How to run with chef-client zero (-z, local mode, or solo)

[PAStheLoD]

Hello,

Could you help me with this? I've tried to run it through bundler, and did bekshelf install too.

I want to clone this onto a server (or preferably not) and issue a few one liners to get the dependencies (on both sides) and have a running kafka 20 minutes later.

Thanks!

Starting Chef Client, version 11.12.8
resolving cookbooks for run list: ["kafka-cookbook::binary"]
Synchronizing Cookbooks:
  - kafka-cookbook
Compiling Cookbooks...

================================================================================
Recipe Compile Error in /root/.chef/local-mode-cache/cache/cookbooks/kafka-cookbook/recipes/binary.rb
================================================================================


NoMethodError
-------------
No resource or method named `kafka_download' for `Chef::Recipe "binary"'


Cookbook Trace:
---------------
  /root/.chef/local-mode-cache/cache/cookbooks/kafka-cookbook/recipes/binary.rb:9:in `from_file'


Relevant File Content:
----------------------
/root/.chef/local-mode-cache/cache/cookbooks/kafka-cookbook/recipes/binary.rb:

  2:  # Cookbook Name:: kafka
  3:  # Recipe:: binary
  4:  #
  5:  
  6:  kafka_tar_gz = [kafka_base, kafka_archive_ext].join('.')
  7:  local_file_path = ::File.join(Chef::Config.file_cache_path, kafka_tar_gz)
  8:  
  9>> kafka_download local_file_path do
 10:    source kafka_download_uri(kafka_tar_gz)
 11:    checksum node.kafka.checksum
 12:    md5_checksum node.kafka.md5_checksum
 13:    not_if { kafka_installed? }
 14:  end
 15:  
 16:  execute 'extract-kafka' do
 17:    user 'root'
 18:    group 'root'

[mthssdrbrg]

Chef uses the name of the directory when "creating" LWRPs, so in this case the directory is named kafka-cookbook which unfortunately won't work that well. If you clone it to a kafka directory everything should work as expected.

Let me know whether this solves the issue for you.

[PAStheLoD]

Thanks, it helps get the ball rolling, then it

• fails on missing /opt/kafka/build,
• missing kafka user and group,
• and then it needed a chown kafka /opt/kafka.

And I get a nice extracted deployment, but attributes doesn't seem to have any effect.

I run chef-client -z -c /var/chef/cookbooks/solo.rb -j /var/chef/cookbooks/attr.json and attr.json is:

{
    "run_list": [ "recipe[kafka::binary]" ],
    "kafka": {
        "init_style": "systemd",
        "automatic_start": true
     }
}

[mthssdrbrg]

Use the default recipe instance rather than the binary one (it will use the binary recipe by default, but the default recipe also includes some other "internal" recipes, for setting up user/group and whatnot).

[mthssdrbrg]

@PAStheLoD: did that solve your issue?

[PAStheLoD]

Yes, thanks, now I'm trying to use attributes, but so far I haven't been able to figure it out completely, and the cookbook seems to assume RedHat style layout, because it fails with the following:

* Parent directory /etc/sysconfig does not exist.

So I see that somehow I should tweak init_opts (kafka_init_opts {:env_path=>"/etc/sysconfig/kafka", :env_template=>"systemd/kafka.env.erb", :source=>"systemd/default.erb", :script_path=>"/etc/systemd/system/kafka.service", :provider=>Chef::Provider::Service::Systemd, :permissions=>"644"}), right?

[mthssdrbrg]

Hm, I was under the assumption that /etc/sysconfig for Environment files was something generic for systemd, but perhaps I'm mistaken. What Linux distro are you using?

[PAStheLoD]

Debian 7

For systemd the /etc/systemd is the generic directory.

[mthssdrbrg]

It appears as if /etc/sysconfig for EnvironmentFile(s) is a Fedora specific thing, my bad. There doesn't really seem to be any generic place to put EnvironmentFile(s), but I'm gonna do some more digging to see where exactly this file should be placed. Another alternative would be to create a wrapper script for starting Kafka, but that's a slightly larger task than just placing the file in a different location.

[PAStheLoD]

Debian uses /etc/default/.

Also, I'd link kafka.service into multi-user.wants target.

And, it seems that if I modify scala_version, nothing happens. (So, maybe I should be asking how to force a reinstall? [I'm using current master.])

