def test_get_items_to_be_processed(self):
p = PillowError(
date_created=datetime.datetime.utcnow() - datetime.timedelta(days=1),
date_next_attempt=datetime.datetime.utcnow() - datetime.timedelta(days=2),
date_last_attempt=datetime.datetime.utcnow() - datetime.timedelta(days=3),
)
p.save()
self.addCleanup(p.delete)
errors = list(PillowRetryEnqueuingOperation.get_items_to_be_processed(
datetime.datetime.utcnow()))
self.assertTrue(errors)
def test_bulk_reset(self):
for i in range(0, 5):
error = create_error(_change(id=i), attempts=const.PILLOW_RETRY_QUEUE_MAX_PROCESSING_ATTEMPTS)
error.save()
errors = PillowError.get_errors_to_process(datetime.utcnow()).all()
self.assertEqual(len(errors), 0)
PillowError.bulk_reset_attempts(datetime.utcnow())
errors = PillowError.get_errors_to_process(datetime.utcnow()).all()
self.assertEqual(len(errors), 5)
def test_get_or_create(self):
message = 'abcd'
id = '12335'
error = create_error({'id': id}, message=message, attempts=2)
error.save()
get = PillowError.get_or_create({'id': id}, FakePillow())
self.assertEqual(get.total_attempts, 2)
self.assertEqual(get.current_attempt, 2)
self.assertTrue(message in error.error_traceback)
new = PillowError.get_or_create({'id': id}, FakePillow1())
self.assertIsNone(new.id)
self.assertEqual(new.current_attempt, 0)
def test_bulk_reset_cutoff(self):
for i in range(0, 3):
error = create_error({'id': i}, attempts=1)
if i >= 1:
error.total_attempts = PillowError.multi_attempts_cutoff() + 1
error.save()
errors = PillowError.get_errors_to_process(datetime.utcnow()).all()
self.assertEqual(len(errors), 0)
PillowError.bulk_reset_attempts(datetime.utcnow())
errors = PillowError.get_errors_to_process(datetime.utcnow()).all()
self.assertEqual(len(errors), 2)
def test_bulk_reset_cutoff(self):
for i in range(0, 3):
error = create_error(_change(id=i), attempts=1)
if i >= 1:
error.total_attempts = const.PILLOW_RETRY_MULTI_ATTEMPTS_CUTOFF + 1
error.save()
errors = PillowError.get_errors_to_process(datetime.utcnow()).all()
self.assertEqual(len(errors), 0)
PillowError.bulk_reset_attempts(datetime.utcnow())
errors = PillowError.get_errors_to_process(datetime.utcnow()).all()
self.assertEqual(len(errors), 2)
def test_get_or_create(self):
message = 'abcd'
id = '12335'
error = create_error(_change(id=id), message=message, attempts=2)
error.save()
get = PillowError.get_or_create(_change(id=id), FakePillow())
self.assertEqual(get.total_attempts, 2)
self.assertEqual(get.current_attempt, 2)
self.assertTrue(message in error.error_traceback)
another_pillow = make_fake_constructed_pillow('FakePillow1', '')
new = PillowError.get_or_create(_change(id=id), another_pillow)
self.assertIsNone(new.id)
self.assertEqual(new.current_attempt, 0)
def create_error(change, message='message', attempts=0, pillow=None, ex_class=None):
change = force_to_change(change)
change.metadata = ChangeMeta(data_source_type='couch', data_source_name='test_commcarehq', document_id=change.id)
error = PillowError.get_or_create(change, pillow or FakePillow())
for n in range(0, attempts):
error.add_attempt(*get_ex_tb(message, ex_class=ex_class))
return error
def handle(self, *args, **options):
root_dir = settings.FILEPATH
git_snapshot = gitinfo.get_project_snapshot(root_dir, submodules=True)
git_snapshot['diff_url'] = options.get('url', None)
deploy = HqDeploy(
date=datetime.utcnow(),
user=options['user'],
environment=options['environment'],
code_snapshot=git_snapshot,
)
deploy.save()
# reset PillowTop errors in the hope that a fix has been deployed
rows_updated = PillowError.bulk_reset_attempts(datetime.utcnow())
if rows_updated:
print "\n---------------- Pillow Errors Reset ----------------\n" \
"{} pillow errors queued for retry\n".format(rows_updated)
if options['mail_admins']:
snapshot_table = render_to_string('hqadmin/partials/project_snapshot.html', dictionary={'snapshot': git_snapshot})
message = "Deployed by %s, cheers!" % options['user']
snapshot_body = "<html><head><title>Deploy Snapshot</title></head><body><h2>%s</h2>%s</body></html>" % (message, snapshot_table)
call_command('mail_admins', snapshot_body, **{'subject': 'Deploy successful', 'html': True})
def process_pillow_retry(error_doc_id):
# Redis error logged in get_redis_client
try:
client = cache_core.get_redis_client()
except cache_core.RedisClientError:
