433 lines
18 KiB
Bash
433 lines
18 KiB
Bash
#!/usr/bin/env bash
|
|
|
|
function retrieve_tests_metadata() {
|
|
mkdir -p $(dirname "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}") $(dirname "${FLAKY_RSPEC_SUITE_REPORT_PATH}") "${RSPEC_PROFILING_FOLDER_PATH}"
|
|
|
|
if [[ -n "${RETRIEVE_TESTS_METADATA_FROM_PAGES}" ]]; then
|
|
if [[ ! -f "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}" ]]; then
|
|
curl --location -o "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}" "https://gitlab-org.gitlab.io/gitlab/${KNAPSACK_RSPEC_SUITE_REPORT_PATH}" ||
|
|
echo "{}" > "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}"
|
|
fi
|
|
|
|
if [[ ! -f "${FLAKY_RSPEC_SUITE_REPORT_PATH}" ]]; then
|
|
curl --location -o "${FLAKY_RSPEC_SUITE_REPORT_PATH}" "https://gitlab-org.gitlab.io/gitlab/${FLAKY_RSPEC_SUITE_REPORT_PATH}" ||
|
|
curl --location -o "${FLAKY_RSPEC_SUITE_REPORT_PATH}" "https://gitlab-org.gitlab.io/gitlab/rspec_flaky/report-suite.json" || # temporary back-compat
|
|
echo "{}" > "${FLAKY_RSPEC_SUITE_REPORT_PATH}"
|
|
fi
|
|
else
|
|
# ${CI_DEFAULT_BRANCH} might not be master in other forks but we want to
|
|
# always target the canonical project here, so the branch must be hardcoded
|
|
local project_path="gitlab-org/gitlab"
|
|
local artifact_branch="master"
|
|
local username="gitlab-bot"
|
|
local job_name="update-tests-metadata"
|
|
local test_metadata_job_id
|
|
|
|
# Ruby
|
|
test_metadata_job_id=$(scripts/api/get_job_id.rb --endpoint "https://gitlab.com/api/v4" --project "${project_path}" -q "status=success" -q "ref=${artifact_branch}" -q "username=${username}" -Q "scope=success" --job-name "${job_name}")
|
|
|
|
if [[ -n "${test_metadata_job_id}" ]]; then
|
|
echo "test_metadata_job_id: ${test_metadata_job_id}"
|
|
|
|
if [[ ! -f "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}" ]]; then
|
|
scripts/api/download_job_artifact.rb --endpoint "https://gitlab.com/api/v4" --project "${project_path}" --job-id "${test_metadata_job_id}" --artifact-path "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}" || echo "{}" > "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}"
|
|
fi
|
|
|
|
if [[ ! -f "${FLAKY_RSPEC_SUITE_REPORT_PATH}" ]]; then
|
|
scripts/api/download_job_artifact.rb --endpoint "https://gitlab.com/api/v4" --project "${project_path}" --job-id "${test_metadata_job_id}" --artifact-path "${FLAKY_RSPEC_SUITE_REPORT_PATH}" ||
|
|
scripts/api/download_job_artifact.rb --endpoint "https://gitlab.com/api/v4" --project "${project_path}" --job-id "${test_metadata_job_id}" --artifact-path "rspec_flaky/report-suite.json" || # temporary back-compat
|
|
echo "{}" > "${FLAKY_RSPEC_SUITE_REPORT_PATH}"
|
|
|
|
# temporary back-compat
|
|
if [[ -f "rspec_flaky/report-suite.json" ]]; then
|
|
mv "rspec_flaky/report-suite.json" "${FLAKY_RSPEC_SUITE_REPORT_PATH}"
|
|
fi
|
|
fi
|
|
else
|
|
echo "test_metadata_job_id couldn't be found!"
