Remove the healthcheck step

lib/kamal/cli/app.rb

@@ -14,12 +14,16 @@ def boot
             end
           end
 
+          #  Primary hosts and roles are returned first, so they can open the barrier
+          barrier = Kamal::Cli::Healthcheck::Barrier.new if KAMAL.roles.many?
+
           on(KAMAL.hosts, **KAMAL.boot_strategy) do |host|
             KAMAL.roles_on(host).each do |role|
-              Kamal::Cli::App::Boot.new(host, role, version, self).run
+              Kamal::Cli::App::Boot.new(host, role, self, version, barrier).run
             end
           end
 
+          #  Tag once the app booted on all hosts
           on(KAMAL.hosts) do |host|
             execute *KAMAL.auditor.record("Tagging #{KAMAL.config.absolute_image} as the latest image"), verbosity: :debug
             execute *KAMAL.app.tag_latest_image

lib/kamal/cli/app/boot.rb

@@ -1,38 +1,37 @@
 class Kamal::Cli::App::Boot
-  attr_reader :host, :role, :version, :sshkit
-  delegate :execute, :capture_with_info, :info, to: :sshkit
-  delegate :uses_cord?, :assets?, to: :role
+  attr_reader :host, :role, :version, :barrier, :sshkit
+  delegate :execute, :capture_with_info, :capture_with_pretty_json, :info, :error, to: :sshkit
+  delegate :uses_cord?, :assets?, :running_traefik?, to: :role
 
-  def initialize(host, role, version, sshkit)
+  def initialize(host, role, sshkit, version, barrier)
     @host = host
     @role = role
     @version = version
+    @barrier = barrier
     @sshkit = sshkit
   end
 
   def run
     old_version = old_version_renamed_if_clashing
 
-    start_new_version
+    wait_at_barrier if queuer?
+
+    begin
+      start_new_version
+    rescue => e
+      close_barrier if gatekeeper?
+      stop_new_version
+      raise
+    end
+
+    release_barrier if gatekeeper?
 
     if old_version
       stop_old_version(old_version)
     end
   end
 
   private
-    def app
-      @app ||= KAMAL.app(role: role, host: host)
-    end
-
-    def auditor
-      @auditor = KAMAL.auditor(role: role)
-    end
-
-    def audit(message)
-      execute *auditor.record(message), verbosity: :debug
-    end
-
     def old_version_renamed_if_clashing
       if capture_with_info(*app.container_id_for_version(version), raise_on_non_zero_exit: false).present?
         renamed_version = "#{version}_replaced_#{SecureRandom.hex(8)}"
@@ -46,12 +45,17 @@ def old_version_renamed_if_clashing
 
     def start_new_version
       audit "Booted app version #{version}"
+
       execute *app.tie_cord(role.cord_host_file) if uses_cord?
       hostname = "#{host.to_s[0...51].gsub(/\.+$/, '')}-#{SecureRandom.hex(6)}"
       execute *app.run(hostname: hostname)
       Kamal::Cli::Healthcheck::Poller.wait_for_healthy(pause_after_ready: true) { capture_with_info(*app.status(version: version)) }
     end
 
+    def stop_new_version
+      execute *app.stop(version: version), raise_on_non_zero_exit: false
+    end
+
     def stop_old_version(version)
       if uses_cord?
         cord = capture_with_info(*app.cord(version: version), raise_on_non_zero_exit: false).strip
@@ -65,4 +69,51 @@ def stop_old_version(version)
 
       execute *app.clean_up_assets if assets?
     end
+
+    def release_barrier
+      if barrier.open
+        info "First #{KAMAL.primary_role} container is healthy on #{host}, booting other roles"
+      end
+    end
+
+    def wait_at_barrier
+      info "Waiting for the first healthy #{KAMAL.primary_role} container before booting #{role} on #{host}..."
+      barrier.wait
+      info "First #{KAMAL.primary_role} container is healthy, booting #{role} on #{host}..."
+    rescue Kamal::Cli::Healthcheck::Error
+      info "First #{KAMAL.primary_role} container is unhealthy, not booting #{role} on #{host}"
+      raise
+    end
+
+    def close_barrier
+      if barrier.close
+        info "First #{KAMAL.primary_role} container is unhealthy on #{host}, not booting other roles"
+        error capture_with_info(*app.logs(version: version))
+        error capture_with_info(*app.container_health_log(version: version))
+      end
+    end
+
+    def barrier_role?
+      role == KAMAL.primary_role
+    end
+
+    def app
+      @app ||= KAMAL.app(role: role, host: host)
+    end
+
+    def auditor
+      @auditor = KAMAL.auditor(role: role)
+    end
+
+    def audit(message)
+      execute *auditor.record(message), verbosity: :debug
+    end
+
+    def gatekeeper?
+      barrier && barrier_role?
+    end
+
+    def queuer?
+      barrier && !barrier_role?
+    end
 end

lib/kamal/cli/healthcheck/barrier.rb

@@ -0,0 +1,31 @@
+class Kamal::Cli::Healthcheck::Barrier
+  def initialize
+    @ivar = Concurrent::IVar.new
+  end
+
+  def close
+    set(false)
+  end
+
+  def open
+    set(true)
+  end
+
+  def wait
+    unless opened?
+      raise Kamal::Cli::Healthcheck::Error.new("Halted at barrier")
+    end
+  end
+
+  private
+    def opened?
+      @ivar.value
+    end
+
+    def set(value)
+      @ivar.set(value)
+      true
+    rescue Concurrent::MultipleAssignmentError
+      false
+    end
+end