return
# Prevent more than one task from processing this error, just in case
# it got enqueued twice.
lock = client.lock(
"pillow-retry-processing-%s" % error_doc_id,
timeout=settings.PILLOW_RETRY_PROCESSING_LOCK_TIMEOUT*60
)
if lock.acquire(blocking=False):
try:
error_doc = PillowError.objects.get(id=error_doc_id)
except PillowError.DoesNotExist:
release_lock(lock, True)
return
pillow_name_or_class = error_doc.pillow
try:
pillow = get_pillow_by_name(pillow_name_or_class)
except PillowNotFoundError:
pillow = None
if not pillow:
notify_error((
"Could not find pillowtop class '%s' while attempting a retry. "
"If this pillow was recently deleted then this will be automatically cleaned up eventually. "
"If not, then this should be looked into."
) % pillow_name_or_class)
try:
error_doc.total_attempts = PillowError.multi_attempts_cutoff() + 1
error_doc.save()
finally:
release_lock(lock, True)
return
change = error_doc.change_object
try:
change_metadata = change.metadata
if change_metadata:
document_store = get_document_store(
data_source_type=change_metadata.data_source_type,
data_source_name=change_metadata.data_source_name,
domain=change_metadata.domain
)
change.document_store = document_store
pillow.process_change(change)
except Exception:
ex_type, ex_value, ex_tb = sys.exc_info()
error_doc.add_attempt(ex_value, ex_tb)
error_doc.queued = False
error_doc.save()
else:
error_doc.delete()
finally:
release_lock(lock, True)
def test_empty_metadata(self):
change = _change(id='123')
error = PillowError.get_or_create(change, GetDocPillow())
error.save()
process_pillow_retry(error)
error = PillowError.objects.get(pk=error.id)
self.assertEquals(error.total_attempts, 1)
def test_get_errors_to_process_queued_update(self):
date = datetime.utcnow()
error = create_error({'id': 1}, attempts=0)
error.date_next_attempt = date
error.save()
errors = PillowError.get_errors_to_process(
date,
).all()
self.assertEqual(len(errors), 1)
# check that calling update on the return value has the desired effect
errors.update(queued=True)
errors = PillowError.get_errors_to_process(
date,
).all()
self.assertEqual(len(errors), 0)
def test_get_errors_to_process(self):
# Only re-process errors with
# current_attempt < const.PILLOW_RETRY_QUEUE_MAX_PROCESSING_ATTEMPTS
date = datetime.utcnow()
for i in range(0, 5):
error = create_error(_change(id=i), attempts=i+1)
error.date_next_attempt = date.replace(day=i+1)
error.save()
errors = PillowError.get_errors_to_process(
date.replace(day=1),
).all()
self.assertEqual(len(errors), 1)
errors = PillowError.get_errors_to_process(
date.replace(day=5),
).all()
self.assertEqual(len(errors), 3)
def test_null_meta_date(self):
id = '12335'
meta = {
'domains': ['a' * 247, '123456789'],
'doc_type': 'something',
'date': None,
}
get = PillowError.get_or_create({'id': id}, FakePillow(), meta)
self.assertEqual(None, get.doc_date)
def handle(self, *args, **options):
root_dir = settings.FILEPATH
git_snapshot = gitinfo.get_project_snapshot(root_dir, submodules=True)
compare_url = git_snapshot['diff_url'] = options.get('url', None)
deploy = HqDeploy(
date=datetime.utcnow(),
user=options['user'],
environment=options['environment'],
code_snapshot=git_snapshot,
)
deploy.save()
# reset PillowTop errors in the hope that a fix has been deployed
rows_updated = PillowError.bulk_reset_attempts(datetime.utcnow())
if rows_updated:
print "\n---------------- Pillow Errors Reset ----------------\n" \
"{} pillow errors queued for retry\n".format(rows_updated)
deploy_notification_text = (
"CommCareHQ has been successfully deployed to *{}* by *{}*. "
"Find the diff {{diff_link}}".format(
