升级ruby版本和puma时，Capistrano重新启动错误版本的puma

Question

我正在升级一个使用capistrano部署并运行puma的Rails应用程序，从ruby2.2.0/puma 3.2升级到ruby2.3.1/puma 3.4。当我部署的时候，puma似乎在错误的版本中重新启动。

在我部署之前，Puma在pid 4717 /版本3.4下运行

vagrant@vagrant-master:/mnt/code/shared/tmp/pids$ ps -ef | grep -i puma
vagrant   4717     1  0 20:53 ?        00:00:00 puma 3.4.0 (unix:///mnt/code/shared/tmp/sockets/code-puma.sock) [20160510205039]                                                                                                                                                                                                                                                        
vagrant   4720  4717 12 20:53 ?        00:00:10 puma: cluster worker 0: 4717 [20160510205039]            

当我部署旧版本(puma 3.2 / Ruby 2.2.0)时，puma重新启动，它现在在pid 5466下运行：

vagrant@vagrant-master:/mnt/code/shared/tmp/pids$ ps -ef | grep -i puma
vagrant   5466     1  0 20:56 ?        00:00:00 puma 3.4.0 (unix:///mnt/code/shared/tmp/sockets/code-puma.sock) [20160510205440]                                                                                                                                                                                                                                                        
vagrant   6154  5466  0 20:56 ?        00:00:00 puma: cluster worker 0: 5466 [20160510205440]     

我的puma错误日志显示它运行的是2.3.1而不是2.2.0

/usr/local/rbenv/versions/2.3.1/lib/ruby/gems/2.3.0/gems/bundler-1.12.3/lib/bundler/spec_set.rb:95:
in `block in materialize': Could not find active_attr-0.8.5 in any of the sources (Bundler::GemNotFound)

然而，它应该已经切换到Puma 3.2/ruby 2.2.0。当我强制终止美洲狮时，它会用正确的版本重启(通过monit) (本例中为美洲狮3.2 )

vagrant@vagrant-master:/mnt/code/shared/tmp/pids$ kill -9 5466

vagrant@vagrant-master:/mnt/code/shared/tmp/pids$ ps -ef | grep -i puma
vagrant  17834     1  0 21:04 ?        00:00:00 puma 3.2.0 (unix:///mnt/code/shared/tmp/sockets/code-puma.sock) [20160510205440]                                                                                                                                                                                                                                                        
vagrant  17837 17834 90 21:04 ?        00:00:01 puma: cluster worker 0: 17834 [20160510205440]   

部署日志中的相关行：

DEBUG [07f95004] Running /usr/bin/env [ -f /mnt/code/shared/tmp/pids/puma.pid ] as vagrant@vagrant-master
DEBUG [07f95004] Command: [ -f /mnt/code/shared/tmp/pids/puma.pid ]
DEBUG [07f95004] Finished in 0.007 seconds with exit status 0 (successful).

DEBUG [42d98483] Running /usr/bin/env kill -0 $( cat /mnt/code/shared/tmp/pids/puma.pid ) as vagrant@vagrant-master
DEBUG [42d98483] Command: kill -0 $( cat /mnt/code/shared/tmp/pids/puma.pid )
DEBUG [42d98483] Finished in 0.007 seconds with exit status 0 (successful).

INFO [e81875a3] Running RBENV_ROOT=/usr/local/rbenv RBENV_VERSION=2.2.0 /usr/local/rbenv/bin/rbenv exec bundle exec pumactl -S /mnt/code/shared/tmp/pids/puma.state restart as vagrant@vagrant-master
DEBUG [e81875a3] Command: cd /mnt/code/current && ( export PATH="/usr/local/rbenv/shims:/usr/local/rbenv/bin:$PATH" RBENV_ROOT="/usr/local/rbenv" RBENV_VERSION="2.2.0" RACK_ENV="vagrant" ; RBENV_ROOT=/usr/local/rbenv RBENV_VERSION=2.2.0 /usr/local/rbenv/bin/rbenv exec bundle exec pumactl -S /mnt/code/shared/tmp/pids/puma.state restart )
DEBUG [e81875a3]    Command restart sent success

这是我在capistrano中的puma配置的一部分

set :puma_bind,       "unix://#{shared_path}/tmp/sockets/#{fetch(:application)}-puma.sock"
set :puma_state,      "#{shared_path}/tmp/pids/puma.state"
set :puma_pid,        "#{shared_path}/tmp/pids/puma.pid"
set :puma_preload_app, false
set :puma_prune_bundler, true

如何在不终止puma进程的情况下部署ruby版本和puma版本的更改？为什么puma从错误的版本开始？

谢谢

Answer 1

检查puma.rb配置文件是否指定了puma.pid和puma.state路径：

pidfile '/mnt/code/shared/tmp/pids/puma.pid'
state_path '/mnt/code/shared/tmp/pids/puma.state'

还要检查deploy/production.rb是否指定了指向以下位置的路径：

set :puma_state,      "#{shared_path}/tmp/pids/puma.state"
set :puma_pid,        "#{shared_path}/tmp/pids/puma.pid"

我也有类似的问题，在设置了这个之后，它工作得很好。