|
|
echo "{}" > "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}"
|
|
echo "{}" > "${FLAKY_RSPEC_SUITE_REPORT_PATH}"
|
|
fi
|
|
fi
|
|
}
|
|
|
|
function update_tests_metadata() {
|
|
local rspec_flaky_folder_path="$(dirname "${FLAKY_RSPEC_SUITE_REPORT_PATH}")/"
|
|
local knapsack_folder_path="$(dirname "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}")/"
|
|
|
|
echo "{}" > "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}"
|
|
|
|
scripts/merge-reports "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}" ${knapsack_folder_path}rspec*.json
|
|
|
|
export FLAKY_RSPEC_GENERATE_REPORT="true"
|
|
scripts/merge-reports "${FLAKY_RSPEC_SUITE_REPORT_PATH}" ${rspec_flaky_folder_path}all_*.json
|
|
scripts/flaky_examples/prune-old-flaky-examples "${FLAKY_RSPEC_SUITE_REPORT_PATH}"
|
|
|
|
if [[ "$CI_PIPELINE_SOURCE" == "schedule" ]]; then
|
|
scripts/insert-rspec-profiling-data
|
|
else
|
|
echo "Not inserting profiling data as the pipeline is not a scheduled one."
|
|
fi
|
|
|
|
cleanup_individual_job_reports
|
|
}
|
|
|
|
function retrieve_tests_mapping() {
|
|
mkdir -p $(dirname "$RSPEC_PACKED_TESTS_MAPPING_PATH")
|
|
|
|
if [[ -n "${RETRIEVE_TESTS_METADATA_FROM_PAGES}" ]]; then
|
|
if [[ ! -f "${RSPEC_PACKED_TESTS_MAPPING_PATH}" ]]; then
|
|
(curl --location -o "${RSPEC_PACKED_TESTS_MAPPING_PATH}.gz" "https://gitlab-org.gitlab.io/gitlab/${RSPEC_PACKED_TESTS_MAPPING_PATH}.gz" && gzip -d "${RSPEC_PACKED_TESTS_MAPPING_PATH}.gz") || echo "{}" > "${RSPEC_PACKED_TESTS_MAPPING_PATH}"
|
|
fi
|
|
else
|
|
# ${CI_DEFAULT_BRANCH} might not be master in other forks but we want to
|
|
# always target the canonical project here, so the branch must be hardcoded
|
|
local project_path="gitlab-org/gitlab"
|
|
local artifact_branch="master"
|
|
local username="gitlab-bot"
|
|
local job_name="update-tests-metadata"
|
|
local test_metadata_with_mapping_job_id
|
|
|
|
test_metadata_with_mapping_job_id=$(scripts/api/get_job_id.rb --endpoint "https://gitlab.com/api/v4" --project "${project_path}" -q "status=success" -q "ref=${artifact_branch}" -q "username=${username}" -Q "scope=success" --job-name "${job_name}")
|
|
|
|
if [[ -n "${test_metadata_with_mapping_job_id}" ]]; then
|
|
echo "test_metadata_with_mapping_job_id: ${test_metadata_with_mapping_job_id}"
|
|
|
|
if [[ ! -f "${RSPEC_PACKED_TESTS_MAPPING_PATH}" ]]; then
|
|
(scripts/api/download_job_artifact.rb --endpoint "https://gitlab.com/api/v4" --project "${project_path}" --job-id "${test_metadata_with_mapping_job_id}" --artifact-path "${RSPEC_PACKED_TESTS_MAPPING_PATH}.gz" && gzip -d "${RSPEC_PACKED_TESTS_MAPPING_PATH}.gz") || echo "{}" > "${RSPEC_PACKED_TESTS_MAPPING_PATH}"
|
|
fi
|
|
else
|
|
echo "test_metadata_with_mapping_job_id couldn't be found!"
