fix(ci): wait for the standalone/single-node process to be killed bef…

…ore starting again (#18010)

Signed-off-by: Bugen Zhao <[email protected]>

ci/Dockerfile

@@ -11,7 +11,7 @@ ENV LANG en_US.utf8
 # Use AWS ubuntu mirror
 RUN sed -i 's|http://archive.ubuntu.com/ubuntu|http://us-east-2.ec2.archive.ubuntu.com/ubuntu/|g' /etc/apt/sources.list
 RUN apt-get update -yy && \
-    DEBIAN_FRONTEND=noninteractive apt-get -y install sudo make build-essential cmake protobuf-compiler curl parallel python3 python3-pip python3-venv software-properties-common \
+    DEBIAN_FRONTEND=noninteractive apt-get -y install sudo make build-essential cmake protobuf-compiler curl parallel python3 python3-pip python3-venv software-properties-common psmisc \
     openssl libssl-dev libsasl2-dev libcurl4-openssl-dev pkg-config bash openjdk-17-jdk wget unzip git tmux lld postgresql-client kcat netcat-openbsd mysql-client \
     maven zstd libzstd-dev locales \
     python3.12 python3.12-dev \

ci/build-ci-image.sh

@@ -10,7 +10,7 @@ cat ../rust-toolchain
 # shellcheck disable=SC2155
 
 # REMEMBER TO ALSO UPDATE ci/docker-compose.yml
-export BUILD_ENV_VERSION=v20240731
+export BUILD_ENV_VERSION=v20240812
 
 export BUILD_TAG="public.ecr.aws/w1p7b4n3/rw-build-env:${BUILD_ENV_VERSION}"
 

ci/docker-compose.yml

@@ -71,7 +71,7 @@ services:
       retries: 5
 
   source-test-env:
-    image: public.ecr.aws/w1p7b4n3/rw-build-env:v20240731
+    image: public.ecr.aws/w1p7b4n3/rw-build-env:v20240812
     depends_on:
       - mysql
       - sqlserver-server
@@ -85,7 +85,7 @@ services:
       - ..:/risingwave
 
   sink-test-env:
-    image: public.ecr.aws/w1p7b4n3/rw-build-env:v20240731
+    image: public.ecr.aws/w1p7b4n3/rw-build-env:v20240812
     depends_on:
       - mysql
       - db
@@ -108,12 +108,12 @@ services:
 
 
   rw-build-env:
-    image: public.ecr.aws/w1p7b4n3/rw-build-env:v20240731
+    image: public.ecr.aws/w1p7b4n3/rw-build-env:v20240812
     volumes:
       - ..:/risingwave
 
   ci-flamegraph-env:
-    image: public.ecr.aws/w1p7b4n3/rw-build-env:v20240731
+    image: public.ecr.aws/w1p7b4n3/rw-build-env:v20240812
     # NOTE(kwannoel): This is used in order to permit
     # syscalls for `nperf` (perf_event_open),
     # so it can do CPU profiling.
@@ -124,7 +124,7 @@ services:
       - ..:/risingwave
 
   regress-test-env:
-    image: public.ecr.aws/w1p7b4n3/rw-build-env:v20240731
+    image: public.ecr.aws/w1p7b4n3/rw-build-env:v20240812
     depends_on:
       db:
         condition: service_healthy

ci/scripts/single-node-utils.sh

@@ -19,7 +19,7 @@ start_single_node() {
 }
 
 stop_single_node() {
-  pkill risingwave
+  killall --wait risingwave
   rm -rf "$HOME/.risingwave/state_store"
   rm -rf "$HOME/.risingwave/meta_store"
 }
@@ -47,7 +47,6 @@ wait_single_node() {
 
 restart_single_node() {
   stop_single_node
-  sleep 5
   start_single_node "$PREFIX_LOG"/single-node-restarted.log &
   wait_single_node
 }

ci/scripts/standalone-utils.sh

@@ -75,7 +75,7 @@ start_standalone() {
 }
 
 stop_standalone() {
-  pkill standalone
+  killall --wait standalone
 }
 
 wait_standalone() {
@@ -101,7 +101,6 @@ wait_standalone() {
 
 restart_standalone() {
   stop_standalone
-  sleep 5
   start_standalone "$PREFIX_LOG"/standalone-restarted.log &
   wait_standalone
 }