feat[utest]: repeat the automatic run and sample free/ps between loops

This commit is contained in:
sexenteen authored and Rbb666 committed 2026-11-09 09:31:47 +08:00
1 parent b3a0eb5e12
commit cc533832d3
3 files changed
+123 -4

No files matched your search

+47 -4
View File
@@ -51,6 +51,11 @@ on:
required: false
default: 'manual'
type: string
utest_loop:
description: '自动运行重复次数 (1-18),大于 1 时每轮结束后执行 free/ps 观测内存与线程变化'
required: false
default: '1'
type: string
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
@@ -274,7 +279,31 @@ jobs:
- name: Build BSP
if: ${{ env.TOOLCHAIN_INSTALLED != '' && success() }}
env:
# Passed through env rather than interpolated into the script, so a value
# like "$(...)" or a backtick can never be executed by the shell.
UTEST_LOOP_INPUT: ${{ github.event.inputs.utest_loop }}
run: |
# Validate the loop count before anything is appended to .config. The value
# must be a plain decimal integer in 1..18: 18 loops x 20 minutes is the
# 360 minute step budget of the monitor step, so a larger value could only
# be cancelled by GitHub without a useful message.
UTEST_LOOP="${UTEST_LOOP_INPUT:-1}"
case "$UTEST_LOOP" in
''|*[!0-9]*)
echo "ERROR: utest_loop must be a decimal integer in 1..18, got '$UTEST_LOOP'"
exit 1
;;
esac
# 10# forces a decimal parse, so 08/09 are not treated as octal
UTEST_LOOP=$((10#$UTEST_LOOP))
if [ "$UTEST_LOOP" -lt 1 ] || [ "$UTEST_LOOP" -gt 18 ]; then
echo "ERROR: utest_loop must be in 1..18, got '$UTEST_LOOP'"
exit 1
fi
echo "UTEST_LOOP=$UTEST_LOOP" >> $GITHUB_ENV
echo "utest auto run loop: $UTEST_LOOP"
# Keep regular RISC-V jobs UP; SMP coverage has its own matrix entry.
if [[ "$TEST_BSP_ROOT" == "bsp/qemu-virt64-riscv" && "${{ matrix.platform.SMP_RUN }}" != "smp" ]]; then
echo "# CONFIG_RT_USING_SMP is not set" >> $TEST_BSP_ROOT/.config
@@ -300,6 +329,12 @@ jobs:
cat $UTEST_CONFIG_ROOT/$TEST_CONFIG_FILE >> $TEST_BSP_ROOT/.config
echo >> $TEST_BSP_ROOT/.config
# Repeat the automatic utest run when a loop count is requested, so that
# resources leaked by a test case show up as growing memory/threads.
if [ "$UTEST_LOOP" -ne 1 ]; then
echo "CONFIG_RT_UTEST_AUTO_RUN_LOOP=$UTEST_LOOP" >> $TEST_BSP_ROOT/.config
fi
# Keep standard qemu tests free of PM/tickless side effects.
if [[ "${{ matrix.platform.KERNEL }}" == "standard" && ( "$TEST_BSP_ROOT" == "bsp/qemu-virt64-riscv" || "$TEST_BSP_ROOT" == "bsp/qemu-virt64-aarch64" ) ]]; then
echo "# CONFIG_RT_USING_PM is not set" >> $TEST_BSP_ROOT/.config
@@ -419,7 +454,7 @@ jobs:
if: ${{ env.TOOLCHAIN_INSTALLED != '' && success() }}
# Same known-flaky qemu-virt64-riscv standard kernel utest run as the previous step.
continue-on-error: ${{ matrix.platform.UTEST == 'RISCV' && (matrix.config_file == 'kernel/ipc.cfg' || matrix.config_file == 'kernel/kernel_basic.cfg') }}
timeout-minutes: 20
timeout-minutes: 360
run: |
LOG_FILE="qemu_output_${TEST_QEMU_ARCH}.log"
FAILURE_DETECTED=false
@@ -437,7 +472,15 @@ jobs:
echo " VV VV "
echo "=========================================================================================="
set +e
timeout 20m bash -c '
# Bound the QEMU wait by the loop count (20 minutes per loop). The step level
# timeout-minutes cannot take an expression, so the real limit is applied here.
LOOP_NUM="${UTEST_LOOP:-1}"
case "$LOOP_NUM" in
''|*[!0-9]*) LOOP_NUM=1 ;;
esac
MONITOR_TIMEOUT_MIN=$(( LOOP_NUM * 20 ))
echo "Monitoring the QEMU log for at most ${MONITOR_TIMEOUT_MIN} minutes (utest_loop=${LOOP_NUM})"
timeout "${MONITOR_TIMEOUT_MIN}m" bash -c '
tail -n +1 -f "$1" | while IFS= read -r line; do
echo "$line"
if [[ "$line" == *"[ FAILED ] [ result ]"* ]]; then
@@ -454,7 +497,7 @@ jobs:
exit 1
fi