|
|
echo "{}" > "${RSPEC_PACKED_TESTS_MAPPING_PATH}"
|
|
fi
|
|
fi
|
|
|
|
scripts/unpack-test-mapping "${RSPEC_PACKED_TESTS_MAPPING_PATH}" "${RSPEC_TESTS_MAPPING_PATH}"
|
|
}
|
|
|
|
function retrieve_frontend_fixtures_mapping() {
|
|
mkdir -p $(dirname "$FRONTEND_FIXTURES_MAPPING_PATH")
|
|
|
|
if [[ -n "${RETRIEVE_TESTS_METADATA_FROM_PAGES}" ]]; then
|
|
if [[ ! -f "${FRONTEND_FIXTURES_MAPPING_PATH}" ]]; then
|
|
(curl --location -o "${FRONTEND_FIXTURES_MAPPING_PATH}" "https://gitlab-org.gitlab.io/gitlab/${FRONTEND_FIXTURES_MAPPING_PATH}") || echo "{}" > "${FRONTEND_FIXTURES_MAPPING_PATH}"
|
|
fi
|
|
else
|
|
# ${CI_DEFAULT_BRANCH} might not be master in other forks but we want to
|
|
# always target the canonical project here, so the branch must be hardcoded
|
|
local project_path="gitlab-org/gitlab"
|
|
local artifact_branch="master"
|
|
local username="gitlab-bot"
|
|
local job_name="generate-frontend-fixtures-mapping"
|
|
local test_metadata_with_mapping_job_id
|
|
|
|
test_metadata_with_mapping_job_id=$(scripts/api/get_job_id.rb --endpoint "https://gitlab.com/api/v4" --project "${project_path}" -q "ref=${artifact_branch}" -q "username=${username}" -Q "scope=success" --job-name "${job_name}")
|
|
|
|
if [[ $? -eq 0 ]] && [[ -n "${test_metadata_with_mapping_job_id}" ]]; then
|
|
echo "test_metadata_with_mapping_job_id: ${test_metadata_with_mapping_job_id}"
|
|
|
|
if [[ ! -f "${FRONTEND_FIXTURES_MAPPING_PATH}" ]]; then
|
|
(scripts/api/download_job_artifact.rb --endpoint "https://gitlab.com/api/v4" --project "${project_path}" --job-id "${test_metadata_with_mapping_job_id}" --artifact-path "${FRONTEND_FIXTURES_MAPPING_PATH}") || echo "{}" > "${FRONTEND_FIXTURES_MAPPING_PATH}"
|
|
fi
|
|
else
|
|
echo "test_metadata_with_mapping_job_id couldn't be found!"
|
|
echo "{}" > "${FRONTEND_FIXTURES_MAPPING_PATH}"
|
|
fi
|
|
fi
|
|
}
|
|
|
|
function update_tests_mapping() {
|
|
if ! crystalball_rspec_data_exists; then
|
|
echo "No crystalball rspec data found."
|
|
return 0
|
|
fi
|
|
|
|
scripts/generate-test-mapping "${RSPEC_TESTS_MAPPING_PATH}" crystalball/rspec*.yml
|
|
scripts/pack-test-mapping "${RSPEC_TESTS_MAPPING_PATH}" "${RSPEC_PACKED_TESTS_MAPPING_PATH}"
|
|
gzip "${RSPEC_PACKED_TESTS_MAPPING_PATH}"
|
|
rm -f crystalball/rspec*.yml "${RSPEC_PACKED_TESTS_MAPPING_PATH}"
|
|
}
|
|
|
|
function crystalball_rspec_data_exists() {
|
|
compgen -G "crystalball/rspec*.yml" >/dev/null
|
|
}
|
|
|
|
function retrieve_previous_failed_tests() {
|
|
local directory_for_output_reports="${1}"
|
|
local rspec_pg_regex="${2}"
|
|
local rspec_ee_pg_regex="${3}"
|
|
local pipeline_report_path="test_results/previous/test_reports.json"
|
|
|
|
# Used to query merge requests. This variable reflects where the merge request has been created
|
|
local target_project_path="${CI_MERGE_REQUEST_PROJECT_PATH}"
|
|
local instance_url="${CI_SERVER_URL}"
|
|
|
|
echo 'Attempting to build pipeline test report...'