options['environment'],
options['user'],
)
)
if hasattr(settings, 'MIA_THE_DEPLOY_BOT_API'):
link = diff_link(STYLE_SLACK, compare_url)
requests.post(settings.MIA_THE_DEPLOY_BOT_API, data=json.dumps({
"username": "******",
"text": deploy_notification_text.format(diff_link=link),
}))
if settings.DATADOG_API_KEY:
tags = ['environment:{}'.format(options['environment'])]
link = diff_link(STYLE_MARKDOWN, compare_url)
datadog_api.Event.create(
title="Deploy Success",
text=deploy_notification_text.format(diff_link=link),
tags=tags,
alert_type="success"
)
print "\n=============================================================\n" \
"Congratulations! Deploy Complete.\n\n" \
"Don't forget to keep an eye on the deploy dashboard to " \
"make sure everything is running smoothly.\n\n" \
"https://p.datadoghq.com/sb/5c4af2ac8-1f739e93ef" \
"\n=============================================================\n"
if options['mail_admins']:
message_body = get_deploy_email_message_body(
environment=options['environment'], user=options['user'],
compare_url=compare_url)
call_command('mail_admins', message_body, **{'subject': 'Deploy successful', 'html': True})
def test_pillow_not_found(self):
error = PillowError.objects.create(
doc_id='missing-pillow',
pillow='badmodule.NotARealPillow',
date_created=datetime.utcnow(),
date_last_attempt=datetime.utcnow()
)
# make sure this doesn't error
process_pillow_retry(error.id)
# and that its total_attempts was bumped above the threshold
self.assertTrue(PillowError.objects.get(pk=error.pk).total_attempts > PillowError.multi_attempts_cutoff())
def test_get_or_create_meta(self):
id = '12335'
date = '2013-12-05T08:52:19Z'
meta = {
'domains': ['a' * 247, '123456789'],
'doc_type': 'something',
'date': date,
}
get = PillowError.get_or_create({'id': id}, FakePillow(), meta)
self.assertEqual(get.domains, 'a' * 247 + ',1234...')
self.assertEqual(get.doc_type, 'something')
self.assertEqual(get.doc_date, parse(date))
get.save()
def handle(self, *args, **options):
root_dir = settings.FILEPATH
git_snapshot = gitinfo.get_project_snapshot(root_dir, submodules=True)
git_snapshot['diff_url'] = options.get('url', None)
deploy = HqDeploy(
date=datetime.utcnow(),
user=options['user'],
environment=options['environment'],
code_snapshot=git_snapshot,
)
deploy.save()
# reset PillowTop errors in the hope that a fix has been deployed
rows_updated = PillowError.bulk_reset_attempts(datetime.utcnow())
if rows_updated:
print "\n---------------- Pillow Errors Reset ----------------\n" \
"{} pillow errors queued for retry\n".format(rows_updated)
deploy_notification_text = (
"CommCareHQ has been successfully deployed to *{}* by *{}*. "
"Find the diff {{diff_link}}".format(
options['environment'],
options['user'],
)
)
if hasattr(settings, 'MIA_THE_DEPLOY_BOT_API'):
link = diff_link(STYLE_SLACK, git_snapshot['diff_url'])
requests.post(settings.MIA_THE_DEPLOY_BOT_API, data=json.dumps({
"channel": "#dev",
"username": "******",
"text": deploy_notification_text.format(diff_link=link),
"icon_emoji": ":see_no_evil:"
}))
if settings.DATADOG_API_KEY:
tags = ['environment:{}'.format(options['environment'])]
link = diff_link(STYLE_MARKDOWN, git_snapshot['diff_url'])
datadog_api.Event.create(
title="Deploy Success",
text=deploy_notification_text.format(diff_link=link),
tags=tags
)
if options['mail_admins']:
snapshot_table = render_to_string('hqadmin/partials/project_snapshot.html', dictionary={'snapshot': git_snapshot})
message = "Deployed by %s, cheers!" % options['user']
snapshot_body = "<html><head><title>Deploy Snapshot</title></head><body><h2>%s</h2>%s</body></html>" % (message, snapshot_table)
