Workflows-friendly adaptation of CircleCI's build locking https://github.com/bellkev/circle-lock-test

circle-lock.sh

#!/usr/bin/env bash

set -o xtrace -o errexit -o pipefail -o nounset

########################################################################################
# CircleCI's current recommendation for roughly serializing a subset
# of build commands for a given branch
#
# circle discussion thread - https://discuss.circleci.com/t/serializing-deployments/153
# Code from - https://github.com/bellkev/circle-lock-test
########################################################################################


# sets $branch, $job_name, $tag, $rest
parse_args() {
    while [[ $# -gt 0 ]]; do
        case $1 in
            -b|--branch) branch="$2" ;;
            -j|--job-name) job_name="$2" ;;
            -t|--tag) tag="$2" ;;
            *) break ;;
        esac
        shift 2
    done
    rest=("$@")
}

# reads $branch, $tag, $commit_message
should_skip() {
    if [[ "$branch" && "$CIRCLE_BRANCH" != "$branch" ]]; then
        echo "Not on branch $branch. Skipping..."
        return 0
    fi

    if [[ "$job_name" && "$CIRCLE_JOB" != "$job_name" ]]; then
        echo "Not running $job_name. Skipping..."
        return 0
    fi

    if [[ "$tag" && "$commit_message" != *\[$tag\]* ]]; then
        echo "No [$tag] commit tag found. Skipping..."
        return 0
    fi

    return 1
}

# reads $branch, $job_name, $tag
# sets $jq_prog
make_jq_prog() {
    local jq_filters=""

    if [[ $branch ]]; then
        jq_filters+=" and .branch == \"$branch\""
    fi

    if [[ $job_name ]]; then
        jq_filters+=" and (.workflows?.job_name? == \"$job_name\" or .build_parameters?.CIRCLE_JOB? == \"$job_name\")"
    fi

    if [[ $tag ]]; then
        jq_filters+=" and (.subject | contains(\"[$tag]\"))"
    fi

    jq_prog=".[] | select(.build_num < $CIRCLE_BUILD_NUM and (.status | test(\"running|pending|queued\")) $jq_filters) | .build_num"
}


if [[ "$0" != *bats* ]]; then
    set -e
    set -u
    set -o pipefail

    branch=""
    tag=""
    job_name=""
    rest=()
    api_url="https://circleci.com/api/v1/project/$CIRCLE_PROJECT_USERNAME/$CIRCLE_PROJECT_REPONAME?circle-token=$CIRCLE_TOKEN&limit=100"

    parse_args "$@"
    commit_message=$(git log -1 --pretty=%B)
    if should_skip; then exit 0; fi
    make_jq_prog

    echo "Checking for running builds..."

    # unset errexit so we can detect and handle temporary circleci api failures
    set +e
    consecutive_failures=0
    while true; do
        builds=$(curl --fail --silent --connect-timeout 5 --max-time 10 -H "Accept: application/json" "$api_url" | jq "$jq_prog")

        if [[ $? -ne 0 ]]; then
            echo "CircleCI api call failed"
            consecutive_failures=$(($consecutive_failures + 1))
        elif [[ ! -z ${builds} ]]; then
            # reset failure counter
            consecutive_failures=0

            echo "Waiting on builds:"
            echo "$builds"
        else
            break
        fi

        # limit the number of consecutive failures that we're willing to tolerate
        if [[ ${consecutive_failures} -gt 5 ]]; then
            echo "Failed $consecutive_failures consecutive attempts...giving up"
            exit 1
        fi

        echo "Retrying in 10 seconds..."
        sleep 10
    done

    echo "Acquired lock"

    if [[ "${#rest[@]}" -ne 0 ]]; then
        "${rest[@]}"
    fi
fi

FYI The -j|--job-name argument is required.

Super awesome 👍

@jmcvetta, I'd forgotten that I even had this in a public gist, but now that you've reminded me, I've updated it to include some error handling around occasional circleci api failures. We had to add that a year or so back, because we were seeing builds fail due to CircleCI api error causing the lock script to fail, which then bubbled up to the job and forced us to re-run. Tolerating occasional failures like this in the lock script drastically improved our satisfaction w/ this rudimentary approach. Enjoy!