test: add single-node binary tests (#15016)

Author: kwannoel

ci/scripts/run-e2e-test.sh

@@ -26,13 +26,24 @@ if [[ $mode == "standalone" ]]; then
   source ci/scripts/standalone-utils.sh
 fi
 
+if [[ $mode == "single-node" ]]; then
+  source ci/scripts/single-node-utils.sh
+fi
+
 cluster_start() {
   if [[ $mode == "standalone" ]]; then
     mkdir -p "$PREFIX_LOG"
     cargo make clean-data
     cargo make pre-start-dev
     start_standalone "$PREFIX_LOG"/standalone.log &
     cargo make dev standalone-minio-etcd
+  elif [[ $mode == "single-node" ]]; then
+    mkdir -p "$PREFIX_LOG"
+    cargo make clean-data
+    cargo make pre-start-dev
+    start_single_node "$PREFIX_LOG"/single-node.log &
+    # Give it a while to make sure the single-node is ready.
+    sleep 3
   else
     cargo make ci-start "$mode"
   fi
@@ -44,6 +55,9 @@ cluster_stop() {
     stop_standalone
     # Don't check standalone logs, they will exceed the limit.
     cargo make kill
+  elif [[ $mode == "single-node" ]]
+  then
+    stop_single_node
   else
     cargo make ci-kill
   fi
@@ -74,7 +88,9 @@ echo "--- e2e, $mode, batch"
 RUST_LOG="info,risingwave_stream=info,risingwave_batch=info,risingwave_storage=info" \
 cluster_start
 sqllogictest -p 4566 -d dev './e2e_test/ddl/**/*.slt' --junit "batch-ddl-${profile}"
-sqllogictest -p 4566 -d dev './e2e_test/background_ddl/basic.slt' --junit "batch-ddl-${profile}"
+if [[ "$mode" != 'single-node' ]]; then
+  sqllogictest -p 4566 -d dev './e2e_test/background_ddl/basic.slt' --junit "batch-ddl-${profile}"
+fi
 sqllogictest -p 4566 -d dev './e2e_test/visibility_mode/*.slt' --junit "batch-${profile}"
 sqllogictest -p 4566 -d dev './e2e_test/database/prepare.slt'
 sqllogictest -p 4566 -d test './e2e_test/database/test.slt'

ci/scripts/single-node-utils.sh

@@ -0,0 +1,50 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+export RW_PREFIX=$PWD/.risingwave
+export PREFIX_BIN=./target/debug
+export PREFIX_LOG=$RW_PREFIX/log
+
+# You can fill up this section by consulting
+# .risingwave/log/risedev.log, after calling ./risedev d full.
+# It is expected that minio, etcd will be started after this is called.
+start_single_node() {
+  mkdir -p "$HOME/.risingwave/state_store"
+  mkdir -p "$HOME/.risingwave/meta_store"
+  RUST_BACKTRACE=1 "$PREFIX_BIN"/risingwave >"$1" 2>&1
+}
+
+stop_single_node() {
+  pkill risingwave
+  rm -rf "$HOME/.risingwave/state_store"
+  rm -rf "$HOME/.risingwave/meta_store"
+}
+
+wait_single_node() {
+  set +e
+  timeout 20s bash -c '
+    while true; do
+      echo "Polling every 1s for single_node to be ready for 20s"
+      if psql -h localhost -p 4566 -d dev -U root -c "SELECT 1;" </dev/null
+      then exit 0;
+      else sleep 1;
+      fi
+    done
+  '
+  STATUS=$?
+  set -e
+  if [[ $STATUS -ne 0 ]]; then
+    echo "Single node failed to start with status: $STATUS"
+    exit 1
+  else
+    echo "Single node is ready"
+  fi
+}
+
+restart_single_node() {
+  stop_single_node
+  sleep 5
+  start_single_node "$PREFIX_LOG"/single-node-restarted.log &
+  wait_single_node
+}

ci/workflows/main-cron.yml

@@ -777,6 +777,26 @@ steps:
     timeout_in_minutes: 25
     retry: *auto-retry
 
+  - label: "e2e single-node binary test"
+    key: "e2e-single-node-binary-tests"
+    command: "ci/scripts/e2e-test.sh -p ci-release -m single-node"
+    if: |
+      !(build.pull_request.labels includes "ci/main-cron/skip-ci") && build.env("CI_STEPS") == null
+      || build.pull_request.labels includes "ci/run-e2e-single-node-tests"
+      || build.env("CI_STEPS") =~ /(^|,)e2e-single-node-tests?(,|$$)/
+    depends_on:
+      - "build"
+      - "build-other"
+      - "docslt"
+    plugins:
+      - docker-compose#v4.9.0:
+          run: rw-build-env
+          config: ci/docker-compose.yml
+          mount-buildkite-agent: true
+      - ./ci/plugins/upload-failure-logs
+    timeout_in_minutes: 25
+    retry: *auto-retry
+
   - label: "end-to-end test for opendal (parallel)"
     key: "e2e-test-opendal-parallel"
     command: "ci/scripts/e2e-test-parallel-for-opendal.sh -p ci-release"

ci/workflows/pull-request.yml

@@ -698,6 +698,22 @@ steps:
     timeout_in_minutes: 30
     retry: *auto-retry
 
+  - label: "e2e single-node binary test"
+    command: "ci/scripts/e2e-test.sh -p ci-dev -m single-node"
+    if: build.pull_request.labels includes "ci/run-e2e-single-node-tests" || build.env("CI_STEPS") =~ /(^|,)e2e-single-node-tests?(,|$$)/
+    depends_on:
+      - "build"
+      - "build-other"
+      - "docslt"
+    plugins:
+      - docker-compose#v4.9.0:
+          run: rw-build-env
+          config: ci/docker-compose.yml
+          mount-buildkite-agent: true
+      - ./ci/plugins/upload-failure-logs
+    timeout_in_minutes: 30
+    retry: *auto-retry
+
   # FIXME(kwannoel): Let the github PR labeller label it, if sqlsmith source files has changes.
   - label: "fuzz test"
     command: "ci/scripts/pr-fuzz-test.sh -p ci-dev"