|
|
|
|
scripts/pipeline_test_report_builder.rb --instance-base-url "${instance_url}" --target-project "${target_project_path}" --mr-id "${CI_MERGE_REQUEST_IID}" --output-file-path "${pipeline_report_path}"
|
|
|
|
echo 'Generating failed tests lists...'
|
|
|
|
scripts/failed_tests.rb --previous-tests-report-path "${pipeline_report_path}" --output-directory "${directory_for_output_reports}" --rspec-pg-regex "${rspec_pg_regex}" --rspec-ee-pg-regex "${rspec_ee_pg_regex}"
|
|
}
|
|
|
|
function rspec_args() {
|
|
local rspec_opts="${1}"
|
|
local junit_report_file="${2:-${JUNIT_RESULT_FILE}}"
|
|
|
|
echo "-Ispec -rspec_helper --color --format documentation --format RspecJunitFormatter --out ${junit_report_file} ${rspec_opts}"
|
|
}
|
|
|
|
function rspec_simple_job() {
|
|
export NO_KNAPSACK="1"
|
|
|
|
local rspec_cmd="bin/rspec $(rspec_args "${1}" "${2}")"
|
|
echoinfo "Running RSpec command: ${rspec_cmd}"
|
|
|
|
eval "${rspec_cmd}"
|
|
}
|
|
|
|
function rspec_db_library_code() {
|
|
local db_files="spec/lib/gitlab/database/"
|
|
|
|
rspec_simple_job "-- ${db_files}"
|
|
}
|
|
|
|
function debug_rspec_variables() {
|
|
echoinfo "SKIP_FLAKY_TESTS_AUTOMATICALLY: ${SKIP_FLAKY_TESTS_AUTOMATICALLY}"
|
|
echoinfo "RETRY_FAILED_TESTS_IN_NEW_PROCESS: ${RETRY_FAILED_TESTS_IN_NEW_PROCESS}"
|
|
|
|
echoinfo "KNAPSACK_GENERATE_REPORT: ${KNAPSACK_GENERATE_REPORT}"
|
|
echoinfo "FLAKY_RSPEC_GENERATE_REPORT: ${FLAKY_RSPEC_GENERATE_REPORT}"
|
|
|
|
echoinfo "KNAPSACK_TEST_FILE_PATTERN: ${KNAPSACK_TEST_FILE_PATTERN}"
|
|
echoinfo "KNAPSACK_LOG_LEVEL: ${KNAPSACK_LOG_LEVEL}"
|
|
echoinfo "KNAPSACK_REPORT_PATH: ${KNAPSACK_REPORT_PATH}"
|
|
|
|
echoinfo "FLAKY_RSPEC_SUITE_REPORT_PATH: ${FLAKY_RSPEC_SUITE_REPORT_PATH}"
|
|
echoinfo "FLAKY_RSPEC_REPORT_PATH: ${FLAKY_RSPEC_REPORT_PATH}"
|
|
echoinfo "NEW_FLAKY_RSPEC_REPORT_PATH: ${NEW_FLAKY_RSPEC_REPORT_PATH}"
|
|
echoinfo "SKIPPED_FLAKY_TESTS_REPORT_PATH: ${SKIPPED_FLAKY_TESTS_REPORT_PATH}"
|
|
echoinfo "RETRIED_TESTS_REPORT_PATH: ${RETRIED_TESTS_REPORT_PATH}"
|
|
|
|
echoinfo "CRYSTALBALL: ${CRYSTALBALL}"
|
|
}
|
|
|
|
function rspec_paralellized_job() {
|
|
read -ra job_name <<< "${CI_JOB_NAME}"
|
|
local test_tool="${job_name[0]}"
|
|
local test_level="${job_name[1]}"
|
|
local report_name=$(echo "${CI_JOB_NAME}" | sed -E 's|[/ ]|_|g') # e.g. 'rspec unit pg12 1/24' would become 'rspec_unit_pg12_1_24'
|
|
local rspec_opts="${1}"
|
|
local spec_folder_prefixes=""