if [[ "$line" == *"[==========] [ utest ] finished"* ]]; then
if [[ "$line" == *"[==========] [ utest ] all loops finished"* ]]; then
if [[ "$FAILURE_DETECTED" == "true" ]]; then
echo "=========================================================================================="
echo " || || "
@@ -477,7 +520,7 @@ jobs:
if [[ $monitor_status -eq 124 ]]; then
echo "=========================================================================================="
echo " || || "
echo " || Error: Utest run timed out after 20 minutes. Exiting... || "
echo " || Error: Utest run timed out after ${MONITOR_TIMEOUT_MIN} minutes. Exiting... || "
echo " VV VV "
echo "=========================================================================================="
pkill -f "qemu-system-${TEST_QEMU_ARCH}.*${TEST_BSP_ROOT}/rtthread.bin" || true
+14
View File
@@ -226,6 +226,20 @@ config RT_USING_UTEST
help
If enable this option, the test cases will be run automatically when board boot up.
config RT_UTEST_AUTO_RUN_LOOP
int "Loop count of the automatic run"
depends on RT_UTEST_USING_AUTO_RUN
range 1 18
default 1
help
How many times the automatic run repeats the registered test cases.
When it is greater than 1, "free" and "ps" are executed after each
loop, so memory or thread leaks show up in the log as growing numbers.
The upper bound matches the CI budget: the utest_auto_run workflow
budgets 20 minutes per loop with a 360 minute step timeout, so 18
loops are the most that can finish before GitHub cancels the step.
if FINSH_USING_OPTION_COMPLETION
config RT_UTEST_MAX_OPTIONS
int "Maximum number of utest cases for auto-completion"
+62
View File
@@ -14,6 +14,10 @@
#include "utest.h"
#include "utest_log.h"
#ifdef RT_USING_FINSH
#include <msh.h>
#endif
#define DBG_TAG "utest"
#ifdef UTEST_DEBUG
#define DBG_LVL DBG_LOG
@@ -222,6 +226,44 @@ static int utest_help(void)
return 0;
}
#ifdef RT_UTEST_AUTO_RUN_LOOP
static void utest_loop_runtime_info(rt_uint32_t loop)
{
if (tc_loop <= 1)
{
return;
}
LOG_I("[==========] [ utest ] loop %u/%u runtime info", (rt_uint32_t)loop, (rt_uint32_t)tc_loop);
#ifdef RT_USING_FINSH
{
static char free_cmd[] = "free";
static char ps_cmd[] = "ps";
int ret = 0;
ret |= msh_exec(free_cmd, sizeof(free_cmd) - 1);
ret |= msh_exec(ps_cmd, sizeof(ps_cmd) - 1);
if (ret != 0)
{
/* do not fail the test run just because the commands are absent, but make it visible */
LOG_W("[==========] [ utest ] 'free'/'ps' are not available, runtime info is incomplete");
}
}
#else
{
rt_size_t total = 0;
rt_size_t used = 0;
rt_size_t max_used = 0;
rt_memory_info(&total, &used, &max_used);
LOG_I("[==========] [ utest ] memory total:%u used:%u maximum:%u",
(rt_uint32_t)total, (rt_uint32_t)used, (rt_uint32_t)max_used);
}
#endif /* RT_USING_FINSH */
}
#endif /* RT_UTEST_AUTO_RUN_LOOP */
static void utest_do_run(const char *utest_name)
{
rt_size_t i;
@@ -352,7 +394,18 @@ static void utest_do_run(const char *utest_name)
}
LOG_I("[==========] [ utest ] finished");
#ifdef RT_UTEST_AUTO_RUN_LOOP
utest_loop_runtime_info(index + 1);
#endif
}
#ifdef RT_UTEST_AUTO_RUN_LOOP
/* Emitted once, after every loop and its runtime info. CI uses this line
* instead of the per-loop "finished" one, otherwise it would stop after the
* first loop and miss the remaining loops and their free/ps output. */
LOG_I("[==========] [ utest ] all loops finished");
#endif
}
static void utest_thr_entry(void *para)
@@ -378,7 +431,16 @@ static void utest_thread_create(const char *utest_name)
#ifdef RT_UTEST_USING_AUTO_RUN
static int utest_auto_run(void)
{
#ifdef RT_UTEST_AUTO_RUN_LOOP
tc_loop = RT_UTEST_AUTO_RUN_LOOP;
#else
tc_loop = 1;
#endif
if (tc_loop == 0)
{
tc_loop = 1;
}
utest_thread_create(RT_NULL);
return RT_EOK;
}