partition slow udf tests

risingwavelabs · May 27, 2024 · 64acfec · 64acfec
1 parent a0d7323
commit 64acfec
Show file tree

Hide file tree

Showing 4 changed files with 54 additions and 5 deletions.
diff --git a/ci/scripts/run-e2e-test.sh b/ci/scripts/run-e2e-test.sh
@@ -80,7 +80,6 @@ RUST_LOG="info,risingwave_stream=info,risingwave_batch=info,risingwave_storage=i
 cluster_start
 # Please make sure the regression is expected before increasing the timeout.
 sqllogictest -p 4566 -d dev './e2e_test/streaming/**/*.slt' --junit "streaming-${profile}"
-sqllogictest -p 4566 -d dev './e2e_test/backfill/rate_limit/*.slt'
 sqllogictest -p 4566 -d dev './e2e_test/backfill/sink/different_pk_and_dist_key.slt'
 
 echo "--- Kill cluster"
@@ -115,7 +114,6 @@ pkill python3
 
 sqllogictest -p 4566 -d dev './e2e_test/udf/alter_function.slt'
 sqllogictest -p 4566 -d dev './e2e_test/udf/graceful_shutdown_python.slt'
-sqllogictest -p 4566 -d dev './e2e_test/udf/always_retry_python.slt'
 # FIXME: flaky test
 # sqllogictest -p 4566 -d dev './e2e_test/udf/retry_python.slt'
 

diff --git a/ci/scripts/slow-e2e-test.sh b/ci/scripts/slow-e2e-test.sh
@@ -0,0 +1,39 @@
+#!/usr/bin/env bash
+
+# Exits as soon as any line fails.
+set -euo pipefail
+
+while getopts 'p:m:' opt; do
+    case ${opt} in
+        p )
+            profile=$OPTARG
+            ;;
+        m )
+            mode=$OPTARG
+            ;;
+        \? )
+            echo "Invalid Option: -$OPTARG" 1>&2
+            exit 1
+            ;;
+        : )
+            echo "Invalid option: $OPTARG requires an argument" 1>&2
+            ;;
+    esac
+done
+shift $((OPTIND -1))
+
+source ci/scripts/common.sh
+
+download_and_prepare_rw "$profile" common
+
+
+echo "--- Download artifacts"
+mkdir -p e2e_test/udf/wasm/target/wasm32-wasi/release/
+buildkite-agent artifact download udf.wasm e2e_test/udf/wasm/target/wasm32-wasi/release/
+buildkite-agent artifact download udf.jar ./
+
+echo "--- e2e, $mode, slow-udf-tests"
+RUST_LOG="info" \
+risedev ci-start "$mode"
+sqllogictest -p 4566 -d dev './e2e_test/udf/always_retry_python.slt'
+sqllogictest -p 4566 -d dev './e2e_test/backfill/rate_limit/slow-udf.slt'
diff --git a/ci/workflows/main-cron.yml b/ci/workflows/main-cron.yml
@@ -90,6 +90,18 @@ steps:
     timeout_in_minutes: 15
     retry: *auto-retry
 
+  - label: "slow end-to-end test (release)"
+    key: "slow-e2e-test-release"
+    command: "ci/scripts/slow-e2e-test.sh -p ci-release -m ci-3streaming-2serving-3fe"
+    if: |
+        !(build.pull_request.labels includes "ci/main-cron/skip-ci") && build.env("CI_STEPS") == null
+        || build.pull_request.labels includes "ci/run-slow-e2e-test"
+        || build.env("CI_STEPS") =~ /(^|,)slow-e2e-tests?(,|$$)/
+    depends_on:
+      - "build"
+      - "build-other"
+      - "docslt"
+
   - label: "meta backup test (release)"
     key: "e2e-meta-backup-test-release"
     command: "ci/scripts/run-meta-backup-test.sh -p ci-release -m ci-3streaming-2serving-3fe"

diff --git a/e2e_test/udf/always_retry_python.slt b/e2e_test/udf/always_retry_python.slt
@@ -2,7 +2,7 @@ system ok
 python3 e2e_test/udf/test.py &
 
 # wait for server to start
-sleep 5s
+sleep 10s
 
 statement ok
 CREATE FUNCTION sleep_always_retry(INT) RETURNS INT AS 'sleep' USING LINK 'http://localhost:8815' WITH ( always_retry_on_network_error = true );
@@ -33,9 +33,9 @@ CREATE MATERIALIZED VIEW mv_no_retry AS SELECT sleep_no_retry(v1) as s1 from t;
 statement ok
 CREATE MATERIALIZED VIEW mv_always_retry AS SELECT sleep_always_retry(v1) as s1 from t;
 
-# Immediately kill the server, sleep for 10s.
+# Immediately kill the server, sleep
 system ok
-pkill -9 -i python && sleep 10
+pkill -9 -i python && sleep 50
 
 # Restart the server
 system ok