File indexing completed on 2026-04-28 07:24:56
0001 """Seed the `swf-alarms` context and two initial alarm configs."""
0002 from __future__ import annotations
0003
0004 import time
0005 import uuid
0006
0007 from django.db import migrations
0008
0009
0010 CONTEXT_NAME = 'swf-alarms'
0011
0012 ALARM_CONFIGS = [
0013 {
0014 'title': "PanDA task failure rate — Sakib's tasks",
0015 'data': {
0016 'entry_id': 'alarm_panda_failure_rate_sakib',
0017 'enabled': True,
0018 'severity': 'warning',
0019 'recipients': ['srahman1@bnl.gov', 'wenaus@gmail.com'],
0020 'renotification_window_hours': 24,
0021 'params': {
0022 'threshold': 0.03,
0023 'since_days': 1,
0024 'username': 'Sakib Rahman',
0025 'min_terminal_jobs': 5,
0026 },
0027 },
0028 'content': (
0029 "Alert on PanDA tasks owned by Sakib Rahman whose computed "
0030 "failure rate exceeds the configured threshold over the "
0031 "configured window. Threshold, window, and minimum terminal "
0032 "jobs are in the Check params below.\n"
0033 "\n"
0034 "Dashboard: https://epic-devcloud.org/prod/alarms/\n"
0035 ),
0036 },
0037 {
0038 'title': 'PanDA task failure rate — catch-all',
0039 'data': {
0040 'entry_id': 'alarm_panda_failure_rate_eic_all',
0041 'enabled': True,
0042 'severity': 'info',
0043 'recipients': ['wenaus@gmail.com'],
0044 'renotification_window_hours': 48,
0045 'params': {
0046 'threshold': 0.05,
0047 'since_days': 1,
0048 'min_terminal_jobs': 5,
0049 },
0050 },
0051 'content': (
0052 "Catch-all alert on any PanDA task whose computed failure rate "
0053 "exceeds the configured threshold over the configured window. "
0054 "Torre-only tuning channel for shaping future per-owner "
0055 "alarms. Threshold and window live in the Check params below.\n"
0056 ),
0057 },
0058 ]
0059
0060
0061 def seed(apps, schema_editor):
0062 Entry = apps.get_model('remote_app', 'Entry')
0063 EntryContext = apps.get_model('remote_app', 'EntryContext')
0064 now = time.time()
0065
0066 ctx, _ = EntryContext.objects.get_or_create(
0067 name=CONTEXT_NAME,
0068 defaults={
0069 'title': 'swf-alarms',
0070 'description': 'Alarm configs, firings, and engine-run records.',
0071 'timestamp_created': now,
0072 'timestamp_modified': now,
0073 },
0074 )
0075 for cfg in ALARM_CONFIGS:
0076 eid = cfg['data']['entry_id']
0077 if Entry.objects.filter(context=ctx, kind='alarm',
0078 data__entry_id=eid).exists():
0079 continue
0080 Entry.objects.create(
0081 id=str(uuid.uuid4()),
0082 title=cfg['title'],
0083 content=cfg['content'],
0084 kind='alarm',
0085 context=ctx,
0086 data=cfg['data'],
0087 status='active',
0088 archived=False,
0089 timestamp_created=now,
0090 timestamp_modified=now,
0091 )
0092
0093
0094 def unseed(apps, schema_editor):
0095 Entry = apps.get_model('remote_app', 'Entry')
0096 EntryContext = apps.get_model('remote_app', 'EntryContext')
0097 Entry.objects.filter(context__name=CONTEXT_NAME).delete()
0098 EntryContext.objects.filter(name=CONTEXT_NAME).delete()
0099
0100
0101 class Migration(migrations.Migration):
0102 dependencies = [('remote_app', '0001_initial')]
0103 operations = [migrations.RunPython(seed, unseed)]