call_command('mail_admins', snapshot_body, **{'subject': 'Deploy successful', 'html': True})
def test_get_errors_to_process_queued(self):
date = datetime.utcnow()
error = create_error({'id': 1}, attempts=0)
error.date_next_attempt = date
error.save()
queued_error = create_error({'id': 2}, attempts=0)
queued_error.date_next_attempt = date
queued_error.queued = True
queued_error.save()
errors = PillowError.get_errors_to_process(
date,
).all()
self.assertEqual(len(errors), 1)
self.assertEqual(error.id, errors[0]['id'])
def handle_pillow_error(pillow, change, exception):
from pillow_retry.models import PillowError
error_id = None
if pillow.retry_errors:
error = PillowError.get_or_create(change, pillow)
error.add_attempt(exception, sys.exc_info()[2])
error.save()
error_id = error.id
pillow_logging.exception(
"[%s] Error on change: %s, %s. Logged as: %s" % (
pillow.get_name(),
change['id'],
exception,
error_id
)
)
def _handle_pillow_error(self, change, exception):
try:
# This breaks the module boundary by using a show function defined in commcare-hq
# but it was decided that it wasn't worth the effort to maintain the separation.
meta = self.get_couch_db().show('domain/domain_date', change['id'])
except ResourceNotFound:
# Show function does not exist
meta = None
error = PillowError.get_or_create(change, self, change_meta=meta)
error.add_attempt(exception, sys.exc_info()[2])
error.save()
pillow_logging.exception(
"[%s] Error on change: %s, %s. Logged as: %s" % (
self.get_name(),
change['id'],
exception,
error.id
)
)
def test_get_errors_to_process_max_limit(self):
# see settings.PILLOW_RETRY_MULTI_ATTEMPTS_CUTOFF
date = datetime.utcnow()
def make_error(id, current_attempt, total_attempts):
error = create_error({'id': id})
error.date_next_attempt = date
error.current_attempt = current_attempt
error.total_attempts = total_attempts
error.save()
# current_attempts <= limit, total_attempts <= limit
make_error(
'to-process1',
settings.PILLOW_RETRY_QUEUE_MAX_PROCESSING_ATTEMPTS,
settings.PILLOW_RETRY_MULTI_ATTEMPTS_CUTOFF
)
# current_attempts = 0, total_attempts > limit
make_error(
'to-process2',
0,
settings.PILLOW_RETRY_MULTI_ATTEMPTS_CUTOFF + 1
)
# current_attempts > limit, total_attempts <= limit
make_error(
'not-processed1',
settings.PILLOW_RETRY_QUEUE_MAX_PROCESSING_ATTEMPTS + 1,
settings.PILLOW_RETRY_MULTI_ATTEMPTS_CUTOFF
)
# current_attempts <= limit, total_attempts > limit
make_error(
'not-processed2',
settings.PILLOW_RETRY_QUEUE_MAX_PROCESSING_ATTEMPTS,
settings.PILLOW_RETRY_MULTI_ATTEMPTS_CUTOFF + 1
)
errors = PillowError.get_errors_to_process(date, fetch_full=True).all()
self.assertEqual(len(errors), 2)
docs_to_process = {e.doc_id for e in errors}
self.assertEqual({'to-process1', 'to-process2'}, docs_to_process)
def handle_pillow_error(pillow, change, exception):
from couchdbkit import ResourceNotFound
from pillow_retry.models import PillowError
meta = None
if hasattr(pillow, 'get_couch_db'):
try:
meta = pillow.get_couch_db().show('domain_shows/domain_date', change['id'])
except ResourceNotFound:
pass
error = PillowError.get_or_create(change, pillow, change_meta=meta)
error.add_attempt(exception, sys.exc_info()[2])
error.save()
pillow_logging.exception(
"[%s] Error on change: %s, %s. Logged as: %s" % (
pillow.get_name(),
change['id'],
exception,
error.id
)
)
def handle_pillow_error(pillow, change, exception):
from pillow_retry.models import PillowError
pillow_logging.exception("[%s] Error on change: %s, %s" % (
pillow.get_name(),
change['id'],
exception,
))
datadog_counter('commcare.change_feed.changes.exceptions', tags=[
