fix(services): manage dependencies manually

Since fleet's dependency management cross-cluster is lacking
(see coreos/fleet#338), we resort to starting the services in
a certain order. We also remove Requires=/After= if the dependency
could potentially exist on another server in the cluster.
We're essentially waiting for X-Requires/X-After to be implemented
in fleet.

Also, we make the contrib/ scripts use this Makefile to start
services so we don't repeat logic. Lastly, fleet version checking
is added to the Makefile so that it will error if fleetctl locally
is a different version than the server - many unstable/incomplete
changes are in master.

fixes #810, #805

Author: carmstrong

Makefile

@@ -2,54 +2,104 @@
 # Deis Makefile
 #
 
+ifndef FLEETCTL_TUNNEL
+	$(error You need to set FLEETCTL_TUNNEL to the IP address of a server in the cluster.)
+endif
+
 ifndef DEIS_NUM_INSTANCES
-    DEIS_NUM_INSTANCES = 1
+	DEIS_NUM_INSTANCES = 1
 endif
 
+# TODO refactor to support non-vagrant installations, since this Makefile
+# is now used by the various contrib/ scripts.
 define ssh_all
-  i=1 ; while [ $$i -le $(DEIS_NUM_INSTANCES) ] ; do \
-      vagrant ssh deis-$$i -c $(1) ; \
-      i=`expr $$i + 1` ; \
-  done
+	i=1 ; while [ $$i -le $(DEIS_NUM_INSTANCES) ] ; do \
+			vagrant ssh deis-$$i -c $(1) ; \
+			i=`expr $$i + 1` ; \
+	done
 endef
 
-# ordered list of deis components
-# we don't manage the router if we're setting up a local cluster
+define echo_yellow
+	@echo "\033[0;33m$(subst ",,$(1))\033[0m"
+endef
+
+# due to scheduling problems with fleet 0.2.0, start order of components
+# is fragile. hopefully this can be changed soon...
+ALL_COMPONENTS=builder cache controller database logger registry router
+
 ifeq ($(DEIS_NUM_INSTANCES),1)
-	COMPONENTS=builder cache controller database logger registry router
+	export SKIP_ROUTER=false
+	START_COMPONENTS=registry logger cache database router
 else
-	COMPONENTS=builder cache controller database logger registry
+	export SKIP_ROUTER=true
+	START_COMPONENTS=registry logger cache database
 endif
 
-UNIT_FILES = $(foreach C, $(COMPONENTS), $(wildcard $(C)/systemd/*))
+ALL_UNITS = $(foreach C, $(ALL_COMPONENTS), $(wildcard $(C)/systemd/*))
+START_UNITS = $(foreach C, $(START_COMPONENTS), $(wildcard $(C)/systemd/*))
 
 all: build run
 
-pull:
-	$(call ssh_all,'for c in $(COMPONENTS); do docker pull deis/$$c; done')
-
 build:
-	$(call ssh_all,'cd share && for c in $(COMPONENTS); do cd $$c && docker build -t deis/$$c . && cd ..; done')
+	$(call ssh_all,'cd share && for c in $(ALL_COMPONENTS); do cd $$c && docker build -t deis/$$c . && cd ..; done')
 
-install:
-	fleetctl --strict-host-key-checking=false submit $(UNIT_FILES)
+check-fleet:
+	@LOCAL_VERSION=`fleetctl -version`; \
+	REMOTE_VERSION=`ssh core@$(FLEETCTL_TUNNEL) fleetctl -version`; \
+	if [ "$$LOCAL_VERSION" != "$$REMOTE_VERSION" ]; then \
+			echo "Your fleetctl client version should match the server. Local version: $$LOCAL_VERSION, server version: $$REMOTE_VERSION. Uninstall your local version and install the latest build from https://github.com/coreos/fleet/releases"; exit 1; \
+	fi
+
+clean: uninstall
+	$(call ssh_all,'for c in $(ALL_COMPONENTS); do docker rm -f deis-$$c; done')
 