|
|
local rspec_flaky_folder_path="$(dirname "${FLAKY_RSPEC_SUITE_REPORT_PATH}")/"
|
|
local knapsack_folder_path="$(dirname "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}")/"
|
|
local rspec_run_status=0
|
|
|
|
if [[ "${test_tool}" =~ "-ee" ]]; then
|
|
spec_folder_prefixes="'ee/'"
|
|
fi
|
|
|
|
if [[ "${test_tool}" =~ "-jh" ]]; then
|
|
spec_folder_prefixes="'jh/'"
|
|
fi
|
|
|
|
if [[ "${test_tool}" =~ "-all" ]]; then
|
|
spec_folder_prefixes="['', 'ee/', 'jh/']"
|
|
fi
|
|
|
|
export KNAPSACK_LOG_LEVEL="debug"
|
|
export KNAPSACK_REPORT_PATH="${knapsack_folder_path}${report_name}_report.json"
|
|
|
|
# There's a bug where artifacts are sometimes not downloaded. Since specs can run without the Knapsack report, we can
|
|
# handle the missing artifact gracefully here. See https://gitlab.com/gitlab-org/gitlab/-/issues/212349.
|
|
if [[ ! -f "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}" ]]; then
|
|
echo "{}" > "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}"
|
|
fi
|
|
|
|
cp "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}" "${KNAPSACK_REPORT_PATH}"
|
|
|
|
export KNAPSACK_TEST_FILE_PATTERN=$(ruby -r./tooling/quality/test_level.rb -e "puts Quality::TestLevel.new(${spec_folder_prefixes}).pattern(:${test_level})")
|
|
export FLAKY_RSPEC_REPORT_PATH="${rspec_flaky_folder_path}all_${report_name}_report.json"
|
|
export NEW_FLAKY_RSPEC_REPORT_PATH="${rspec_flaky_folder_path}new_${report_name}_report.json"
|
|
export SKIPPED_FLAKY_TESTS_REPORT_PATH="${rspec_flaky_folder_path}skipped_flaky_tests_${report_name}_report.txt"
|
|
export RETRIED_TESTS_REPORT_PATH="${rspec_flaky_folder_path}retried_tests_${report_name}_report.txt"
|
|
|
|
if [[ -d "ee/" ]]; then
|
|
export KNAPSACK_GENERATE_REPORT="true"
|
|
export FLAKY_RSPEC_GENERATE_REPORT="true"
|
|
|
|
if [[ ! -f $FLAKY_RSPEC_REPORT_PATH ]]; then
|
|
echo "{}" > "${FLAKY_RSPEC_REPORT_PATH}"
|
|
fi
|
|
|
|
if [[ ! -f $NEW_FLAKY_RSPEC_REPORT_PATH ]]; then
|
|
echo "{}" > "${NEW_FLAKY_RSPEC_REPORT_PATH}"
|
|
fi
|
|
fi
|
|
|
|
debug_rspec_variables
|
|
|
|
if [[ -n $RSPEC_TESTS_MAPPING_ENABLED ]]; then
|
|
tooling/bin/parallel_rspec --rspec_args "$(rspec_args "${rspec_opts}")" --filter "tmp/matching_tests.txt" || rspec_run_status=$?
|
|
else
|
|
tooling/bin/parallel_rspec --rspec_args "$(rspec_args "${rspec_opts}")" || rspec_run_status=$?
|
|
fi
|
|
|
|
echoinfo "RSpec exited with ${rspec_run_status}."
