dash 4.0.5 → 4.0.8

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ee8ed58f4d81341ced43dc7d970c081a4c92becc3492860cf3ec8fd1a80687ec
-  data.tar.gz: 588e1b33ca3f041a54beb773472cab9b8144892a32972b71ae9c58306cbb8612
+  metadata.gz: a417156134dd79a4724f87192f70baa55fdbc4478a115e021845e3742bacee88
+  data.tar.gz: ebdbb8d7f192b692c31c8cdaa9834584c01dec3653b744b1a481672fee179d28
 SHA512:
-  metadata.gz: b24d5758340f3e3b7b7b6ecf1df765ffabab7360ddd5bc0df99dad3b90ba3662fc65dcbbe6bfa99af29dea90efb977f4b3d5a2d68454c1e652d57602857f9166
-  data.tar.gz: bbc5bfea941e10310fd01776548b462e9e0e82d4228ed25bb66a0c16b35083979473cb2232ccb7bab3820ccaa45ae32303f424ffebfe76ac321b84354c54cf2e
+  metadata.gz: 3b263730b7ccf4eeeb857304477945fd9d2476bdf52cdb30a964b12b09143b36b5f585ae672d776eac29a6e009606add151aa6e9dd98385f9146980bbc40a9db
+  data.tar.gz: a3ebb3e0141a0f692bd70205aa3d438f9a3cbd1022d1b92c229bf1daf19294c4826a7420c0fefbb077f5a6d6cb46016ee6ab3926bb08f11383bb7a6eb0aaa23e

data/lib/dash/cli/doctor/host_checks.rb

@@ -113,7 +113,12 @@ class Dash::Cli::Doctor::HostChecks
       elsif version.nil?
         result :proxy_version, :ok, "not running (will be started on deploy)"
       elsif Dash::Utils.older_version?(version, minimum)
-        result :proxy_version, :fail, "#{version} is older than the minimum #{minimum}, run `dash proxy reboot` to update"
+        if deploy_reboots_proxy?
+          result :proxy_version, :ok,
+            "#{version} is older than the minimum #{minimum}; the next deploy reboots the proxy to update it"
+        else
+          result :proxy_version, :fail, "#{version} is older than the minimum #{minimum}, run `dash proxy reboot` to update"
+        end
       else
         result :proxy_version, :ok, "#{version} (minimum #{minimum})"
       end
@@ -121,6 +126,18 @@ class Dash::Cli::Doctor::HostChecks
       result :proxy_version, :warn, "running image tag #{version} is not a version number"
     end
 
+    # Mirrors `proxy boot`: with reboot_on_deploy enabled it reboots a drifted
+    # proxy on its own, and a MINIMUM_VERSION bump moves the expected config
+    # digest, so a merely-stale proxy is converged by the very deploy this
+    # doctor gates. Only a version the deploy would not touch — pinned in the
+    # config, or with automatic reboots disabled — is a real failure. Any error
+    # probing drift falls back to the conservative answer.
+    def deploy_reboots_proxy?
+      DASH.config.proxy.reboot_on_deploy? && Dash::Cli::Proxy::Drift.new(host, sshkit).drifted?
+    rescue StandardError
+      false
+    end
+
     # What a container path has to look like to count as a container runtime
     # socket when the config no longer names one.
     DOCKER_SOCKET_PATTERN = %r{docker\.sock\z}

data/lib/dash/configuration/docs/ssh.yml

@@ -79,3 +79,11 @@ ssh:
   # host or tunnel that does not support agent forwarding (for example,
   # Cloudflare Access for Infrastructure with SSH).
   forward_agent: false
+
+  # Connect timeout
+  #
+  # Seconds to wait for the TCP connection and the SSH handshake with a
+  # host before giving up with `Net::SSH::ConnectionTimeout`. Defaults to
+  # 30. Without a bound, a host that accepts the connection but never
+  # answers the handshake hangs the deploy indefinitely.
+  connect_timeout: 30

data/lib/dash/configuration/docs/sshkit.yml

@@ -23,7 +23,8 @@ sshkit:
   #
   # A pooled connection that the network dropped while idle is evicted and the command
   # retried once on a fresh connection, so a long wait elsewhere (a server-lock poll, a
-  # loadbalancer reboot) does not fail the next command on another host.
+  # loadbalancer reboot) does not fail the next command on another host. A host that
+  # stops answering keepalives mid-command is not retried; that deploy fails.
   pool_idle_timeout: 300
 
   # DNS retry settings

data/lib/dash/configuration/registry.rb

@@ -4,6 +4,7 @@ class Dash::Configuration::Registry
   def initialize(config:, secrets:, context: "registry")
     @registry_config = config["registry"] || {}
     @secrets = secrets
+    @context = context
     validate! registry_config, context: context, with: Dash::Configuration::Validator::Registry
   end
 
