On Tue, 27 Nov 2018 at 08:29, Oleksandr Terentiev <oterenti@cisco.com> wrote:

Hi Abibal,

In our project we need to analyze a total number of passed and failed tests for each packet. To distinguish packets we use lava-test-set feature.
In order to implement that I modified ptest.py and send-to-lava.sh scripts. Could you please look at the patch and express your opinion?
Maybe this code can be added to git.linaro.org/qa/test-definitions.git ?

Hi Oleksandr,

The code looks good, can you have an example of the LAVA test run to see the actual results?

Regards,

Anibal

Best regards,
Alex

automated/linux/ptest: Analyze each test in package tests

Currently ptest.py analyze only exit code of each package test
to decide if it passed or not. However, ptest-runner can return
success code even though some tests failed. So we need to parse
test output and analyze it.

It also quite useful to see exactly which tests failed. So results are
recorded for each particular test, and lava-test-set feature is used
to distinguish packages.

Signed-off-by: Oleksandr Terentiev <oterenti@cisco.com>

diff --git a/automated/linux/ptest/ptest.py b/automated/linux/ptest/ptest.py
index 13feb4d..a28d7f0 100755
--- a/automated/linux/ptest/ptest.py
+++ b/automated/linux/ptest/ptest.py
@@ -84,20 +84,60 @@ def filter_ptests(ptests, requested_ptests, exclude):

     return filter_ptests

+def parse_line(line):
+    test_status_list = {
+        'pass': re.compile("^PASS:(.+)"),
+        'fail': re.compile("^FAIL:(.+)"),
+        'skip': re.compile("^SKIP:(.+)")
+    }
+
+    for test_status, status_regex in test_status_list.items():
+            test_name = status_regex.search(line)
+            if test_name:
+                return [test_name.group(1), test_status]

-def check_ptest(ptest_dir, ptest_name, output_log):
-    status = 'pass'
+    return None

-    try:
-        output = subprocess.check_call('ptest-runner -d %s %s' %
-                                       (ptest_dir, ptest_name), shell=True,
-                                       stderr=subprocess.STDOUT)
-    except subprocess.CalledProcessError:
-        status = 'fail'
+def parse_ptest(log_file):
+    result = []

-    with open(output_log, 'a+') as f:
-        f.write("%s %s\n" % (ptest_name, status))
+    with open(log_file, 'r') as f:
+        for line in f:
+            result_tuple = parse_line(line)
+            if not result_tuple:
+                continue
+            print(result_tuple)
+            result.append(result_tuple)
+            continue

+    return result
+
+def run_command(command, log_file):
+    process = subprocess.Popen(command,
+                               shell=True,
+                               stdout=subprocess.PIPE,
+                               stderr=subprocess.STDOUT)
+    with open(log_file, 'w') as f:
+        while True:
+            output = process.stdout.readline()
+            if output == '' and process.poll() is not None:
+                break
+            if output:
+                print output.strip()
+                f.write("%s\n" % output.strip())
+    rc = process.poll()
+    return rc
+
+def check_ptest(ptest_dir, ptest_name, output_log):
+    log_name = os.path.join(os.getcwd(), '%s.log' % ptest_name)
+    status = run_command('ptest-runner -d %s %s' % (ptest_dir, ptest_name), log_name)
+
+    with open(output_log, 'a+') as f:
+        f.write("lava-test-set start %s\n" % ptest_name)
+        f.write("%s %s\n" % (ptest_name, "pass" if status == 0 else "fail"))
+        for test, test_status in parse_ptest(log_name):
+            f.write("%s %s\n" % (re.sub(r'[^\w-]', '', test), test_status))
+        f.write("lava-test-set stop %s\n" % ptest_name)

def main():
     parser = argparse.ArgumentParser(description="LAVA/OE ptest script",
diff --git a/automated/utils/send-to-lava.sh b/automated/utils/send-to-lava.sh
index bf2a477..db4442c 100755
--- a/automated/utils/send-to-lava.sh
+++ b/automated/utils/send-to-lava.sh
@@ -4,6 +4,8 @@ RESULT_FILE="$1"

which lava-test-case > /dev/null 2>&1
lava_test_case="$?"
+which lava-test-set > /dev/null 2>&1
+lava_test_set="$?"

if [ -f "${RESULT_FILE}" ]; then
     while read -r line; do
@@ -31,6 +33,18 @@ if [ -f "${RESULT_FILE}" ]; then
             else
                echo "<TEST_CASE_ID=${test} RESULT=${result} MEASUREMENT=${measurement} UNITS=${units}>"
             fi
+        elif echo "${line}" | egrep -iq "^lava-test-set.*"; then
+            test_set_status="$(echo "${line}" | awk '{print $2}')"
+            test_set_name="$(echo "${line}" | awk '{print $3}')"
+            if [ "${lava_test_set}" -eq 0 ]; then
+                lava-test-set "${test_set_status}" "${test_set_name}"
+            else
+                if [ "${test_set_status}" == "start" ]; then
+                    echo "<LAVA_SIGNAL_TESTSET START ${test_set_name}>"
+                else
+                    echo "<LAVA_SIGNAL_TESTSET STOP>"
+                fi
+            fi
         fi
     done < "${RESULT_FILE}"
else

On 01.10.18 17:09, Anibal Limon wrote:

Hi,

I was on vacation, that's the reason for the slow response, comments below,

On Fri, 28 Sep 2018 at 03:59, Oleksandr Terentiev <oterenti@cisco.com> wrote:

Hi,

I would like to discuss the following question.
As it was said now we have to analyze pass/fail of every ptest. From my point of view there are couple options.