-uninstall:  stop
-	fleetctl --strict-host-key-checking=false destroy $(UNIT_FILES)
+full-clean: clean
+	$(call ssh_all,'for c in $(ALL_COMPONENTS); do docker rmi deis-$$c; done')
 
-start:
-	echo "\033[0;33mStarting services can take some time... grab some coffee!\033[0m"
-	fleetctl --strict-host-key-checking=false start $(UNIT_FILES)
+install: check-fleet
+	fleetctl --strict-host-key-checking=false submit $(START_UNITS)
 
-stop:
-	fleetctl --strict-host-key-checking=false stop $(UNIT_FILES)
+pull:
+	$(call ssh_all,'for c in $(ALL_COMPONENTS); do docker pull deis/$$c; done')
 
 restart: stop start
 
 run: install start
 
-clean: uninstall
-	$(call ssh_all,'for c in $(COMPONENTS); do docker rm -f deis-$$c; done')
+start: check-fleet
+	fleetctl --strict-host-key-checking=false start $(START_UNITS)
+	$(call echo_yellow,"Use 'make status' to monitor these services")
+	$(call echo_yellow,"Run 'make start-builder' to continue once all are running")
 
-full-clean: clean
-	$(call ssh_all,'for c in $(COMPONENTS); do docker rmi deis-$$c; done')
+start-builder: check-fleet
+	fleetctl --strict-host-key-checking=false submit builder/systemd/*
+	fleetctl --strict-host-key-checking=false start builder/systemd/*
+	$(call echo_yellow,"Use 'make status' to monitor the service")
+	$(call echo_yellow,"Run 'make start-controller' to continue once the service is running")
+
+start-controller: check-fleet
+	fleetctl --strict-host-key-checking=false submit controller/systemd/*
+	fleetctl --strict-host-key-checking=false start controller/systemd/*
+	$(call echo_yellow,"Use 'make status' to monitor the service")
+	@if [ "$$SKIP_ROUTER" == true ]; then \
+		echo "\033[0;33mYou'll need to configure DNS and start the router manually for multi-node clusters.\033[0m" ; \
+		echo "\033[0;33mRun 'make start-router' to schedule and start deis-router.\033[0m" ; \
+	else \
+		echo "\033[0;33mYour Deis cluster is ready to go once the controller is live! Follow the README to login and use Deis.\033[0m" ; \
+	fi
+
+start-router: check-fleet
+	fleetctl --strict-host-key-checking=false submit router/systemd/*
+	fleetctl --strict-host-key-checking=false start router/systemd/*
+	$(call echo_yellow,"Use 'make status' to monitor the service and note the IP it has been scheduled to.")
+	$(call echo_yellow,"Create a wildcard DNS domain which resolves to this host and use that domain when creating clusters/apps in the README.")
+	$(call echo_yellow,"Your Deis cluster is ready to go! Follow the README to login and use Deis.")
+
+status: check-fleet
+	fleetctl --strict-host-key-checking=false list-units
+
+stop: check-fleet
+	fleetctl --strict-host-key-checking=false stop $(ALL_UNITS)
+
+uninstall: check-fleet stop
+	fleetctl --strict-host-key-checking=false destroy $(ALL_UNITS)

README.md

@@ -26,8 +26,7 @@ Trying out Deis? Continue following these instructions for a local cluster setup
 ## Install prerequisites
 On your workstation:
 * Install [Vagrant](http://www.vagrantup.com/downloads.html) and [VirtualBox](https://www.virtualbox.org/wiki/Downloads)
-* Install [Go](http://golang.org/doc/install) and configure your GOPATH, if necessary
-* Install the fleetctl client: `go get github.com/coreos/fleet && go install github.com/coreos/fleet/fleetctl`
+* Install the fleetctl client: Install v0.2.0 from the [fleet GitHub page](https://github.com/coreos/fleet/releases).
 * Install the Docker client if you want to run Docker commands locally (optional)
 
