puma 是 rails5 的标配可是你知道 puma 的线程数与数据库连接池的关系吗？

当我们配置 rails 的时候应注意 puma 的线程数与数据库连接池数一致。

否则会出来一些奇怪的现象：

例如在并发大，但是在内存，cpu,mysql 压力不大的情况下，ruby 耗时特别的长。使用 Newrelic 与 ab 工具 尝试了解项目的性能

例如得到一个错误：ActiveRecord::ConnectionTimeoutError (could not obtain a connection from the pool within 5.000 seconds (waited 5.002 seconds); all pooled connections were in use)

下面我们通过 connection_pool 的源码来了解一下

这里有着一个缓存，缓存着一个 conn, 这个 conn，在此次的请求中可以被使用任意次数。缓存不存在时，将从连接池中获取相应的 conn

#/activerecord/lib/active_record/connection_adapters/abstract/connection_pool.rb#363
def connection
  @thread_cached_conns[connection_cache_key(Thread.current)] ||= checkout
end

#/activerecord/lib/active_record/connection_adapters/abstract/connection_pool.rb#389

# If all connections are leased and the pool is at capacity (meaning the
# number of currently leased connections is greater than or equal to 
# size limit set), an ActiveRecord::ConnectionTimeoutError exception will be raised.
def checkout(checkout_timeout = @checkout_timeout)
  checkout_and_verify(acquire_connection(checkout_timeout))
end

获取 conn 的方式有以下三种，第一种是直接返回可用的 conn, 我们现在先关注第二种，尝试创建 conn，并控制着进程中所有的线程创建的连接数。

#/activerecord/lib/active_record/connection_adapters/abstract/connection_pool.rb#706

# Acquire a connection by one of 1) immediately removing one
# from the queue of available connections, 2) creating a new
# connection if the pool is not at capacity, 3) waiting on the
# queue for a connection to become available.
def acquire_connection(checkout_timeout)
  if conn = @available.poll || try_to_checkout_new_connection
    conn
  else
    reap
    @available.poll(checkout_timeout)
  end
end

#/activerecord/lib/active_record/connection_adapters/abstract/connection_pool.rb#739

# If the pool is not at a +@size+ limit, establish new connection. Connecting
# to the DB is done outside main synchronized section.
def try_to_checkout_new_connection
  # first in synchronized section check if establishing new conns is allowed
  # and increment @now_connecting, to prevent overstepping this pool's @size
  # constraint
  do_checkout = synchronize do
    if @threads_blocking_new_connections.zero? && (@connections.size + @now_connecting) < @size
      @now_connecting += 1
    end
  end
  if do_checkout
    begin
      conn = checkout_new_connection
    ensure
      synchronize do
        if conn
          adopt_connection(conn)
          # returned conn needs to be already leased
          conn.lease
        end
        @now_connecting -= 1
      end
    end
  end
end

当没有可用的 conns，并且也已经达到创建 conn 的上限时，该怎么办呢？这时就会等待使用中的 conns 的释放，等待超时了就会报错

#/activerecord/lib/active_record/connection_adapters/abstract/connection_pool.rb#147
def poll(timeout = nil)
  synchronize { internal_poll(timeout) }
end

def internal_poll(timeout)
  no_wait_poll || (timeout && wait_poll(timeout))
end

def wait_poll(timeout)
  @num_waiting += 1

  t0 = Time.now
  elapsed = 0
  loop do
    @cond.wait(timeout - elapsed)

    return remove if any?

    elapsed = Time.now - t0
    if elapsed >= timeout
      msg = 'could not obtain a connection from the pool within %0.3f seconds (waited %0.3f seconds); all pooled connections were in use' %
        [timeout, elapsed]
      raise ConnectionTimeoutError, msg
    end
  end
ensure
  @num_waiting -= 1
end
end

总结

1. 当使用多线程的 rails 服务器时，应当注意：连接池的数量与线程数量相同。
2. 当连接池数小于线程数并且高负载时，某些线程会等待 conn 的释放，而影响访问速度与性能。具体情况可以查看该帖子：使用 Newrelic 与 ab 工具 尝试了解项目的性能
3. 当偶尔发生 ActiveRecord::ConnectionTimeoutError 错误时，请检查连接池数
4. puma 配置中默认线程是 0-16，数据库连接池默认是 5。可能大多人在并发高的时候都需要增加连接池。