feat: detect flaky tests

Makefile

@@ -314,6 +314,7 @@ docker-test: ##@tests Run tests in a docker container with golang.
 test: test-unit ##@tests Run basic, short tests during development
 
 test-unit: export BUILD_TAGS ?=
+test-unit: export UNIT_TEST_FAILFAST ?= true
 # Ensure 'waku' and 'wakuv2' tests are executed first to reduce the impact of flaky tests.
 # Otherwise, the entire target might fail at the end, making re-runs time-consuming.
 test-unit: export UNIT_TEST_PACKAGES ?= $(shell go list ./... | \

_assets/ci/Jenkinsfile

@@ -1,3 +1,4 @@
+#!/usr/bin/env groovy
 library 'status-jenkins-lib@v1.7.0'
 
 pipeline {

_assets/ci/Jenkinsfile.android

@@ -1,3 +1,4 @@
+#!/usr/bin/env groovy
 library 'status-jenkins-lib@v1.7.0'
 
 pipeline {

_assets/ci/Jenkinsfile.docker

@@ -1,3 +1,4 @@
+#!/usr/bin/env groovy
 library 'status-jenkins-lib@v1.7.0'
 
 pipeline {

_assets/ci/Jenkinsfile.ios

@@ -1,3 +1,4 @@
+#!/usr/bin/env groovy
 library 'status-jenkins-lib@v1.7.0'
 
 pipeline {

_assets/ci/Jenkinsfile.linux

@@ -1,3 +1,4 @@
+#!/usr/bin/env groovy
 library 'status-jenkins-lib@v1.7.0'
 
 pipeline {

_assets/ci/Jenkinsfile.tests

@@ -1,3 +1,4 @@
+#!/usr/bin/env groovy
 library 'status-jenkins-lib@v1.7.0'
 
 pipeline {
@@ -9,12 +10,22 @@ pipeline {
       defaultValue: 'develop',
       description: 'Name of branch to build.'
     )
+    string(
+      name: 'UNIT_TEST_COUNT',
+      defaultValue: getDefaultUnitTestCount(),
+      description: 'How many times to run tests?'
+    )
+    booleanParam(
+      name: 'UNIT_TEST_FAILFAST',
+      defaultValue: getDefaultUnitTestFailfast(),
+      description: 'Should the job fail fast on first test failure?'
+    )
   }
 
   options {
     timestamps()
     /* Prevent Jenkins jobs from running forever */
-    timeout(time: 40, unit: 'MINUTES')
+    timeout(time: getDefaultTimeout(), unit: 'MINUTES')
     disableConcurrentBuilds()
     /* manage how many builds we keep */
     buildDiscarder(logRotator(
@@ -99,3 +110,24 @@ pipeline {
     cleanup { dir(env.TMPDIR) { deleteDir() } }
   } // post
 } // pipeline
+
+def getDefaultUnitTestCount() {
+  if (env.JOB_BASE_NAME == 'tests-nightly') {
+    return '10'
+  }
+  return '1'
+}
+
+def getDefaultUnitTestFailfast() {
+  if (env.JOB_BASE_NAME == 'tests-nightly') {
+    return false
+  }
+  return true
+}
+
+def getDefaultTimeout() {
+  if (env.JOB_BASE_NAME == 'tests-nightly') {
+    return 8*60
+  }
+  return 40
+}

_assets/scripts/run_unit_tests.sh

@@ -5,21 +5,53 @@ GIT_ROOT=$(cd "${BASH_SOURCE%/*}" && git rev-parse --show-toplevel)
 
 source "${GIT_ROOT}/_assets/scripts/colors.sh"
 
+if [[ $UNIT_TEST_FAILFAST == 'true' ]]; then
+  GOTEST_EXTRAFLAGS="${GOTEST_EXTRAFLAGS} --failfast"
+fi
+
+if [[ -z "${UNIT_TEST_COUNT}" ]]; then
+  UNIT_TEST_COUNT=1
+fi
+
+redirect_stdout() {
+  output_file=$1
+
+  if [[ "${CI}" == 'true' ]];
+  then
+    cat > "${output_file}";
+  else
+    tee "${output_file}";
+  fi
+}
+
+last_failing_exit_code=0
+
 for package in ${UNIT_TEST_PACKAGES}; do
   echo -e "${GRN}Testing:${RST} ${package}"
   package_dir=$(go list -f "{{.Dir}}" "${package}")
   output_file=${package_dir}/test.log
 
-  go test -tags "${BUILD_TAGS}" -timeout 30m -v -failfast "${package}" ${GOTEST_EXTRAFLAGS} | \
+  go test -count="${UNIT_TEST_COUNT}" -tags "${BUILD_TAGS}" -v "${package}" ${GOTEST_EXTRAFLAGS} | \
-     if [ "${CI}" = "true" ]; then cat > "${output_file}"; else tee "${output_file}"; fi
+    redirect_stdout "${output_file}"
   go_test_exit=$?
 
-  if [ "${CI}" = "true" ]; then
+  if [[ "${CI}" == 'true' ]]; then
     go-junit-report -in "${output_file}" -out "${package_dir}"/report.xml
   fi
 
-  if [ ${go_test_exit} -ne 0 ]; then
+  if [[ "${go_test_exit}" -ne 0 ]]; then
     echo -e "${YLW}Failed, see the log:${RST} ${BLD}${output_file}${RST}"
+    if [[ "$UNIT_TEST_FAILFAST" == 'true' ]]; then
       exit "${go_test_exit}"
     fi
+    last_failing_exit_code="${go_test_exit}"
+  fi
 done
+
+if [[ "${last_failing_exit_code}" -ne 0 ]]; then
+  if [[ "${UNIT_TEST_COUNT}" -gt 1 ]]; then
+    "${GIT_ROOT}/_assets/scripts/test_stats.py"
+  fi
+
+  exit "${last_failing_exit_code}"
+fi

_assets/scripts/test_stats.py

@@ -0,0 +1,36 @@
+#!/usr/bin/env python
+
+import glob
+import xml.etree.ElementTree as ET
+from collections import defaultdict
+
+test_stats = defaultdict(lambda: defaultdict(int))
+
+for file in glob.glob("**/report.xml", recursive=True):
+    tree = ET.parse(file)
+    root = tree.getroot()
+    for testcase in root.iter("testcase"):
+        test_name = testcase.attrib["name"]
+
+        test_stats[test_name]["total"] += 1
+
+        if testcase.find("failure") is not None:
+            test_stats[test_name]["failed"] += 1
+        elif testcase.find("error") is not None:
+            test_stats[test_name]["failed"] += 1
+
+failing_test_stats = [
+    {"name": name, "failure_rate": stats["failed"] / stats["total"]}
+    for name, stats in test_stats.items() if stats["failed"] != 0
+]
+
+sorted_failing_test_stats = sorted(failing_test_stats,
+                                   key=lambda x: x["failure_rate"],
+                                   reverse=True)
+
+print("---")
+print("Failing tests stats")
+print("(test name: failure rate)")
+print("---")
+for test_stat in sorted_failing_test_stats:
+    print(f"{test_stat['name']}: {test_stat['failure_rate'] * 100}%")