 ## Additional setup for a multi-node cluster

builder/systemd/deis-builder.service

@@ -1,7 +1,5 @@
 [Unit]
 Description=deis-builder
-Requires=deis-registry.service
-After=deis-registry.service
 
 [Service]
 TimeoutStartSec=20m
@@ -13,5 +11,7 @@ ExecStop=/usr/bin/docker rm -f deis-builder
 [Install]
 WantedBy=multi-user.target
 
+# We only need this so that when we start controller, it's guaranteed to be scheduled
+# (If logger and builder are on separate machines, controller can never be scheduled)
 [X-Fleet]
-X-ConditionMachineOf=deis-controller.service
+X-ConditionMachineOf=deis-logger.service

contrib/ec2/README.md

@@ -48,12 +48,11 @@ Please wait for it to come up, then run ./initialize-ec2-cluster.sh
 
 ## Initialize the cluster
 Once the cluster is up, get the hostname of any of the machines from EC2, set
-FLEETCTL_TUNNEL, then run [the init script][init-script] to bootstrap the cluster
-remotely:
+FLEETCTL_TUNNEL, and issue a `make run` from the project root:
 ```console
 $ ssh-add ~/.ssh/id_rsa
 $ export FLEETCTL_TUNNEL=ec2-12-345-678-90.us-west-1.compute.amazonaws.com
-$ ./initialize-ec2-cluster.sh
+$ cd ../.. && make run
 The authenticity of host '54.215.248.50:22' can't be established.
 RSA key fingerprint is 86:10:74:b9:6a:ee:3b:21:d0:0f:b4:63:cc:10:64:c9.
 Are you sure you want to continue connecting (yes/no)? yes
@@ -92,4 +91,3 @@ email: info@opdemand.com
 [template]: https://s3.amazonaws.com/coreos.com/dist/aws/coreos-alpha.template
 [cf-params]: cloudformation.json
 [pro-script]: provision-ec2-cluster.sh
-[init-script]: initialize-ec2-cluster.sh

contrib/ec2/initialize-ec2-cluster.sh

@@ -48,12 +48,11 @@ $ DEIS_NUM_INSTANCES=5 ./provision-rackspace-cluster.sh deis-key
 ```
 
 ### Initialize the cluster
-Once the cluster is up, get the IP address for any of the machines in the cluster, set
-FLEETCTL_TUNNEL, and run [the init script](initialize-rackspace-cluster.sh) to bootstrap the cluster
-remotely:
+Once the cluster is up, get the hostname of any of the machines from EC2, set
+FLEETCTL_TUNNEL, and issue a `make run` from the project root:
 ```console
 $ export FLEETCTL_TUNNEL=23.253.219.94
-$ ./initialize-rackspace-cluster.sh
+$ cd ../.. && make run
 The authenticity of host '23.253.219.94:22' can't be established.
 RSA key fingerprint is ce:3a:c1:3a:ad:11:bd:60:84:8e:60:a8:2f:19:1a:a6.
 Are you sure you want to continue connecting (yes/no)? yes

contrib/rackspace/README.md

@@ -32,7 +32,7 @@ fi
 
 i=1 ; while [[ $i -le $DEIS_NUM_INSTANCES ]] ; do \
     echo_yellow "Provisioning deis-$i..."
-    supernova production boot --image 6bdbd558-e66c-49cc-9ff3-126e7411f602 --flavor $FLAVOR --key-name $1 --user-data ../coreos/user-data --config-drive true deis-$i ; \
+    supernova production boot --image c431c19c-4c09-48ef-a7c1-f4a43f65a1de --flavor $FLAVOR --key-name $1 --user-data ../coreos/user-data --config-drive true deis-$i ; \
     ((i = i + 1)) ; \
 done
 

contrib/rackspace/initialize-rackspace-cluster.sh

@@ -14,3 +14,4 @@ WantedBy=multi-user.target
 
 [X-Fleet]
 X-ConditionMachineOf=deis-logger.service
+X-ConditionMachineOf=deis-builder.service