Merge branch 'master' of github.com:zalando/spilo

docs/README.md

@@ -1,3 +1,3 @@
 Introduction
 ============
-Spilo (სფილო, small elephants) is a Highly Available PostgreSQL cluster (HA-cluster). It will run a number of clusters with one cluster being a master and the others being slaves. Its purpose is to provide a very resilient, highly available PostgreSQL cluster which can be configured and started within minutes.
+Spilo (სპილო, elephant) is a Highly Available PostgreSQL cluster (HA-cluster). It will run a number of clusters with one cluster being a master and the others being slaves. Its purpose is to provide a very resilient, highly available PostgreSQL cluster which can be configured and started within minutes.

docs/User's Guide/Deployment.md

@@ -0,0 +1,79 @@
+Prerequisites
+=============
+
+You will need a VPC and be allowed to create new infrastructure.
+
+To deploy spilo to AWS you will need the tooling from [stups](http://stups.readthedocs.org/en/latest).
+The gist of it is to:
+
+* have an AWS account with enough rights for deployment
+* install Python 3.4
+* install `stups-mai`
+* install `stups-senza`
+
+Before going any further ensure you can login to your VPC using the `stups` tooling.
+
+Deploying etcd using senza
+==========================
+
+Deploying etcd should be a once in a VPC-lifetime thing. We can use `senza` to deploy the etcd appliance.
+If you already have other spilo-instances running, you may reuse the already running etcd-appliance.
+
+The following prerequisites need to be met.
+
+* `stups-senza`
+* A [Senza Definition](http://stups.readthedocs.org/en/latest/components/senza.html#senza-definition), use the provided `etcd-appliance.yaml` as an example.
+* The Security Group defined in the Senza Definition needs to be created
+
+To deploy the etcd-appliance, use the following:
+
+	senza create DEFINITION.yaml VERSION HOSTED_ZONE DOCKER_IMAGE
+
+This will create and execute a cloud formation template for you.
+
+Example:
+
+Argument   		   | Value
+-------------------|-------
+Definition         | etcd-appliance.yaml
+Hosted zone 	   | repository.example.com
+Version 		   | 1
+Docker repository  | docker.registry.example.com
+Docker image       | repository/etcd-appliance
+Image tag          | 0.2-SNAPSHOT
+
+	senza create etcd-cluster.yaml 1 repository.example.com docker.registry.example.com/repository/etcd-appliance:0.2-SNAPSHOT
+
+Deploying Spilo using senza
+===========================
+
+* Have some global idea about the usage characteristics of the appliance
+* Have unique name for the cluster
+
+You can use senza init to create a senza definition for the spilo appliance,
+the `ETCD_DISCOVERY_URL` should point to `HOSTED_ZONE` from the etcd-appliance that you want to use.
+
+	senza init DEFINITION.yaml
+
+Choose postgresapp as the template, senza will now prompt you for some information, you may want to override the defaults.
+
+If you want, now is the time to edit `DEFINITION.yaml` to your needs.
+
+To deploy the appliance using senza, do the following (we use `CLUSTER_NAME` for the `VERSION` that senza requires):
+
+	senza create [OPTIONS] DEFINITION.yaml CLUSTER_NAME DOCKER_IMAGE
+
+Example:
+
+Argument   		   | Value
+-------------------|-------
+Definition		   | spilo.yaml
+Cluster Name	   | pompeii
+Docker repository  | docker.registry.example.com
+Docker image       | repository/spilo
+Image tag          | 0.7-SNAPSHOT
+
+	senza create spilo.yaml pompeii docker.registry.example.com/repository/spilo:0.7-SNAPSHOT
+
+You can now monitor the progress using:
+	senza watch -n 2 DEFINITION.yaml CLUSTER_NAME

postgres-appliance/Dockerfile

@@ -33,8 +33,7 @@ RUN usermod -d /home/postgres -m postgres
 
 # install WAL-e
 RUN apt-get install build-essential -y
-RUN apt-get install python-pip -y
-RUN apt-get install python-dev libxml2-dev libxslt-dev libffi-dev lzop pv -y
+RUN apt-get install python-pip python-dev libxml2-dev libxslt-dev libffi-dev lzop pv daemontools -y
 RUN pip install six --upgrade
 RUN pip install wal-e
 
