fix(CI): resolve flaky tests (#3560)

Leiyks · web-flow · commit 88a859c68edd · 2026-01-14T14:05:43.000+01:00
* fix(test_integrations_kafka): properly wait for zookeeper to be ready before starting kafka

Signed-off-by: Alexandre Rulleau &lt;alexandre.rulleau@datadoghq.com&gt;

* fix(verify_centos): increase timeout to avoid failure

Signed-off-by: Alexandre Rulleau &lt;alexandre.rulleau@datadoghq.com&gt;

* fix(profiling_test): skip jit related test on 7.X

Signed-off-by: Alexandre Rulleau &lt;alexandre.rulleau@datadoghq.com&gt;

* fix(test_extension_ci): increase timeout for valgrind runs

Signed-off-by: Alexandre Rulleau &lt;alexandre.rulleau@datadoghq.com&gt;

---------

Signed-off-by: Alexandre Rulleau &lt;alexandre.rulleau@datadoghq.com&gt;
diff --git a/.gitlab/generate-common.php b/.gitlab/generate-common.php
@@ -143,10 +143,27 @@ function dockerhub_login() {
       ZOOKEEPER_TICK_TIME: 2000
       ALLOW_ANONYMOUS_LOGIN: "yes"
       ZOOKEEPER_ADMIN_ENABLE_SERVER: "false"
+      KAFKA_OPTS: "-Dzookeeper.4lw.commands.whitelist=srvr,ruok"
 
   kafka:
     name: registry.ddbuild.io/images/mirror/confluentinc/cp-kafka:7.8.0
     alias: kafka-integration
+    entrypoint: ["/bin/bash"]
+    command:
+      - -c
+      - |
+        # Wait for Zookeeper to be ready before starting Kafka
+        echo "Waiting for Zookeeper to be ready..."
+        for i in $(seq 1 30); do
+          if echo "ruok" | nc zookeeper 2181 2>/dev/null | grep -q "imok"; then
+            echo "Zookeeper is ready, starting Kafka..."
+            break
+          fi
+          echo "Waiting for Zookeeper... attempt $i/30"
+          sleep 2
+        done
+        # Start Kafka with original entrypoint
+        exec /etc/confluent/docker/run
     variables:
       KAFKA_BROKER_ID: 111
       KAFKA_CREATE_TOPICS: test-lowlevel:1:1,test-highlevel:1:1
diff --git a/.gitlab/generate-package.php b/.gitlab/generate-package.php
@@ -1168,7 +1168,6 @@
     - job: "prepare code"
       artifacts: true
   before_script:
-<?php dockerhub_login() ?>
     - |
       # Setup cache dirs
       mkdir -p $PIP_CACHE_DIR
@@ -1187,6 +1186,7 @@
 
       # Install Python dependencies
       pip install -U pip virtualenv
+<?php dockerhub_login() ?>
     - git clone https://github.com/DataDog/system-tests.git
     - mv packages/{datadog-setup.php,dd-library-php-*x86_64-linux-gnu.tar.gz} system-tests/binaries
     - cd system-tests
diff --git a/.gitlab/wait-for-service-ready.sh b/.gitlab/wait-for-service-ready.sh
@@ -8,6 +8,7 @@ detect_service_type() {
     test-agent) echo "test-agent" ;;
     mysql-integration) echo "mysql" ;;
     elasticsearch*) echo "elasticsearch" ;;
+    zookeeper*) echo "zookeeper" ;;
     kafka*) echo "kafka" ;;
     redis*) echo "redis" ;;
     httpbin*) echo "httpbin" ;;
@@ -51,7 +52,7 @@ wait_for_single_service() {
       kafka)
         # Kafka readiness via nc check + settle time
         if timeout 5 nc -z "${HOST}" "${PORT}" 2>/dev/null; then
-          sleep 2  # Additional settle time for Kafka
+          sleep 5  # Additional settle time for Kafka
           echo "Kafka is ready"
           return 0
         fi
@@ -69,6 +70,13 @@ wait_for_single_service() {
           return 0
         fi
         ;;
+      zookeeper)
+        # Zookeeper readiness via "ruok" four-letter-word command
+        if echo "ruok" | nc -w1 -q1 "${HOST}" "${PORT}" 2>/dev/null | grep -q "imok"; then
+          echo "Zookeeper is ready"
+          return 0
+        fi
+        ;;
       generic|*)
         # For generic services, just verify port + HTTP 200/health endpoint
         if curl -sf "http://${HOST}:${PORT}/" > /dev/null 2>&1 || curl -sf "http://${HOST}:${PORT}/health" > /dev/null 2>&1; then
diff --git a/dockerfiles/verify_packages/centos/install.sh b/dockerfiles/verify_packages/centos/install.sh
@@ -6,26 +6,30 @@ OS_VERSION=$(source /etc/os-release; echo $VERSION_ID)
 
