Protocol error when pushing to Heroku from CircleCI

Question

# ./circle.yml

deployment:
  acceptance:
    branch: master
    commands:
      - ./script/heroku_deploy.sh thecatch-staging:
          timeout: 300

# ./script/heroku_deploy.sh

#!/bin/sh -e
APP_NAME=$1

git remote add heroku git@heroku.com:$APP_NAME.git
git fetch heroku
MIGRATION_CHANGES=$(git diff HEAD heroku/master --name-only -- db | wc -l)
echo "$MIGRATION_CHANGES db changes."

PREV_WORKERS=$(heroku ps --app $APP_NAME | grep "^worker." | wc -l | tr -d ' ')

# migrations require downtime so enter maintenance mode
if test $MIGRATION_CHANGES -gt 0; then
  heroku maintenance:on --app $APP_NAME

  # Make sure workers are not running during a migration
  heroku scale worker=0 --app $APP_NAME
fi

# deploy code changes (and implicitly restart the app and any running workers)
git push heroku $CIRCLE_SHA1:refs/heads/master

# run database migrations if needed and restart background workers once finished
if test $MIGRATION_CHANGES -gt 0; then
  heroku run rake db:migrate db:seed --app $APP_NAME
  heroku scale worker=$PREV_WORKERS --app $APP_NAME
  heroku restart --app $APP_NAME
fi

heroku run rake cache:flush --app $APP_NAME
heroku maintenance:off --app $APP_NAME

When I push to master, circleci runs my tests, tests pass and I get the following error while deploying to Heroku:

fatal: protocol error: expected old/new/ref, got 'shallow 0966e85b342cb4daaace954d68c928acf743490f'
fatal: The remote end hung up unexpectedly

Full deploy log:

$ ./script/heroku_deploy.sh thecatch-staging
 Warning: Permanently added 'heroku.com,50.19.85.132' (RSA) to the list of known hosts.

Fetching repository, done.
From heroku.com:thecatch-staging
 * [new branch]      master     -> heroku/master
0 db changes.
Warning: Permanently added the RSA host key for IP address '50.19.85.156' to the list of known hosts.

Fetching repository, done.
Unshallowing... 
remote: Counting objects: 2126, done.
remote: Compressing objects: 100% (614/614), done.
remote: Total 1810 (delta 1255), reused 1663 (delta 1140)
Receiving objects: 100% (1810/1810), 1006.69 KiB | 330.00 KiB/s, done.
Resolving deltas: 100% (1255/1255), completed with 130 local objects.
Counting objects: 55, done.
Delta compression using up to 32 threads.
Compressing objects: 100% (39/39), done.
Writing objects: 100% (41/41), 4.40 KiB | 0 bytes/s, done.
Total 41 (delta 27), reused 0 (delta 0)
fatal: protocol error: expected old/new/ref, got 'shallow 0966e85b342cb4daaace954d68c928acf743490f'
fatal: The remote end hung up unexpectedly
./script/heroku_deploy.sh thecatch-staging returned exit code 128action ./script/heroku_deploy.sh thecatch-staging failed

Answer 1

You are using git push heroku $CIRCLE_SHA1:refs/heads/master instead of git push heroku $CIRCLE_SHA1:master in your code.

Also, you are defining the $CIRCLE_SHA1 variable, but its value is not being assigned anywhere in your script.

Correct these two and check if that solves your problem.

Answer 2

For anyone trying this out, after talking to Circle folks, I had to change the timeout call to be between brackets circle.yml to

deployment:
  production:
    branch: master
    commands:
      - ./script/heroku_deploy.sh myHerokuApp $CIRCLE_SHA1: { timeout: 400 }
      - ./script/heroku_deploy_restart.sh myHerokuApp

My script now has takes variables, appName and CircleSha:

#!/bin/sh -e
APP_NAME=$1
CIRCLE_SHA1=$2

#and so on...

(I also had problems with the PREV_WORKERS line not working as expected so I hard coded it in my case.)

Another issue to consider is that in the case of a timeout you might not be restarting workers or putting application back off maintenance mode.

Thus I move those commands to a second script called heroku_deploy_restart.sh

#!/bin/sh -e
APP_NAME=$1

#restart background workers once finished
heroku scale worker=1 --app $APP_NAME
heroku run rake memcached:flush --app $APP_NAME
heroku maintenance:off --app $APP_NAME