@@ -43,18 +42,15 @@ ENV ETCDVERSION 2.0.9
 RUN apt-get install curl -y
 RUN curl -L https://github.com/coreos/etcd/releases/download/v${ETCDVERSION}/etcd-v${ETCDVERSION}-linux-amd64.tar.gz -o etcd-v${ETCDVERSION}-linux-amd64.tar.gz && tar vzxf etcd-v${ETCDVERSION}-linux-amd64.tar.gz && cp etcd-v${ETCDVERSION}-linux-amd64/etcd* /bin/
 
-# install haproxy
-RUN apt-get install haproxy
-
 ENV PATH $PATH:/usr/lib/postgresql-${PGVERSION}/bin
 
 ADD postgres_ha.sh /home/postgres/
-RUN chown postgres:postgres /home/postgres/postgres_ha.sh
+RUN chown postgres:postgres /home/postgres -R
 RUN chmod 700 /home/postgres/postgres_ha.sh
 
-ENV ETCD_ADDRESS=""
-ENV ETCD_DISCOVERY_URL postgres.acid.zalan.do
+ENV ETCD_DISCOVERY_URL postgres.acid.example.com
 ENV SCOPE test
+ENV WAL_S3_BUCKET spilo-example-com
 ENV DEBUG 0
 # run subsequent commands as user postgres
 USER postgres

postgres-appliance/TODO

@@ -9,7 +9,7 @@ Done:
 Done:
    run etcd proxy instead of full-scale etcd
 
-Cancelled, since we are not going to use haproxye:
+Cancelled, since we are not going to use haproxy:
   configure haproxy to fetch information from
   the etcd-proxy and point to the acting master
 

postgres-appliance/postgres_ha.sh

@@ -1,20 +1,21 @@
 #!/bin/bash
 
 PATH=$PATH:/usr/lib/postgresql/${PGVERSION}/bin
+WALE_ENV_DIR=/home/postgres/etc/wal-e.d/env
 
 function write_postgres_yaml
 {
-  local_address=$(cat /etc/hosts |grep ${HOSTNAME}|cut -f1)
+  aws_private_ip=$(curl -s http://169.254.169.254/latest/meta-data/local-ipv4)
   cat >> postgres.yml <<__EOF__
 loop_wait: 10
-aws_use_host_address: "on"
 etcd:
   scope: $SCOPE
   ttl: 30
   host: 127.0.0.1:8080
 postgresql:
   name: postgresql_${HOSTNAME}
-  listen: ${local_address}:5432
+  listen: 0.0.0.0:5432
+  connect_address: ${aws_private_ip}:5432
   data_dir: $PGDATA
   replication:
     username: standby
@@ -28,31 +29,35 @@ postgresql:
   parameters:
     archive_mode: "on"
     wal_level: hot_standby
-    archive_command: /bin/true
+    archive_command: "envdir ${WALE_ENV_DIR} wal-e --aws-instance-profile wal-push \"%p\" -p 1"
     max_wal_senders: 5
     wal_keep_segments: 8
     archive_timeout: 1800s
     max_replication_slots: 5
     hot_standby: "on"
+  recovery_conf:
+    restore_command: "envdir ${WALE_ENV_DIR} wal-e --aws-instance-profile wal-fetch \"%f\" \"%p\" -p 1"
 __EOF__
 }
 
+function write_archive_command_environment
+{
+  mkdir -p ${WALE_ENV_DIR}
+  echo "s3://${WAL_S3_BUCKET}/spilo/${SCOPE}/wal/" > ${WALE_ENV_DIR}/WALE_S3_PREFIX
+}
+
 # get governor code
 git clone https://github.com/zalando/governor.git
 
 write_postgres_yaml
 
+write_archive_command_environment
+
 # start etcd proxy
 # for the -proxy on TDB the url of the etcd cluster
 [ "$DEBUG" -eq 1 ] && exec /bin/bash
 
-if [[ -n $ETCD_ADDRESS ]]
-then
-  # address is still useful for local debugging
-  etcd -name "proxy-$SCOPE" -proxy on -bind-addr 127.0.0.1:8080 --data-dir=etcd -initial-cluster $ETCD_ADDRESS &
-else
-  etcd -name "proxy-$SCOPE" -proxy on -bind-addr 127.0.0.1:8080 --data-dir=etcd -discovery-srv $ETCD_DISCOVERY_URL &
-fi
+etcd -name "proxy-$SCOPE" -proxy on -bind-addr 127.0.0.1:8080 --data-dir=etcd -discovery-srv $ETCD_DISCOVERY_URL &
 
 exec governor/governor.py "/home/postgres/postgres.yml"