Anyway, with your tremendous help I've finally managed to reach the end of the cookbook.

[mthssdrbrg]

Hm, alright, maybe I should do the same kind of value_for_platform_family logic that's done for the :sysv install method for the time being.

To force a reinstall simply remove /opt/kafka or whatever node.kafka.install_dir is set to.

I'll look into multi-user.wants, my knowledge of systemd is rather limited...

[mthssdrbrg]

I just merged a branch with some systemd-related fixes, would be great if you could take it for a spin.

[PAStheLoD]

Okay, master is now at 5cd3511.

Removing /opt/kafka, leads me to this:

  * kafka_install[/opt/kafka] action run
    * execute[install-kafka] action run

      ================================================================================
      Error executing action `run` on resource 'execute[install-kafka]'
      ================================================================================

      Mixlib::ShellOut::ShellCommandFailed
      ------------------------------------
      Expected process to exit with [0], but received '1'
      ---- Begin output of cp -r /root/.chef/local-mode-cache/cache/kafka-build/kafka_2.9.2-0.8.1.1/* /opt/kafka ----
      STDOUT: 
      STDERR: cp: cannot stat '/root/.chef/local-mode-cache/cache/kafka-build/kafka_2.9.2-0.8.1.1/*': Permission denied
      ---- End output of cp -r /root/.chef/local-mode-cache/cache/kafka-build/kafka_2.9.2-0.8.1.1/* /opt/kafka ----
      Ran cp -r /root/.chef/local-mode-cache/cache/kafka-build/kafka_2.9.2-0.8.1.1/* /opt/kafka returned 1

I guess the cp -r would run as kafka, but obviously /root is off limits for anyone except root.

Also, I have set scala version to 2.10.1, so seeing 2.9.2 is a bit strange :)

[mthssdrbrg]

Ah, so Chef::Config.file_cache_path is under /root, that's rather unexpected. I've worked around that now though, should be in master in a little bit.

I'm not sure why it's using 2.9.2 if you've specified 2.10.1, could you share your cookbook code?

[mthssdrbrg]

The cp -r issue should be resolved in b996099.

[PAStheLoD]

Still using zero/solo,

attr.json:

{
    "run_list": [ "recipe[kafka::default]" ],
    "kafka":  {
        "default": {
            "init_style": "systemd",
            "automatic_start": true,
            "scala_version": "2.10.1"
        }
    }
}

And solo.rb:

cookbook_path [ "/root/.berkshelf/cookbooks", "/var/chef/cookbooks" ]

In /var/chef/cookbooks there is only attr.json and solo.rb, I just call it as chef-client -z -c /var/chef/cookbooks/solo.rb -j /var/chef/cookbooks/attr.json, and in the berkshelf folder there is only the apt, java and kafka cookbooks.

So, now I get the error message about systemd's system/ directory being in the wrong place:

Recipe: kafka::_configure
  * template[/opt/kafka/config/log4j.properties] action create (up to date)
  * template[/opt/kafka/config/server.properties] action create (up to date)
  * template[/etc/default/kafka] action create (up to date)
  * template[/usr/lib/systemd/system/kafka.service] action create
    * Parent directory /usr/lib/systemd/system does not exist.
    ================================================================================
    Error executing action `create` on resource 'template[/usr/lib/systemd/system/kafka.service]'

It's in /lib/systemd/system.

Strange. The "official" page on unit load paths mentions the one you use in the code, man page recommends using pkg-config:

Currently on Debian Jessie

pkg-config systemd --print-variables | xargs -n1 -I{} bash -c "echo -n {}=; pkg-config systemd --variable={}"


