Cleaning ec2.py cache after instances launch, hints in readme.

Author: arssher

devops/ec2.yml

@@ -8,19 +8,23 @@
     key_name: aws
     security_group: cluster
     count: 3
-    vpc_subnet_id: subnet-99ba45d4
+    vpc_subnet_id: subnet-e0e2d99a
     region: eu-central-1
 
   tasks:
 
-  # image here is basic ubuntu backed by ebs
+  # image here is ubuntu xenial backed by ebs
   - import_tasks: tasks/ec2.yml instance_type="t2.micro" image="ami-1e339e71"
     tags:
       - micro
 
-  - import_tasks: tasks/clean_ec2_cache.yml
+  # image here is ubuntu xenial backed by instance store
+  - import_tasks: tasks/ec2.yml instance_type="c3.2xlarge" image="ami-4199282e"
     tags:
-      - terminate
+      - c3.2xlarge
+
+  # good for both launch and termination
+  - import_tasks: tasks/clean_ec2_cache.yml
 
   - name: terminate all ec2 instances
     ec2:

devops/readme.txt

@@ -43,6 +43,8 @@ Same on hardcoded nodes:
 ansible-playbook -i inventory_manual --tags "build_pg"
 Provision only the second node:
 ansible-playbook -i inventory_ec2 provision.yml --limit nodes[1]
+Build PG, but don't remove src and build with O0:
+ansible-playbook -i inventory_ec2/ provision.yml --skip-tags "rm_pg_src" -e "cflags='-O0'"
 
 Reload configs:
 ansible-playbook -i inventory_ec2/ send_config.yml
@@ -78,6 +80,9 @@ Currently pgbench exits on first error. postgres_fdw currently supports only
 repeatable read / serializable isolation levels which immediately leads to
 serialization errors, so you should use either patched postgres_fdw or pgbench.
 
+If you don't want to measure the dist performance, keep data dir on tmpfs or
+turn of fsync, the effect it similar.
+
 Things that made me wonder during writing this:
 * Reusability of tasks. Playbooks, files with tasks, roles, includes, imports,
   old include API...probably working directly from Python would be easier? There