The first we can parse output and mark ptest as failed if there is even only one failed test found.

Right I will choice this approach changes needs to be done in the ptest lava script [1] to fail when any of the package tests failed like [2].

The second we can analyze each test within some packet and record corresponding results.
I see a few issues here. First of all there will be a large number of test results as each ptest can run lots of tests.

Right but that need to be handled in every OE ptest script, I mean if you want to fail if certain test inside a ptest fails needs to be done in OE.

Another thing is that we need somehow separate test results between particular packets.

Currently we use QA reports to see only the package test result, now If you want to be able look at details you need to see the ptest log.

As an option we can use lava-test-set feature for that. So each test within ptest will be marked as test case and packet name we will see as test set.

What do you think about that?

May be the lava-test-set is an option.

I would go to do the 1st option and then start to review/implement the idea of use lava-test-feature.

Regards,

Anibal

[1] https://git.linaro.org/qa/test-definitions.git/tree/automated/linux/ptest/ptest.py

[2] http://git.openembedded.org/openembedded-core/tree/meta/lib/oeqa/runtime/cases/ptest.py#n87

Regards,
Alex

On 23.08.18 16:10, Anibal Limon wrote:

On Thu, 23 Aug 2018 at 05:54, Oleksandr Terentiev <oterenti@cisco.com> wrote:

Thank you Anibal for the fast response

On 22.08.18 19:50, Anibal Limon wrote:

On Wed, 22 Aug 2018 at 11:39, Oleksandr Terentiev <oterenti@cisco.com> wrote:

Hi,

I launched util-linux ptest using automated/linux/ptest/ptest.yaml from
https://git.linaro.org/qa/test-definitions.git and received the
following results:
https://pastebin.com/nj9PYQzE

As you can see some tests failed. However, case util-linux marked as
passed. It looks like ptest.py only analyze return code of ptest-runner
-d <ptest_dir> <ptest_name> command. And since ptest-runner finishes
correctly exit code is 0. Therefore all tests are always marked as
passed, and users never know when some of the tests fail.

Maybe it worth to analyze each test?

Talking about each ptest the result comes from the ptest script in the OE recipe [1], for convention if the OE ptest returns 0 means pass, so

needs to be fixed in the OE ptest [2].

I’ve read https://wiki.yoctoproject.org/wiki/Ptest carefully a few times more. There are prescriptions about output format. But I didn’t find any mention about return code processing or a reference to the convention you mentioned in the answer.

I looked through some OE run-ptest scripts. I suspect they don’t verify if some of their tests failed, and exit with 0 even if all their tests failed.

http://git.openembedded.org/openembedded-core/tree/meta/recipes-core/util-linux/util-linux/run-ptest
http://git.openembedded.org/openembedded-core/tree/meta/recipes-support/attr/acl/run-ptest
http://git.openembedded.org/openembedded-core/tree/meta/recipes-support/attr/files/run-ptest
http://git.openembedded.org/openembedded-core/tree/meta/recipes-core/dbus/dbus/run-ptest
http://git.openembedded.org/openembedded-core/tree/meta/recipes-devtools/e2fsprogs/e2fsprogs/run-ptest
http://git.openembedded.org/openembedded-core/tree/meta/recipes-extended/gawk/gawk/run-ptest

Right, looks that the OEQA test case was update since i worked on it [1], so now it takes into account the pass/fail of every ptest.

So the ptest.py needs to implement the same behavior.

Regards,

Anibal

[1] http://git.openembedded.org/openembedded-core/tree/meta/lib/oeqa/runtime/cases/ptest.py#n80

Regarding the LAVA ptest.py script, I mark the run as succeed if there is no critical error in the ptest-runner and we have a QA-reports tool to analyse pass/fails

in detail for every ptest executed [3].

I heard about QA-reports tool but I’ve never used it before, so maybe I missed something.
From https://qa-reports.linaro.org/qcomlt/openembedded-rpb-sumo/build/37/testrun/1890442/suite/linux-ptest/tests/
I see all ptests passed. Still, in log
https://qa-reports.linaro.org/qcomlt/openembedded-rpb-sumo/build/37/testrun/1890442/log
I found 54 failed tests and wasn’t able to find a report which indicates those failures.

Is there such a report? It would be really useful to know that some tests failed.

Thanks

[1] http://git.openembedded.org/openembedded-core/tree/meta/recipes-core/util-linux/util-linux/run-ptest

[2] https://wiki.yoctoproject.org/wiki/Ptest

[3] https://qa-reports.linaro.org/qcomlt/openembedded-rpb-sumo/build/37/testrun/1890442/

Regards,

Anibal

Best regards,
Alex