') do |demo|
    demo.vm.provision 'chef_solo' do |chef|
      chef.arguments = '--chef-license accept'
    end
  end
end

Debug output

https://gist.github.com/ralfbetazi/ccf740f9e2267aa629165ef4f3d00f91

Expected behavior

Command vagrant up should launch a guest running Ubuntu 20.04.

Actual behavior

Command vagrant up exits with error. Final output reads:

==> demo: Running provisioner: chef_solo...
    demo: Installing Chef (latest)...
Vagrant could not detect Chef on the guest! Even after Vagrant
attempted to install Chef, it could still not find Chef on the system.
Please make sure you are connected to the Internet and can access
Chef's package distribution servers. If you already have Chef
installed on this guest, you can disable the automatic Chef detection
by setting the 'install' option in the Chef configuration section of
your Vagrantfile:

    chef.install = false

returning exit code 1.

[ralfbetazi]

Addendum to previous comment:

Saw issue #12338 and on assumption chef infra client installed but cleanup failed discovered an ugly stopgap workaround. One can get through provisioning by running vagrant up once, letting it fail (chef is installed, but machine is not provisioned), disabling chef installation, then running vagrant provision to complete provisioning. In other words:

1. Run vagrant up. It will fail with the message described by the original poster, but chef-solo is installed, but the cleanup failed.
2. Edit the Vagrantfile to add inside the chef-solo provision block add the following line:

   chef.install = false

3. Run vagrant provision. The guest will provision using chef-solo installed during Step 1.

This ugly stopgap measure might keep others from being completely stymied by this issue until it can be more properly resolved.

[bastianschwarz]

As an alternative: You could set chef.version to "16" (wich should still work). This was the last version that works for me. Either it was the removal of knife or the version string is different now so Vagrant fails to detect?

[clayg]

I was still having this problem with vagrant 2.2.18

Looks like the problem was that chef 17 removed knife from the install:

INFO ssh: Execute: test -x /opt/chef/bin/knife (sudo=true) DEBUG ssh: stderr: 41e57d38-b4f7-4e46-9c38-13873d338b86-vagrant-ssh DEBUG ssh: Exit status: 1

After some reading I decided the best way to "fix" this is to add:

chef.product = "chef-workstation"

... to my Vagrantfile somewhere around the chef.arguments = "--chef-license accept" bit

[kebugcheckex]

I searched a bit in the codebase, it seems that Vagrant is checking whether Chef is installed by testing if knife binary exists, in plugins/provisioners/chef/cap/linux/chef_installed.rb.

For some reason, the chef-solo installation doesn't come with knife. This also matches @clayg 's solution, where chef-workstation probably has the knife binary.

[collinmcneese]

@rferguson10 This should be fixed now in Vagrant 2.2.19