|
|
|
|
# Experiment to retry failed examples in a new RSpec process: https://gitlab.com/gitlab-org/quality/team-tasks/-/issues/1148
|
|
if [[ $rspec_run_status -ne 0 ]]; then
|
|
if [[ "${RETRY_FAILED_TESTS_IN_NEW_PROCESS}" == "true" ]]; then
|
|
retry_failed_rspec_examples
|
|
rspec_run_status=$?
|
|
fi
|
|
else
|
|
echosuccess "No examples to retry, congrats!"
|
|
fi
|
|
|
|
exit $rspec_run_status
|
|
}
|
|
|
|
function retry_failed_rspec_examples() {
|
|
local rspec_run_status=0
|
|
|
|
# Keep track of the tests that are retried, later consolidated in a single file by the `rspec:flaky-tests-report` job
|
|
local failed_examples=$(grep " failed" ${RSPEC_LAST_RUN_RESULTS_FILE})
|
|
echo "${CI_JOB_URL}" > "${RETRIED_TESTS_REPORT_PATH}"
|
|
echo $failed_examples >> "${RETRIED_TESTS_REPORT_PATH}"
|
|
|
|
echoinfo "Retrying the failing examples in a new RSpec process..."
|
|
|
|
install_junit_merge_gem
|
|
|
|
# Disable Crystalball on retry to not overwrite the existing report
|
|
export CRYSTALBALL="false"
|
|
|
|
# Disable simplecov so retried tests don't override test coverage report
|
|
export SIMPLECOV=0
|
|
|
|
# Retry only the tests that failed on first try
|
|
rspec_simple_job "--only-failures --pattern \"${KNAPSACK_TEST_FILE_PATTERN}\"" "${JUNIT_RETRY_FILE}"
|
|
rspec_run_status=$?
|
|
|
|
# Merge the JUnit report from retry into the first-try report
|
|
junit_merge "${JUNIT_RETRY_FILE}" "${JUNIT_RESULT_FILE}" --update-only
|
|
|
|
exit $rspec_run_status
|
|
}
|
|
|
|
function rspec_rerun_previous_failed_tests() {
|
|
local test_file_count_threshold=${RSPEC_PREVIOUS_FAILED_TEST_FILE_COUNT_THRESHOLD:-10}
|
|
local matching_tests_file=${1}
|
|
local rspec_opts=${2}
|
|
local test_files="$(cat "${matching_tests_file}")"
|
|
local test_file_count=$(wc -w "${matching_tests_file}" | awk {'print $1'})
|
|
|
|
if [[ "${test_file_count}" -gt "${test_file_count_threshold}" ]]; then
|
|
echo "This job is intentionally exited because there are more than ${test_file_count_threshold} test files to rerun."
|
|
exit 0
|
|
fi
|
|
|
|
if [[ -n $test_files ]]; then
|
|
rspec_simple_job "${test_files}"
|
|
else
|
|
echo "No failed test files to rerun"
|
|
fi
|
|
}
|
|
|
|
function rspec_fail_fast() {
|
|
local test_file_count_threshold=${RSPEC_FAIL_FAST_TEST_FILE_COUNT_THRESHOLD:-10}
|
|
local matching_tests_file=${1}
|
|
local rspec_opts=${2}
|
|
local test_files="$(cat "${matching_tests_file}")"
|
|
local test_file_count=$(wc -w "${matching_tests_file}" | awk {'print $1'})
|
|
|
|
if [[ "${test_file_count}" -gt "${test_file_count_threshold}" ]]; then
|
|
echo "This job is intentionally skipped because there are more than ${test_file_count_threshold} test files matched,"
|
|
echo "which would take too long to run in this job."
|
|
echo "All the tests would be run in other rspec jobs."
