Reduce the number of loops in prepare-workspace-git

Our prepare-workspace-git role was trying to express a few related steps
as ansible tasks. The problem is that each one of these steps had to
iterate through the entire project list incurring the ansible startup
time for each task of each loop iteration. We can speed things up a bit
if we use a single loop and a slightly more complicated shell task that
checks if things like directories exist before cloning/initing git
repos.

Note this is still slower than if we removed the Ansible task loop
entirely and move this logic into an ansible module that can run a loop
internally. However, making a change like that is significantly more
involved as we'd likely end up exec'ing git with python and need to
manage logging, errors, and so on. The shell module already handles that
for us in a nice way.

Change-Id: Ic87eb182cc4ca4bd0acdd1aa46c2d72dc1165e90

roles/prepare-workspace-git/tasks/main.yaml

@@ -1,59 +1,26 @@
-- name: Find locally cached git repos
-  stat:
-    path: "{{ cached_repos_root }}/{{ zj_project.canonical_name }}"
-  with_items: "{{ zuul.projects.values() | list }}"
-  loop_control:
-    loop_var: zj_project
-  register: cached_repos
-
-# We do a bare clone here first so that we skip creating a working copy that
-# will be overwritten later anyway.
-- name: Clone cached repo to workspace
+# Do all the steps in a single shell script. This reduces the number of times
+# ansible must loop over the list of projects which reduces the amount of
+# task startup time we incur.
+- name: Set initial repo states in workspace
   shell: |
-    set -e
-    git clone --bare {{ cached_repos_root }}/{{ zj_project.0.canonical_name }} {{ ansible_user_dir }}/{{ zj_project.0.src_dir }}/.git
-    cd {{ ansible_user_dir }}/{{ zj_project.0.src_dir }}
-    git config --local --bool core.bare false
-  args:
-    creates: "{{ ansible_user_dir }}/{{ zj_project.0.src_dir }}"
-  when: zj_project.1.stat.exists
-  with_together:
-    - "{{ zuul.projects.values() | list }}"
-    - "{{ cached_repos.results }}"
-  loop_control:
-    loop_var: zj_project
-  # ANSIBLE0006: If we use the git module, we get warning
-  # ANSIBLE0004 since we  do not give an explicit version
-  tags:
-    - skip_ansible_lint
-
-- name: Initialize non-cached repos
-  command: "git init {{ ansible_user_dir }}/{{ zj_project.0.src_dir }}"
-  args:
-    creates: "{{ ansible_user_dir }}/{{ zj_project.0.src_dir }}"
-  when: not zj_project.1.stat.exists
-  with_together:
-    - "{{ zuul.projects.values() | list }}"
-    - "{{ cached_repos.results }}"
-  loop_control:
-    loop_var: zj_project
-  # ANSIBLE0006: If we use the git module, we get warning
-  # ANSIBLE0004 since we  do not give an explicit version
-  tags:
-    - skip_ansible_lint
-
-- name: Remove origin from local git repos and replace it by the zuul fake origin
-  # To be idempotent, remove origin only if it's found in the local list.
-  shell: |
-    set -e
+    set -ex
+    if [ -d "{{ cached_repos_root }}/{{ zj_project.canonical_name }}" ] ; then
+        git clone --bare {{ cached_repos_root }}/{{ zj_project.canonical_name }} {{ ansible_user_dir }}/{{ zj_project.src_dir }}/.git
+        cd {{ ansible_user_dir }}/{{ zj_project.src_dir }}
+        git config --local --bool core.bare false
+    else
+        git init {{ ansible_user_dir }}/{{ zj_project.src_dir }}
+    fi
+    cd {{ ansible_user_dir }}/{{ zj_project.src_dir }}
     git remote -v | grep origin && git remote rm origin || true
     git remote add origin file:///dev/null
   args:
-    chdir: "{{ ansible_user_dir }}/{{ zj_project.src_dir }}"
+    creates: "{{ ansible_user_dir }}/{{ zj_project.src_dir }}"
   with_items: "{{ zuul.projects.values() | list }}"
   loop_control:
     loop_var: zj_project
-  # ANSIBLE0006: git remote is not supported by ansible module
+  # We're using git in a shell script because it is faster and the module
+  # doesn't support features we need.
   tags:
     - skip_ansible_lint