'pillow_name:{}'.format(pillow.get_name()),
])
# keep track of error attempt count
change.increment_attempt_count()
# always retry document missing errors, because the error is likely with couch
if pillow.retry_errors or isinstance(exception, DocumentMissingError):
error = PillowError.get_or_create(change, pillow)
error.add_attempt(exception, sys.exc_info()[2], change.metadata)
error.save()
def handle(self, **options):
compare_url = options.get('url', None)
minutes = options.get('minutes', None)
deploy = HqDeploy(date=datetime.utcnow(),
user=options['user'],
environment=options['environment'],
diff_url=compare_url,
commit=options['commit'])
deploy.save()
# reset PillowTop errors in the hope that a fix has been deployed
rows_updated = PillowError.bulk_reset_attempts(datetime.utcnow())
if rows_updated:
print("\n---------------- Pillow Errors Reset ----------------\n" \
"{} pillow errors queued for retry\n".format(rows_updated))
deploy_notification_text = (
"CommCareHQ has been successfully deployed to *{}* by *{}* in *{}* minutes. "
.format(
options['environment'],
options['user'],
minutes or '?',
))
if options['environment'] == 'production':
deploy_notification_text += "Monitor the {dashboard_link}. "
if settings.MOBILE_INTEGRATION_TEST_TOKEN:
deploy_notification_text += "Check the integration {integration_tests_link}. "
requests.get(
'https://jenkins.dimagi.com/job/integration-tests/build',
params={'token': settings.MOBILE_INTEGRATION_TEST_TOKEN},
)
requests.get(
'https://jenkins.dimagi.com/job/integration-tests-pipeline/build',
params={'token': settings.MOBILE_INTEGRATION_TEST_TOKEN},
)
deploy_notification_text += "Find the diff {diff_link}"
if settings.DATADOG_API_KEY:
link = diff_link(compare_url)
create_metrics_event(
title="Deploy Success",
text=deploy_notification_text.format(
dashboard_link=dashboard_link(DASHBOARD_URL),
diff_link=link,
integration_tests_link=integration_tests_link(
INTEGRATION_TEST_URL)),
tags={'environment': options['environment']},
alert_type="success")
print(
"\n=============================================================\n"
"Congratulations! Deploy Complete.\n\n"
"Don't forget to keep an eye on the deploy dashboard to "
"make sure everything is running smoothly.\n\n"
"https://app.datadoghq.com/dashboard/xch-zwt-vzv/hq-deploy-dashboard?tpl_var_environment={}"
"\n=============================================================\n"
.format(settings.SERVER_ENVIRONMENT))
def _get_items(self, utcnow):
errors = PillowError.get_errors_to_process(
utcnow=utcnow,
)
return (dict(id=e['id'], key=e['date_next_attempt']) for e in errors)
def create_error(change, message='message', attempts=0, pillow=None, ex_class=None):
change.metadata = ChangeMeta(data_source_type='couch', data_source_name='test_commcarehq', document_id=change.id)
error = PillowError.get_or_create(change, pillow or FakePillow())
for n in range(0, attempts):
error.add_attempt(*get_ex_tb(message, ex_class=ex_class))
return error
def handle(self, *args, **options):
compare_url = options.get('url', None)
minutes = options.get('minutes', None)
deploy = HqDeploy(date=datetime.utcnow(),
user=options['user'],
environment=options['environment'],
diff_url=compare_url)
deploy.save()
# reset PillowTop errors in the hope that a fix has been deployed
rows_updated = PillowError.bulk_reset_attempts(datetime.utcnow())
if rows_updated:
print "\n---------------- Pillow Errors Reset ----------------\n" \
"{} pillow errors queued for retry\n".format(rows_updated)
deploy_notification_text = (
"CommCareHQ has been successfully deployed to *{}* by *{}* in *{}* minutes. "
.format(
