CloudFoundry 学习之 warden 的安装

warden -- 资源隔离和管理。下面我将简单说一下，如何从头开始安装。

本文假设你对 Ruby 不熟悉!

首先请安装 ruby, 推荐使用 rvm ...

建议安装教程：https://rvm.io/rvm/install/　请耐心点自己阅读，并安装。(因为种种原因，不保证能成功) -- PS: 你也可以网上找中文教程，能成功就行。ruby 1.9.2 & 1.9.3 差别不大，都可安装。

安装 bundle, 运行 gem install bundler

=== ==== ====

好了，刚才是准备工作，现在开始安装 warden

1.  git clone git://github.com/cloudfoundry/warden.git

2. 进入 warden 的子目录　warden (名字相同，别奇怪，有 Gemfile 文件的这目录)

3. 执行  bundle

4. 按照README, 执行　sudo apt-get install -y linux-image-generic-lts-backport-natty
sudo apt-get install -y build-essential debootstrap

5. 设置 warden, 执行(这一步可简写 rake setup[config/linux.yml] ,不成功再用下面的)sudo bundle exec rake setup[config/linux.yml]

这一步，请特别注意最后的输出应该和这差不多。

Setting up build-essential (11.4build1) ...
Processing triggers for libc-bin ...
ldconfig deferred processing now taking place

在这之后，你可以在 /tmp/warden 目录下找到　[containers  和 rootfs]，前者为空，后者为刚才安装过程中生成的虚拟机

6. 执行(这一步可简写 rake warden:start[config/linux.yml] ,不成功再用下面的)sudo bundle exec rake warden:start[config/linux.yml]

b/warden/container/spawn.rb","lineno":121,"method":"set_deferred_success"}
{"timestamp":1350281577.2832768,"message":"Listening on /tmp/warden.sock, and ready for action.","log_level":"info","source":"Warden::Server","data":{},"thread_id":15005340,"fiber_id":23490560,"process_id":23434,"file":"/root/warden/warden/lib/warden/server.rb","lineno":277,"method":"block (2 levels) in run!"}    结果，这不是错误!

但这里有个坑，普遍存在，请看：

root@ubuntu-03:/home/wardener/warden/warden# rake warden:start[config/linux.yml]
{"timestamp":1349851024.4173303,"message":"rlimit_nofile: 1024 => 32768","log_level":"debug","source":"Warden::Server","data":{},"thread_id":14814280,"fiber_id":24325740,"process_id":12823,"file":"/home/wardener/warden/warden/lib/warden/server.rb","lineno":238,"method":"run!"}
{"timestamp":1349851024.4284155,"message":"Exited with status 1 (0.007s): [[\"/home/wardener/warden/warden/root/linux/setup.sh\", \"/home/wardener/warden/warden/root/linux/setup.sh\"]]","log_level":"warn","source":"Warden::Container::Spawn::DeferredChild","data":{},"thread_id":14814280,"fiber_id":24325740,"process_id":12823,"file":"/home/wardener/warden/warden/lib/warden/container/spawn.rb","lineno":117,"method":"set_deferred_success"}
{"timestamp":1349851024.4286866,"message":"Stdout: /sys/fs/cgroup does not exist...\n","log_level":"warn","source":"Warden::Container::Spawn::DeferredChild","data":{},"thread_id":14814280,"fiber_id":24325740,"process_id":12823,"file":"/home/wardener/warden/warden/lib/warden/container/spawn.rb","lineno":118,"method":"set_deferred_success"}
{"timestamp":1349851024.4290073,"message":"Stderr: ","log_level":"warn","source":"Warden::Container::Spawn::DeferredChild","data":{},"thread_id":14814280,"fiber_id":24325740,"process_id":12823,"file":"/home/wardener/warden/warden/lib/warden/container/spawn.rb","lineno":119,"method":"set_deferred_success"}
rake aborted!
command exited with failure

出错！ .. ..　运行到这：https://github.com/cloudfoundry/warden/blob/master/warden/root/linux/setup.sh#L25

