Re: [Qemu-devel] [PATCH v3 3/3] tests/migration: Add a test for auto converge

[Yury Kotov]

22.07.2019, 20:35, "Dr. David Alan Gilbert" <address@hidden>:
> * Yury Kotov (address@hidden) wrote:
>>  Signed-off-by: Yury Kotov <address@hidden>
>
> This looks OK to me, but have you tried it on a really really overloaded
> host?
> I worry that you might skip some of the percentage steps or not hit the
> bandwidth on the small overloaded VMs we get in CI.

No, I haven't tried. I saw patchew's logs and you're absolutely right - such
problem exists. Now I'm looking for a way to fix it.

>
> Dave
>
>>  ---
>>   tests/migration-test.c | 119 +++++++++++++++++++++++++++++++++++++----
>>   1 file changed, 108 insertions(+), 11 deletions(-)
>>
>>  diff --git a/tests/migration-test.c b/tests/migration-test.c
>>  index a4feb9545d..bb69517fc8 100644
>>  --- a/tests/migration-test.c
>>  +++ b/tests/migration-test.c
>>  @@ -241,6 +241,17 @@ static int64_t read_ram_property_int(QTestState *who, 
>> const char *property)
>>       return result;
>>   }
>>
>>  +static int64_t read_migrate_property_int(QTestState *who, const char 
>> *property)
>>  +{
>>  + QDict *rsp_return;
>>  + int64_t result;
>>  +
>>  + rsp_return = migrate_query(who);
>>  + result = qdict_get_try_int(rsp_return, property, 0);
>>  + qobject_unref(rsp_return);
>>  + return result;
>>  +}
>>  +
>>   static uint64_t get_migration_pass(QTestState *who)
>>   {
>>       return read_ram_property_int(who, "dirty-sync-count");
>>  @@ -255,20 +266,22 @@ static void read_blocktime(QTestState *who)
>>       qobject_unref(rsp_return);
>>   }
>>
>>  +static bool check_migration_status(QTestState *who, const char *status)
>>  +{
>>  + bool completed;
>>  + char *current_status;
>>  +
>>  + current_status = migrate_query_status(who);
>>  + completed = strcmp(current_status, status) == 0;
>>  + g_assert_cmpstr(current_status, !=, "failed");
>>  + g_free(current_status);
>>  + return completed;
>>  +}
>>  +
>>   static void wait_for_migration_status(QTestState *who,
>>                                         const char *goal)
>>   {
>>  - while (true) {
>>  - bool completed;
>>  - char *status;
>>  -
>>  - status = migrate_query_status(who);
>>  - completed = strcmp(status, goal) == 0;
>>  - g_assert_cmpstr(status, !=, "failed");
>>  - g_free(status);
>>  - if (completed) {
>>  - return;
>>  - }
>>  + while (!check_migration_status(who, goal)) {
>>           usleep(1000);
>>       }
>>   }
>>  @@ -1121,6 +1134,89 @@ static void test_migrate_fd_proto(void)
>>       test_migrate_end(from, to, true);
>>   }
>>
>>  +static void test_migrate_auto_converge(void)
>>  +{
>>  + char *uri = g_strdup_printf("unix:%s/migsocket", tmpfs);
>>  + QTestState *from, *to;
>>  + int i;
>>  + int64_t remaining, downtime;
>>  +
>>  + /*
>>  + * We want the test to be fast enough, but stable.
>>  + * Throttle percentages are chosen to cover all cases (init, increment, 
>> max)
>>  + */
>>  + static const int64_t expected_pcts[] = { 0, 1, 51, 98 };
>>  + const int64_t max_bandwidth = 200000000; /* ~200Mb/s */
>>  + const int64_t downtime_limit = 50; /* 50ms */
>>  + /*
>>  + * We migrate through unix-socket (> 500Mb/s).
>>  + * Thus, expected migration speed ~= bandwidth limit (< 500Mb/s).
>>  + * So, we can predict expected_threshold
>>  + */
>>  + const int64_t expected_threshold = max_bandwidth * downtime_limit / 1000;
>>  +
>>  + if (test_migrate_start(&from, &to, uri, false, false)) {
>>  + return;
>>  + }
>>  +
>>  + migrate_set_capability(from, "auto-converge", true);
>>  + migrate_set_parameter_int(from, "cpu-throttle-initial", expected_pcts[1]);
>>  + migrate_set_parameter_int(from, "cpu-throttle-increment",
>>  + expected_pcts[2] - expected_pcts[1]);
>>  + migrate_set_parameter_int(from, "max-cpu-throttle", expected_pcts[3]);
>>  +
>>  + migrate_set_parameter_int(from, "max-bandwidth", max_bandwidth);
>>  + migrate_set_parameter_int(from, "downtime-limit", downtime_limit);
>>  +
>>  + /* To check remaining size after precopy */
>>  + migrate_set_capability(from, "pause-before-switchover", true);
>>  +
>>  + /* Wait for the first serial output from the source */
>>  + wait_for_serial("src_serial");
>>  +
>>  + migrate(from, uri, "{}");
>>  +
>>  + for (i = 0; i < ARRAY_SIZE(expected_pcts); i++) {
>>  + int64_t pct;
>>  + pct = read_migrate_property_int(from, "cpu-throttle-percentage");
>>  + g_assert_cmpint(pct, ==, expected_pcts[i]);
>>  + while (pct == expected_pcts[i] && !got_stop) {
>>  + usleep(1000);
>>  + pct = read_migrate_property_int(from, "cpu-throttle-percentage");
>>  + }
>>  + /* We break out of this loop only in paused state */
>>  + if (got_stop || i + 1 == ARRAY_SIZE(expected_pcts)) {
>>  + /* Check unexpected throttle percentage change */
>>  + g_assert_true(got_stop);
>>  + /* Check unexpected converge */
>>  + g_assert_cmpint(i + 1, ==, ARRAY_SIZE(expected_pcts));
>>  + g_assert_true(check_migration_status(from, "pre-switchover"));
>>  + }
>>  + }
>>  +
>>  + remaining = read_ram_property_int(from, "remaining");
>>  + g_assert_cmpint(remaining, <, expected_threshold);
>>  +
>>  + wait_command(from, "{ 'execute': 'migrate-continue' , 'arguments':"
>>  + " { 'state': 'pre-switchover' } }");
>>  +
>>  + qtest_qmp_eventwait(to, "RESUME");
>>  +
>>  + wait_for_serial("dest_serial");
>>  + wait_for_migration_complete(from);
>>  +
>>  + downtime = read_migrate_property_int(from, "downtime");
>>  + /*
>>  + * Actual downtime may be greater than downtime limit,
>>  + * but the difference should be small enough (~20ms)
>>  + */
>>  + g_assert_cmpint(downtime, <, downtime_limit + 20);
>>  +
>>  + g_free(uri);
>>  +
>>  + test_migrate_end(from, to, true);
>>  +}
>>  +
>>   int main(int argc, char **argv)
>>   {
>>       char template[] = "/tmp/migration-test-XXXXXX";
>>  @@ -1176,6 +1272,7 @@ int main(int argc, char **argv)
>>       /* qtest_add_func("/migration/ignore_shared", test_ignore_shared); */
>>       qtest_add_func("/migration/xbzrle/unix", test_xbzrle_unix);
>>       qtest_add_func("/migration/fd_proto", test_migrate_fd_proto);
>>  + qtest_add_func("/migration/auto_converge", test_migrate_auto_converge);
>>
>>       ret = g_test_run();
>>
>>  --
>>  2.22.0
> --
> Dr. David Alan Gilbert / address@hidden / Manchester, UK

Regards,
Yury