options['environment'],
options['user'],
minutes or '?',
))
if options['environment'] == 'production':
deploy_notification_text += "Monitor the {dashboard_link}. "
if settings.MOBILE_INTEGRATION_TEST_TOKEN:
deploy_notification_text += "Check the integration {integration_tests_link}. "
requests.get(
'https://jenkins.dimagi.com/job/integration-tests/build',
params={'token': settings.MOBILE_INTEGRATION_TEST_TOKEN},
)
deploy_notification_text += "Find the diff {diff_link}"
if hasattr(settings, 'MIA_THE_DEPLOY_BOT_API'):
link = diff_link(STYLE_SLACK, compare_url)
if options['environment'] == 'staging':
channel = '#staging'
else:
channel = '#hq-ops'
requests.post(
settings.MIA_THE_DEPLOY_BOT_API,
data=json.dumps({
"username":
"******",
"channel":
channel,
"text":
deploy_notification_text.format(
dashboard_link=dashboard_link(STYLE_SLACK,
DASHBOARD_URL),
diff_link=link,
integration_tests_link=integration_tests_link(
STYLE_SLACK, INTEGRATION_TEST_URL)),
}))
if settings.DATADOG_API_KEY:
tags = ['environment:{}'.format(options['environment'])]
link = diff_link(STYLE_MARKDOWN, compare_url)
datadog_api.Event.create(
title="Deploy Success",
text=deploy_notification_text.format(
dashboard_link=dashboard_link(STYLE_MARKDOWN,
DASHBOARD_URL),
diff_link=link,
integration_tests_link=integration_tests_link(
STYLE_MARKDOWN, INTEGRATION_TEST_URL)),
tags=tags,
alert_type="success")
print "\n=============================================================\n" \
"Congratulations! Deploy Complete.\n\n" \
"Don't forget to keep an eye on the deploy dashboard to " \
"make sure everything is running smoothly.\n\n" \
"https://p.datadoghq.com/sb/5c4af2ac8-1f739e93ef" \
"\n=============================================================\n"
if options['mail_admins']:
message_body = get_deploy_email_message_body(
environment=options['environment'],
user=options['user'],
compare_url=compare_url)
call_command('mail_admins', message_body, **{
'subject': 'Deploy successful',
'html': True
})
def handle(self, **options):
compare_url = options.get('url', None)
minutes = options.get('minutes', None)
deploy = HqDeploy(
date=datetime.utcnow(),
user=options['user'],
environment=options['environment'],
diff_url=compare_url
)
deploy.save()
# reset PillowTop errors in the hope that a fix has been deployed
rows_updated = PillowError.bulk_reset_attempts(datetime.utcnow())
if rows_updated:
print("\n---------------- Pillow Errors Reset ----------------\n" \
"{} pillow errors queued for retry\n".format(rows_updated))
deploy_notification_text = (
"CommCareHQ has been successfully deployed to *{}* by *{}* in *{}* minutes. ".format(
options['environment'],
options['user'],
minutes or '?',
)
)
if options['environment'] == 'production':
deploy_notification_text += "Monitor the {dashboard_link}. "
if settings.MOBILE_INTEGRATION_TEST_TOKEN:
deploy_notification_text += "Check the integration {integration_tests_link}. "
requests.get(
'https://jenkins.dimagi.com/job/integration-tests/build',
params={'token': settings.MOBILE_INTEGRATION_TEST_TOKEN},
)
requests.get(
'https://jenkins.dimagi.com/job/integration-tests-pipeline/build',
params={'token': settings.MOBILE_INTEGRATION_TEST_TOKEN},
)
deploy_notification_text += "Find the diff {diff_link}"
if hasattr(settings, 'MIA_THE_DEPLOY_BOT_API'):
link = diff_link(STYLE_SLACK, compare_url)
if options['environment'] == 'staging':
channel = '#staging'
elif options['environment'] == 'icds':
channel = '#nic-server-standup'
else:
channel = '#hq-ops'
requests.post(settings.MIA_THE_DEPLOY_BOT_API, data=json.dumps({
"username": "******",
"channel": channel,