systemdsystemconfdir=/etc/systemd/system
systemdusergeneratordir=/usr/lib/systemd/user-generators
sysctldir=/usr/lib/sysctl.d
systemdshutdowndir=/lib/systemd/system-shutdown
tmpfilesdir=/usr/lib/tmpfiles.d
systemgidmax=999
catalogdir=/usr/lib/systemd/catalog
systemduserconfdir=/etc/systemd/user
systemdsleepdir=/lib/systemd/system-sleep
systemdsystemunitdir=/lib/systemd/system
systemdutildir=/lib/systemd
systemdsystemunitpath=/etc/systemd/system:/etc/systemd/system:/run/systemd/system:/usr/local/lib/systemd/system:/lib/systemd/system:/usr/lib/systemd/system:/lib/systemd/system
prefix=/usr
libdir=/usr/lib/x86_64-linux-gnu
systemduserunitpath=/etc/systemd/user:/etc/systemd/user:/run/systemd/user:/usr/local/lib/systemd/user:/usr/local/share/systemd/user:/usr/lib/systemd/user:/usr/lib/systemd/user:/usr/share/systemd/user
sysusersdir=/usr/lib/sysusers.d
modulesloaddir=/usr/lib/modules-load.d
systemdsystemgeneratordir=/lib/systemd/system-generators
binfmtdir=/usr/lib/binfmt.d
systemuidmax=999
systemduserunitdir=/usr/lib/systemd/user
systemduserpresetdir=/usr/lib/systemd/user-preset
systemdsystempresetdir=/lib/systemd/system-preset

[mthssdrbrg]

Sigh, guess I'll revert the commit that change the path for the unit file. It's kinda funny though, I was under the impression that systemd was supposed to be the "same" on all platforms, but it sure doesn't look like it.

[mthssdrbrg]

16a5f0c should be good to go, give it a spin whenever you get the chance.

[PAStheLoD]

Well, it's supposed to be universal, and you could go with whatever path you feel like, just mkdir -p it, but the real advantage is that you can just write a unit file once and it supposedly runs on every systemd-enabled distro.

And it's working! Sort of. Because I can't figure out how to write a JSON Attributes file to override the defaults. Could you help me out with that?

[mthssdrbrg]

Yeah, that's true, guess I sort of expected paths to be universal as well :). Being able to use a single unit file is definitely nice though.

test-kitchen uses the following JSON attributes file, and it overrides the defaults correctly:

{
    "java": {
        "jdk_version": 7
    },
    "kafka": {
        "broker": {
            "controlled_shutdown_enable": null,
            "log_dirs": [
                "/mnt/kafka-logs-1",
                "/mnt/kafka-logs-2"
            ],
            "zookeeper_connect": [
                "localhost:2181"
            ],
            "zookeeper_connection_timeout_ms": 15000
        },
        "checksum": "",
        "md5_checksum": "",
        "scala_version": "2.10",
        "ulimit_file": 128000,
        "version": "0.8.1.1"
    },
    "run_list": [
        "recipe[apt]",
        "recipe[java::default]",
        "recipe[kafka::default]"
    ]
}

The only thing I can think of is that run_list is specified after the attributes, but I can't imagine that that would make a difference (or at least I certainly hope it doesn't...).

[PAStheLoD]

Thanks, the "default" was causing the problems.

Everything works as intended, so I'm closing this.

[mthssdrbrg]

Cool, I'm glad that it worked out alright.

[PAStheLoD]

Oh, sorry for my lack of thoroughness, but I just noticed that the systemd service file refers to /etc/sysconfig/kafka even on Debian. Do you plan to handle distribution specific differences?

[mthssdrbrg]

Hm, I changed that in b55aba0 and the systemd service file should be in /etc/systemd/system since 16a5f0c.

[PAStheLoD]

I've run systemd-delta that informed me that they are identical. Then edited the system file.

But I just did a run and:

  * template[/etc/systemd/system/kafka.service] action create
    - update content in file /etc/systemd/system/kafka.service from 8535f8 to 5e844d

    --- /etc/systemd/system/kafka.service       2014-11-24 20:34:19.152063214 +0100
    +++ /tmp/chef-rendered-template20141124-2650-1ax6fl1        2014-11-24 20:34:51.403124971 +0100
    @@ -3,6 +3,7 @@

     [Service]
     User=kafka
    +EnvironmentFile=/etc/sysconfig/kafka

Anyway, I've found that with systemd setting rlimits are quite easy, and kafka needs quite a lot of them, so could you also add:

LimitNOFILE=65000 (or whatever someone specified in attributes ulimit_nofile)

[mthssdrbrg]

Oh, wow, just realized that the systemd service template hardcodes /etc/sysconfig/<%= @daemon_name %>, I'll get on that in a minute, and I'll include the limit as well.

[mthssdrbrg]

Latest master contains fixes for the two things that you mentioned, let me know if you find something else that should be fixed.

[PAStheLoD]

Currentely I don't have other kafka nodes to install, sorry :)

[mthssdrbrg]

Alright, then I'll close this again as the integration tests passed :)

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.