@@ -28,13 +29,33 @@ class Dash::Configuration::Registry
   end
 
   private
-    attr_reader :registry_config, :secrets
+    attr_reader :registry_config, :secrets, :context
 
+    # A blank credential would reach the remote host as an empty `docker login -p`,
+    # where docker falls back to prompting and dies with the opaque "Cannot perform
+    # an interactive login from a non TTY device" — fail locally instead.
     def lookup(key)
+      value =
+        if registry_config[key].is_a?(Array)
+          secrets[registry_config[key].first]
+        else
+          registry_config[key]
+        end
+
+      validate_present! key, value unless local?
+      value
+    end
+
+    def validate_present!(key, value)
+      return if value.present?
+
       if registry_config[key].is_a?(Array)
-        secrets[registry_config[key].first]
+        secret = registry_config[key].first
+        raise Dash::ConfigurationError,
+          "#{context}/#{key}: secret '#{secret}' resolved to an empty value — " \
+          "if your secrets file forwards it from the environment (#{secret}=$#{secret}), export the variable before running dash"
       else
-        registry_config[key]
+        raise Dash::ConfigurationError, "#{context}/#{key}: is blank"
       end
     end
 end

data/lib/dash/configuration/ssh.rb

@@ -57,8 +57,15 @@ class Dash::Configuration::Ssh
     ssh_config["forward_agent"]
   end
 
+  # net-ssh only bounds the TCP connect and the version/key exchange when
+  # `timeout` is set. Without it, a host that accepts the TCP connection but
+  # never finishes the handshake blocks the deploy forever.
+  def connect_timeout
+    ssh_config.fetch("connect_timeout", 30)
+  end
+
   def options
-    { user: user, port: port, proxy: proxy, logger: logger, keepalive: true, keepalive_interval: 30, keys_only: keys_only, keys: keys, key_data: key_data, config: config, forward_agent: forward_agent }.compact
+    { user: user, port: port, proxy: proxy, logger: logger, keepalive: true, keepalive_interval: 30, keys_only: keys_only, keys: keys, key_data: key_data, config: config, forward_agent: forward_agent, timeout: connect_timeout }.compact
   end
 
   def to_h

data/lib/dash/sshkit_with_ext.rb

@@ -181,8 +181,15 @@ class SSHKit::Backend::Netssh
   # The block is rerun, so a drop *mid-command* runs the command twice. That
   # is accepted: the errors here are the idle-drop ones, where nothing reached
   # the server, and the alternative is the deploy failing on the next host.
+  #
+  # Net::SSH::Timeout (a Disconnect subclass) is deliberately excluded. net-ssh
+  # raises it when a host has ignored `keepalive_maxcount` keepalives in a row,
+  # which is a host that stopped answering mid-command, not a dropped idle
+  # socket. Retrying that reconnects to a host that has just proven unresponsive
+  # and, if it accepts the connection but never finishes the handshake or the
+  # command, hangs the deploy with no further guard. Let it fail.
   module ReconnectOnStaleConnection
-    STALE_CONNECTION_ERRORS = [ Errno::ECONNRESET, Errno::EPIPE, Net::SSH::Disconnect, Net::SSH::Timeout ].freeze
+    STALE_CONNECTION_ERRORS = [ Errno::ECONNRESET, Errno::EPIPE, Net::SSH::Disconnect ].freeze
 
     private
       def with_ssh
@@ -191,12 +198,12 @@ class SSHKit::Backend::Netssh
         begin
           super do |ssh|
             yield ssh
-          rescue *STALE_CONNECTION_ERRORS
-            evict_stale_session(ssh)
+          rescue *STALE_CONNECTION_ERRORS => e
+            evict_stale_session(ssh) if stale_connection_error?(e)
             raise
           end
         rescue *STALE_CONNECTION_ERRORS => e
-          raise if reconnected
+          raise if reconnected || !stale_connection_error?(e)
 
           reconnected = true
           SSHKit.config.output.warn("Reconnecting to #{host}: #{e.message}")
@@ -204,6 +211,10 @@ class SSHKit::Backend::Netssh
         end
       end
 
+      def stale_connection_error?(error)
+        !error.is_a?(Net::SSH::Timeout)
+      end
+
       # `close` waits for channel-close acknowledgements, which never come over
       # a dead socket. `shutdown!` just closes the socket, and a closed session
       # is what makes the pool drop it instead of caching it again.

data/lib/dash/version.rb

@@ -1,3 +1,3 @@
 module Dash
-  VERSION = "4.0.5"
+  VERSION = "4.0.8"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: dash
 version: !ruby/object:Gem::Version
-  version: 4.0.5
+  version: 4.0.8
 platform: ruby
 authors:
 - Mikael Henriksson