"text": deploy_notification_text.format(
dashboard_link=dashboard_link(STYLE_SLACK, DASHBOARD_URL),
diff_link=link,
integration_tests_link=integration_tests_link(STYLE_SLACK, INTEGRATION_TEST_URL)
),
}))
if settings.DATADOG_API_KEY:
tags = ['environment:{}'.format(options['environment'])]
link = diff_link(STYLE_MARKDOWN, compare_url)
datadog_api.Event.create(
title="Deploy Success",
text=deploy_notification_text.format(
dashboard_link=dashboard_link(STYLE_MARKDOWN, DASHBOARD_URL),
diff_link=link,
integration_tests_link=integration_tests_link(STYLE_MARKDOWN, INTEGRATION_TEST_URL)
),
tags=tags,
alert_type="success"
)
print("\n=============================================================\n" \
"Congratulations! Deploy Complete.\n\n" \
"Don't forget to keep an eye on the deploy dashboard to " \
"make sure everything is running smoothly.\n\n" \
"https://p.datadoghq.com/sb/5c4af2ac8-1f739e93ef" \
"\n=============================================================\n")
if options['mail_admins']:
message_body = get_deploy_email_message_body(user=options['user'], compare_url=compare_url)
call_command('mail_admins', message_body, **{'subject': 'Deploy successful', 'html': True})
if settings.DAILY_DEPLOY_EMAIL:
recipient = settings.DAILY_DEPLOY_EMAIL
subject = 'Deploy Successful - {}'.format(options['environment'])
send_HTML_email(subject=subject,
recipient=recipient,
html_content=message_body)
if settings.SENTRY_CONFIGURED and settings.SENTRY_API_KEY:
create_update_sentry_release()
notify_sentry_deploy(minutes)
def handle(self, **options):
compare_url = options.get('url', None)
minutes = options.get('minutes', None)
deploy = HqDeploy(
date=datetime.utcnow(),
user=options['user'],
environment=options['environment'],
diff_url=compare_url
)
deploy.save()
# reset PillowTop errors in the hope that a fix has been deployed
rows_updated = PillowError.bulk_reset_attempts(datetime.utcnow())
if rows_updated:
print("\n---------------- Pillow Errors Reset ----------------\n" \
"{} pillow errors queued for retry\n".format(rows_updated))
deploy_notification_text = (
"CommCareHQ has been successfully deployed to *{}* by *{}* in *{}* minutes. ".format(
options['environment'],
options['user'],
minutes or '?',
)
)
if options['environment'] == 'production':
deploy_notification_text += "Monitor the {dashboard_link}. "
if settings.MOBILE_INTEGRATION_TEST_TOKEN:
deploy_notification_text += "Check the integration {integration_tests_link}. "
requests.get(
'https://jenkins.dimagi.com/job/integration-tests/build',
params={'token': settings.MOBILE_INTEGRATION_TEST_TOKEN},
)
requests.get(
'https://jenkins.dimagi.com/job/integration-tests-pipeline/build',
params={'token': settings.MOBILE_INTEGRATION_TEST_TOKEN},
)
deploy_notification_text += "Find the diff {diff_link}"
if settings.DATADOG_API_KEY:
tags = ['environment:{}'.format(options['environment'])]
link = diff_link(compare_url)
datadog_api.Event.create(
title="Deploy Success",
text=deploy_notification_text.format(
dashboard_link=dashboard_link(DASHBOARD_URL),
diff_link=link,
integration_tests_link=integration_tests_link(INTEGRATION_TEST_URL)
),
tags=tags,
alert_type="success"
)
print(
"\n=============================================================\n"
"Congratulations! Deploy Complete.\n\n"
"Don't forget to keep an eye on the deploy dashboard to "
"make sure everything is running smoothly.\n\n"
"https://app.datadoghq.com/dashboard/xch-zwt-vzv/hq-deploy-dashboard?tpl_var_environment={}"
"\n=============================================================\n".format(
settings.SERVER_ENVIRONMENT
)
)
if options['mail_admins']:
message_body = get_deploy_email_message_body(user=options['user'], compare_url=compare_url)
subject = 'Deploy Successful - {}'.format(options['environment'])
call_command('mail_admins', message_body, **{'subject': subject, 'html': True})
if settings.DAILY_DEPLOY_EMAIL:
recipient = settings.DAILY_DEPLOY_EMAIL
send_HTML_email(subject=subject,
recipient=recipient,
html_content=message_body)
if settings.SENTRY_CONFIGURED and settings.SENTRY_API_KEY:
create_update_sentry_release()
notify_sentry_deploy(minutes)
def create_error(change, message='message', attempts=0, pillow=None, ex_class=None):
error = PillowError.get_or_create(change, pillow or FakePillow())
for n in range(0, attempts):
error.add_attempt(*get_ex_tb(message, ex_class=ex_class))
return error
def options(self):
return [(p, p) for p in PillowError.get_pillows()]
def options(self):
return [(e, e) for e in PillowError.get_error_types()]
def _get_items(self, utcnow):
errors = PillowError.get_errors_to_process(utcnow=utcnow, )
return (dict(id=e['id'], key=e['date_next_attempt']) for e in errors)
def _get_items(utcnow):
errors = PillowError.get_errors_to_process(utcnow=utcnow, limit=1000)
return [
QueueItem(id=e.id, key=e.date_next_attempt, object=e)
for e in errors
]
def process_pillow_retry(error_doc_id):
# Redis error logged in get_redis_client
try:
client = cache_core.get_redis_client()
except cache_core.RedisClientError:
return
# Prevent more than one task from processing this error, just in case
# it got enqueued twice.
lock = client.lock(
"pillow-retry-processing-%s" % error_doc_id,
timeout=settings.PILLOW_RETRY_PROCESSING_LOCK_TIMEOUT*60
)
if lock.acquire(blocking=False):
try:
error_doc = PillowError.objects.get(id=error_doc_id)
except PillowError.DoesNotExist:
release_lock(lock, True)
return
pillow_name_or_class = error_doc.pillow
try:
pillow = get_pillow_by_name(pillow_name_or_class)
except PillowNotFoundError:
if not settings.UNIT_TESTING:
_assert = soft_assert(to='@'.join(['czue', 'dimagi.com']))
_assert(False, 'Pillow retry {} is still using legacy class {}'.format(
error_doc.pk, pillow_name_or_class
))
pillow = _try_legacy_import(pillow_name_or_class)
if not pillow:
notify_error((
"Could not find pillowtop class '%s' while attempting a retry. "
"If this pillow was recently deleted then this will be automatically cleaned up eventually. "
"If not, then this should be looked into."
) % pillow_name_or_class)
try:
error_doc.total_attempts = PillowError.multi_attempts_cutoff() + 1
error_doc.save()
finally:
release_lock(lock, True)
return
change = error_doc.change_object
if getattr(pillow, 'include_docs', False):
try:
change.set_document(pillow.get_couch_db().open_doc(change.id))
except ResourceNotFound:
change.deleted = True
try:
try:
from corehq.apps.userreports.pillow import ConfigurableReportKafkaPillow
if isinstance(pillow, ConfigurableReportKafkaPillow):
raise Exception('this is temporarily not supported!')
except ImportError:
pass
pillow.process_change(change, is_retry_attempt=True)
except Exception:
ex_type, ex_value, ex_tb = sys.exc_info()
error_doc.add_attempt(ex_value, ex_tb)
error_doc.queued = False
error_doc.save()
else:
error_doc.delete()
finally:
release_lock(lock, True)
def _get_items(utcnow):
errors = PillowError.get_errors_to_process(utcnow=utcnow, limit=1000)
return [dict(id=e['id'], key=e['date_next_attempt']) for e in errors]
def _get_items(utcnow):
errors = PillowError.get_errors_to_process(utcnow=utcnow, limit=1000)
error_pks = [error['id'] for error in errors]
PillowError.objects.filter(pk__in=error_pks).update(queued=True)
return [dict(id=e['id'], key=e['date_next_attempt']) for e in errors]