|
|
exit 0
|
|
fi
|
|
|
|
if [[ -n $test_files ]]; then
|
|
rspec_simple_job "${rspec_opts} ${test_files}"
|
|
else
|
|
echo "No rspec fail-fast tests to run"
|
|
fi
|
|
}
|
|
|
|
function rspec_matched_foss_tests() {
|
|
local test_file_count_threshold=20
|
|
local matching_tests_file=${1}
|
|
local rspec_opts=${2}
|
|
local test_files="$(cat "${matching_tests_file}")"
|
|
local test_file_count=$(wc -w "${matching_tests_file}" | awk {'print $1'})
|
|
|
|
if [[ "${test_file_count}" -gt "${test_file_count_threshold}" ]]; then
|
|
echo "This job is intentionally failed because there are more than ${test_file_count_threshold} FOSS test files matched,"
|
|
echo "which would take too long to run in this job."
|
|
echo "To reduce the likelihood of breaking FOSS pipelines,"
|
|
echo "please add ~\"pipeline:run-as-if-foss\" label to the merge request and trigger a new pipeline."
|
|
echo "This would run all as-if-foss jobs in this merge request"
|
|
echo "and remove this failing job from the pipeline."
|
|
exit 1
|
|
fi
|
|
|
|
if [[ -n $test_files ]]; then
|
|
rspec_simple_job "${rspec_opts} ${test_files}"
|
|
else
|
|
echo "No impacted FOSS rspec tests to run"
|
|
fi
|
|
}
|
|
|
|
function generate_frontend_fixtures_mapping() {
|
|
local pattern=""
|
|
|
|
if [[ -d "ee/" ]]; then
|
|
pattern=",ee/"
|
|
fi
|
|
|
|
if [[ -d "jh/" ]]; then
|
|
pattern="${pattern},jh/"
|
|
fi
|
|
|
|
if [[ -n "${pattern}" ]]; then
|
|
pattern="{${pattern}}"
|
|
fi
|
|
|
|
pattern="${pattern}spec/frontend/fixtures/**/*.rb"
|
|
|
|
export GENERATE_FRONTEND_FIXTURES_MAPPING="true"
|
|
|
|
mkdir -p $(dirname "$FRONTEND_FIXTURES_MAPPING_PATH")
|
|
|
|
rspec_simple_job "--pattern \"${pattern}\""
|
|
}
|
|
|
|
function cleanup_individual_job_reports() {
|
|
local rspec_flaky_folder_path="$(dirname "${FLAKY_RSPEC_SUITE_REPORT_PATH}")/"
|
|
local knapsack_folder_path="$(dirname "${KNAPSACK_RSPEC_SUITE_REPORT_PATH}")/"
|
|
|
|
rm -rf ${knapsack_folder_path}rspec*.json \
|
|
${rspec_flaky_folder_path}all_*.json \
|
|
${rspec_flaky_folder_path}new_*.json \
|
|
${rspec_flaky_folder_path}skipped_flaky_tests_*_report.txt \
|
|
${rspec_flaky_folder_path}retried_tests_*_report.txt \
|
|
${RSPEC_LAST_RUN_RESULTS_FILE} \
|
|
${RSPEC_PROFILING_FOLDER_PATH}/**/*
|
|
rmdir ${RSPEC_PROFILING_FOLDER_PATH} || true
|
|
}
|
|
|
|
function generate_flaky_tests_reports() {
|
|
local rspec_flaky_folder_path="$(dirname "${FLAKY_RSPEC_SUITE_REPORT_PATH}")/"
|
|
|
|
debug_rspec_variables
|
|
|
|
mkdir -p ${rspec_flaky_folder_path}
|
|
|
|
find ${rspec_flaky_folder_path} -type f -name 'skipped_flaky_tests_*_report.txt' -exec cat {} + >> "${SKIPPED_FLAKY_TESTS_REPORT_PATH}"
|
|
find ${rspec_flaky_folder_path} -type f -name 'retried_tests_*_report.txt' -exec cat {} + >> "${RETRIED_TESTS_REPORT_PATH}"
|
|
|
|
cleanup_individual_job_reports
|
|
}
|