错误是：　/sys/fs/cgroup: No such file or directory　（在virtualBox创建的ubuntu 10.04同样没有此目录）

我的物理主机上cgroup目录有：

blkio  cpu  cpuacct  cpuset  devices  freezer  memory  perf_event

在我用virtualBox创建的ubuntu 12.04机子上，有该目录，但为空目录。

---- --- ---

阅读源代码 -- 该 setup.sh 文件主要是设置　cgroup, 还有就是挂载 mount

再就是运行同目录下的 net.sh -- 主要用 iptables 设置网络

--- --- ----

我们已经知道报错在哪了，但现在问题还没能解决。

先看看源代码，了解下一步将要做的事情是？？？

if drained　＃　目前已知为 false
            recover_containers
          end

          FileUtils.rm_f(unix_domain_path) # 删除　/tmp/warden.sock
          server = ::EM.start_unix_domain_server(unix_domain_path, ClientConnection)　# 重建服务对象

          @drainer = Drainer.new(server) #@server, @connection, @state, on_complete_callbacks = server, Set.new, State::INACTIVE, []

          @drainer.on_complete do # @on_complete_callbacks << blk; run_machine
            Fiber.new do
              logger.info("Drain complete")
              # Serialize container state
              container_klass.registry.each { |_, c| c.write_snapshot }

              # Write out sentinel so we know to recover on next startup
              write_drained_sentinel

              EM.stop
            end.resume
          end
          Signal.trap("USR2") { @drainer.drain }

          # 除非有权限，否则阻塞。This is intentionally blocking. We do not want to start accepting
          # connections before permissions have been set on the socket.
          FileUtils.chmod(unix_domain_permissions, unix_domain_path)

          # Let the world know Warden is ready for action.
          logger.info("Listening on #{unix_domain_path}, and ready for action.")

解决办法：

1. 将container/features/cgroup.rb下的"/sys/fs/cgroup"改为"/mnt/cgroup" # 该文件在 这里

2. 同理，将root/linux/setup.sh下的"sys/fs/cgroup"改为"/mnt/cgroup" # 该文件在 这里

若是没有 "/mnt/cgroup" 这个目录，你可以自己创建。(这和cgroup有关，路径不对而矣，别大惊小怪)

再跑一遍配置及启动～～～～(如果你比较倒霉，还是报错。请先执行 :sudo apt-get install cgroup-bin 然后重启，再进行!

8. 打开另一终端，执行　bundle exec bin/warden-repl

root@ubuntu-03:/home/wardener/warden/warden# bin/warden-repl
warden> help
ping                          - ping warden
create [OPTION OPTION ...]    - create container, optionally pass options.
destroy <handle>              - shutdown container <handle>
stop <handle>                 - stop all processes in <handle>
spawn <handle> cmd            - spawns cmd inside container <handle>, returns #jobid
link <handle> #jobid          - do blocking read on results from #jobid
stream <handle> #jobid        - do blocking stream on results from #jobid
run <handle>  cmd             - short hand for stream(spawn(cmd)) i.e. spawns cmd, streams the result
list                          - list containers
info <handle>                 - show metadata for container <handle>
limit <handle> mem  [<value>] - set or get the memory limit for the container (in bytes)
limit <handle> bandwidth <rate> <bandwidth> - set the bandwidth limit for the container <rate> is the maxium transfer rate for both outbound and inbound(in bytes/sec) <burst> is the burst size(in bytes)
net <handle> #in              - forward port #in on external interface to container <handle>
net <handle> #out <address[/mask][:port]> - allow traffic from the container <handle> to address <address>
copy <handle> <in|out> <src path> <dst path> [ownership opts] - Copy files/directories in and out of the container
help                          - show help message

---

The OPTION argument for `create` can be one of:
  * bind_mount:HOST_PATH,CONTAINER_PATH,ro|rw
      e.g. create bind_mount:/tmp/,/home/vcap/tmp,ro
  * grace_time:SECONDS
      e.g. create grace_time:300

Please see README.md for more details.
warden>