cockpit/bots/flakes-refresh

187 lines
6.0 KiB
Python
Executable File

#!/usr/bin/env python3
# This file is part of Cockpit.
#
# Copyright (C) 2018 Red Hat, Inc.
#
# Cockpit is free software; you can redistribute it and/or modify it
# under the terms of the GNU Lesser General Public License as published by
# the Free Software Foundation; either version 2.1 of the License, or
# (at your option) any later version.
#
# Cockpit is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
# Lesser General Public License for more details.
#
# You should have received a copy of the GNU Lesser General Public License
# along with Cockpit; If not, see <http://www.gnu.org/licenses/>.
import sys
import time
import os
import urllib
import json
import re
import ssl
sys.dont_write_bytecode = True
import task
BOTS = os.path.dirname(os.path.realpath(__file__))
NUMBER_OPEN_ISSUES = 7 # How many issues do we want to have open at a given time?
# How far back does our data go? If a flake gets fixed but is still
# flaky after this many days, the bots open another issue.
WINDOW_DAYS = 21
# This parses the output JSONL format discussed here, where various
# values are grouped:
#
# https://github.com/cockpit-project/cockpituous/blob/master/learn/README.md
# Here we're looking for a field in a record that only has one value
def value(record, field):
values = record.get(field, [])
if len(values) == 1:
return values[0][0] or ""
return None
# Here we're looking for the count of a specific field/value in the record
def count(record, field, only):
values = record.get(field, [])
for value, count in values:
if value != only:
continue
return count
return 0
# For linking flakes to test logs
def slurp_one(url, n, logs):
items_url = url + str(n) + "/items.jsonl"
try:
with urllib.request.urlopen(items_url) as f:
for line in f.readlines():
try:
record = json.loads(line.decode('utf-8'))
logs.setdefault(record["test"], [ ]).append(record["url"])
except ValueError as ex:
sys.stderr.write("{0}: {1}\n".format(url, ex))
except urllib.error.URLError as ex:
if ex.code == 404:
return False
raise
return True
def slurp_failure_logs(url):
logs = { }
n = 0
while slurp_one(url, n, logs):
n = n + 1
return logs
def get_failure_logs(failure_logs, name, context):
match = context.replace("/", "-")
return sorted(filter(lambda url: match in url, failure_logs[name]), reverse=True)[0:10]
# Main
def run(context, verbose=False, **kwargs):
api = task.github.GitHub()
open_issues = api.issues(labels=[ "flake" ])
create_count = NUMBER_OPEN_ISSUES - len(open_issues)
if create_count <= 0:
return 0
if verbose:
sys.stderr.write("Going to create %s new flake issue(s)\n" % create_count)
host = os.environ.get("COCKPIT_LEARN_SERVICE_HOST", "learn-cockpit.apps.ci.centos.org")
port = os.environ.get("COCKPIT_LEARN_SERVICE_PORT", "443")
url = "{0}://{1}:{2}/active/".format("https" if port == "443" else "http", host, port)
cafile = os.path.join(BOTS, "images", "files", "ca.pem")
context = ssl.create_default_context(cafile=cafile)
failure_logs = slurp_failure_logs(url)
# Retrieve the URL
statistics = [ ]
with urllib.request.urlopen(url + "statistics.jsonl", context=context) as f:
for line in f.readlines():
try:
record = json.loads(line.decode('utf-8'))
statistics.append(record)
except ValueError as ex:
sys.stderr.write("{0}: {1}\n".format(url, ex))
tests = { }
for record in statistics:
test = value(record, "test")
context = value(record, "context")
status = value(record, "status")
tracker = value(record, "tracker")
# Flaky tests only score on those that fail and are not tracked
if test is not None and status == "failure" and not tracker:
merged = count(record, "merged", True)
not_merged = count(record, "merged", False)
null_merged = count(record, "merged", None)
total = merged + not_merged + null_merged
# And the key is that they were merged anyway
if total > 10:
tests.setdefault(test, [ ]).append((merged / total, context, record))
scores = [ ]
for n, t in tests.items():
scores.append((sum(map(lambda f: f[0], t))/len(t), n, t))
closed_issues = api.issues(labels=["flake"], state="closed", since=(time.time() - (WINDOW_DAYS * 86400)))
def find_in_issues(issues, name):
for issue in issues:
if name in issue["title"]:
return True
return False
def url_desc(url):
m = re.search("pull-[0-9]+", url)
return m.group(0) if m else url
def failure_description(name, f, logs):
return ("%s%% on %s\n" % (int(f[0]*100), f[1]) +
"".join(map(lambda url: " - [%s](%s)\n" % (url_desc(url), url),
get_failure_logs(logs, name, f[1]))))
scores.sort(reverse=True)
for score, name, failures in scores:
if find_in_issues(open_issues, name) or find_in_issues(closed_issues, name):
continue
if verbose:
sys.stderr.write("Opening issue for %s\n" % name)
source = "<details><summary>Source material</summary>\n\n```json\n%s\n```\n</details>\n" % "\n".join(map(lambda f: json.dumps(f[2], indent=2), failures))
data = {
"title": "%s is flaky" % name,
"body": ("\n".join(map(lambda f: failure_description(name, f, failure_logs), failures)) +
"\n\n" + source),
"labels": [ "flake" ]
}
api.post("issues", data)
create_count -= 1
if create_count == 0:
break
return 0
if __name__ == '__main__':
task.main(function=run, title="Create issues for test flakes")