 function do_retry() {
   RETRIES=3
+  ATTEMPT=1
   while
     ! "$@"
   do
     if ! ((--RETRIES)); then
       return 1
     fi
+    echo "Retry attempt $ATTEMPT failed, waiting 5 seconds before retry $((ATTEMPT + 1))..."
+    sleep 5
+    ATTEMPT=$((ATTEMPT + 1))
   done
 }
 
 # Enable epel repo
-do_retry yum install -y epel-release
+do_retry yum install -y --setopt=timeout=300 epel-release
 
 # Installing pre-requisites
-do_retry yum install -y wget nginx httpd
+do_retry yum install -y --setopt=timeout=300 wget nginx httpd
 # Nginx listens on 8080, apache on 8081
 sed -i "s/Listen 80/Listen 127.0.0.1:8081/" /etc/httpd/conf/httpd.conf
 
 # Installing php
 do_retry rpm -Uvh http://rpms.famillecollet.com/enterprise/remi-release-${OS_VERSION}.rpm
-do_retry yum --enablerepo=remi-php${PHP_MINOR_MAJOR} install -y \
+do_retry yum --enablerepo=remi-php${PHP_MINOR_MAJOR} install -y --setopt=timeout=300 \
     php-cli \
     php-fpm \
     php-opcache \
diff --git a/profiling/tests/phpt/allocation_jit_01.phpt b/profiling/tests/phpt/allocation_jit_01.phpt
@@ -11,6 +11,8 @@ where we assume we are hooked into, while we are not.
 <?php
 if (!extension_loaded('datadog-profiling'))
     echo "skip: test requires datadog-profiling", PHP_EOL;
+if (PHP_VERSION_ID < 80000)
+    echo "skip: JIT requires PHP >= 8.0", PHP_EOL;
 ?>
 --ENV--
 USE_ZEND_ALLOC=0
diff --git a/tests/ext/background-sender/agent_sampling-standalone-asm_01.phpt b/tests/ext/background-sender/agent_sampling-standalone-asm_01.phpt
@@ -17,10 +17,12 @@ datadog.trace.agent_test_session_token=background-sender/agent_samplingc
 include __DIR__ . '/../includes/request_replayer.inc';
 
 $rr = new RequestReplayer();
+$rr->maxIteration = 2000;
 
 $picked = 0;
 $notPicked = 0;
-for ($i = 0; $i < 5; $i++)
+$maxIterations = 10;
+for ($i = 0; $i < $maxIterations; $i++)
 {
     //Do call and get sampling
     \DDTrace\start_span();
diff --git a/tests/ext/background-sender/agent_sampling-standalone-asm_02.phpt b/tests/ext/background-sender/agent_sampling-standalone-asm_02.phpt
@@ -17,6 +17,7 @@ datadog.trace.agent_test_session_token=background-sender/agent_samplingb
 include __DIR__ . '/../includes/request_replayer.inc';
 
 $rr = new RequestReplayer();
+$rr->maxIteration = 2000;
 
 $get_sampling = function() use ($rr) {
     $root = json_decode($rr->waitForDataAndReplay()["body"], true);
diff --git a/tests/ext/background-sender/agent_sampling-standalone-asm_03.phpt b/tests/ext/background-sender/agent_sampling-standalone-asm_03.phpt
@@ -22,6 +22,7 @@ datadog.trace.agent_test_session_token=background-sender/agent_samplinga
 include __DIR__ . '/../includes/request_replayer.inc';
 
 $rr = new RequestReplayer();
+$rr->maxIteration = 2000;
 
 $get_sampling = function() use ($rr) {
     $root = json_decode($rr->waitForDataAndReplay()["body"], true);