lib/kamal/cli/healthcheck/error.rb

@@ -0,0 +1,2 @@
+class Kamal::Cli::Healthcheck::Error < StandardError
+end

lib/kamal/cli/healthcheck/poller.rb

@@ -3,7 +3,6 @@ module Kamal::Cli::Healthcheck::Poller
 
   TRAEFIK_UPDATE_DELAY = 5
 
-  class HealthcheckError < StandardError; end
 
   def wait_for_healthy(pause_after_ready: false, &block)
     attempt = 1
@@ -16,9 +15,9 @@ def wait_for_healthy(pause_after_ready: false, &block)
       when "running" # No health check configured
         sleep KAMAL.config.readiness_delay if pause_after_ready
       else
-        raise HealthcheckError, "container not ready (#{status})"
+        raise Kamal::Cli::Healthcheck::Error, "container not ready (#{status})"
       end
-    rescue HealthcheckError => e
+    rescue Kamal::Cli::Healthcheck::Error => e
       if attempt <= max_attempts
         info "#{e.message}, retrying in #{attempt}s (attempt #{attempt}/#{max_attempts})..."
         sleep attempt
@@ -41,9 +40,9 @@ def wait_for_unhealthy(pause_after_ready: false, &block)
       when "unhealthy"
         sleep TRAEFIK_UPDATE_DELAY if pause_after_ready
       else
-        raise HealthcheckError, "container not unhealthy (#{status})"
+        raise Kamal::Cli::Healthcheck::Error, "container not unhealthy (#{status})"
       end
-    rescue HealthcheckError => e
+    rescue Kamal::Cli::Healthcheck::Error => e
       if attempt <= max_attempts
         info "#{e.message}, retrying in #{attempt}s (attempt #{attempt}/#{max_attempts})..."
         sleep attempt

lib/kamal/cli/main.rb

@@ -42,11 +42,6 @@ def deploy
         say "Ensure Traefik is running...", :magenta
         invoke "kamal:cli:traefik:boot", [], invoke_options
 
-        if KAMAL.config.role(KAMAL.config.primary_role).running_traefik?
-          say "Ensure app can pass healthcheck...", :magenta
-          invoke "kamal:cli:healthcheck:perform", [], invoke_options
-        end
-
         say "Detect stale containers...", :magenta
         invoke "kamal:cli:app:stale_containers", [], invoke_options.merge(stop: true)
 
@@ -77,9 +72,6 @@ def redeploy
 
         run_hook "pre-deploy"
 
-        say "Ensure app can pass healthcheck...", :magenta
-        invoke "kamal:cli:healthcheck:perform", [], invoke_options
-
         say "Detect stale containers...", :magenta
         invoke "kamal:cli:app:stale_containers", [], invoke_options.merge(stop: true)
 
@@ -228,9 +220,6 @@ def version
   desc "env", "Manage environment files"
   subcommand "env", Kamal::Cli::Env
 
-  desc "healthcheck", "Healthcheck application"
-  subcommand "healthcheck", Kamal::Cli::Healthcheck
-
   desc "lock", "Manage the deploy lock"
   subcommand "lock", Kamal::Cli::Lock
 
@@ -255,7 +244,7 @@ def container_available?(version)
             raise "Container not found" unless container_id.present?
           end
         end
-      rescue SSHKit::Runner::ExecuteError => e
+      rescue SSHKit::Runner::ExecuteError, SSHKit::Runner::MultipleExecuteError => e
         if e.message =~ /Container not found/
           say "Error looking for container version #{version}: #{e.message}"
           return false

lib/kamal/commands/app/containers.rb

@@ -1,4 +1,6 @@
 module Kamal::Commands::App::Containers
+  DOCKER_HEALTH_LOG_FORMAT    = "'{{json .State.Health}}'"
+
   def list_containers
     docker :container, :ls, "--all", *filter_args
   end
@@ -20,4 +22,10 @@ def rename_container(version:, new_version:)
   def remove_containers
     docker :container, :prune, "--force", *filter_args
   end
+
+  def container_health_log(version:)
+    pipe \
+      container_id_for(container_name: container_name(version)),
+      xargs(docker(:inspect, "--format", DOCKER_HEALTH_LOG_FORMAT))
+  end
 end

lib/kamal/commands/app/logging.rb

@@ -1,7 +1,7 @@
 module Kamal::Commands::App::Logging
-  def logs(since: nil, lines: nil, grep: nil)
+  def logs(version: nil, since: nil, lines: nil, grep: nil)
     pipe \
-      current_running_container_id,
+      version ? container_id_for_version(version) : current_running_container_id,
       "xargs docker logs#{" --since #{since}" if since}#{" --tail #{lines}" if lines} 2>&1",
       ("grep '#{grep}'" if grep)
   end

lib/kamal/commands/base.rb

@@ -3,7 +3,6 @@ class Base
     delegate :sensitive, :argumentize, to: Kamal::Utils
 
     DOCKER_HEALTH_STATUS_FORMAT = "'{{if .State.Health}}{{.State.Health.Status}}{{else}}{{.State.Status}}{{end}}'"
-    DOCKER_HEALTH_LOG_FORMAT    = "'{{json .State.Health}}'"
 
     attr_accessor :config
 

lib/kamal/configuration.rb

@@ -188,7 +188,7 @@ def sshkit
 
 
   def healthcheck
-    { "path" => "/up", "port" => 3000, "max_attempts" => 7, "exposed_port" => 3999, "cord" => "/tmp/kamal-cord", "log_lines" => 50 }.merge(raw_config.healthcheck || {})
+    { "path" => "/up", "port" => 3000, "max_attempts" => 7, "cord" => "/tmp/kamal-cord", "log_lines" => 50 }.merge(raw_config.healthcheck || {})
   end
 
   def healthcheck_service