stacktach/stacktach/views.py
2013-02-14 15:27:02 -05:00

489 lines
16 KiB
Python

# Copyright 2012 - Dark Secret Software Inc.
import datetime
import json
import pprint
from django import db
from django import http
from django.shortcuts import render_to_response
from django import template
from stacktach import datetime_to_decimal as dt
from stacktach import db as stackdb
from stacktach import models
from stacktach import image_type
STACKDB = stackdb
def _extract_states(payload):
return {
'state' : payload.get('state', ""),
'old_state' : payload.get('old_state', ""),
'old_task' : payload.get('old_task_state', ""),
'task' : payload.get('new_task_state', ""),
'image_type' : image_type.get_numeric_code(payload)
}
def _monitor_message(routing_key, body):
event = body['event_type']
publisher = body['publisher_id']
request_id = body['_context_request_id']
parts = publisher.split('.')
service = parts[0]
if len(parts) > 1:
host = ".".join(parts[1:])
else:
host = None
payload = body['payload']
request_spec = payload.get('request_spec', None)
# instance UUID's seem to hide in a lot of odd places.
instance = payload.get('instance_id', None)
instance = payload.get('instance_uuid', instance)
if not instance:
instance = payload.get('exception', {}).get('kwargs', {}).get('uuid')
if not instance:
instance = payload.get('instance', {}).get('uuid')
tenant = body.get('_context_project_id', None)
tenant = payload.get('tenant_id', tenant)
resp = dict(host=host, instance=instance, publisher=publisher,
service=service, event=event, tenant=tenant,
request_id=request_id)
resp.update(_extract_states(payload))
return resp
def _compute_update_message(routing_key, body):
publisher = None
instance = None
args = body['args']
host = args['host']
request_id = body['_context_request_id']
service = args['service_name']
event = body['method']
tenant = args.get('_context_project_id', None)
resp = dict(host=host, instance=instance, publisher=publisher,
service=service, event=event, tenant=tenant,
request_id=request_id)
payload = body.get('payload', {})
resp.update(_extract_states(payload))
return resp
# routing_key : handler
HANDLERS = {'monitor.info':_monitor_message,
'monitor.error':_monitor_message,
'':_compute_update_message}
def start_kpi_tracking(lifecycle, raw):
"""Start the clock for kpi timings when we see an instance.update
coming in from an api node."""
if raw.event != "compute.instance.update":
return
if "api" not in raw.service:
return
tracker = STACKDB.create_request_tracker(request_id=raw.request_id,
start=raw.when,
lifecycle=lifecycle,
last_timing=None,
duration=str(0.0))
STACKDB.save(tracker)
def update_kpi(timing, raw):
"""Whenever we get a .end event, use the Timing object to
compute our current end-to-end duration.
Note: it may not be completely accurate if the operation is
still in-process, but we have no way of knowing it's still
in-process without mapping the original command with the
expected .end event (that's a whole other thing)
Until then, we'll take the lazy route and be aware of these
potential fence-post issues."""
trackers = STACKDB.find_request_trackers(request_id=raw.request_id)
if len(trackers) == 0:
return
tracker = trackers[0]
tracker.last_timing = timing
tracker.duration = timing.end_when - tracker.start
STACKDB.save(tracker)
def aggregate_lifecycle(raw):
"""Roll up the raw event into a Lifecycle object
and a bunch of Timing objects.
We can use this for summarized timing reports.
Additionally, we can use this processing to give
us end-to-end user request timings for kpi reports.
"""
if not raw.instance:
return
# While we hope only one lifecycle ever exists it's quite
# likely we get multiple due to the workers and threads.
lifecycle = None
lifecycles = STACKDB.find_lifecycles(instance=raw.instance)
if len(lifecycles) > 0:
lifecycle = lifecycles[0]
if not lifecycle:
lifecycle = STACKDB.create_lifecycle(instance=raw.instance)
lifecycle.last_raw = raw
lifecycle.last_state = raw.state
lifecycle.last_task_state = raw.old_task
STACKDB.save(lifecycle)
event = raw.event
parts = event.split('.')
step = parts[-1]
name = '.'.join(parts[:-1])
if not step in ['start', 'end']:
# Perhaps it's an operation initiated in the API?
start_kpi_tracking(lifecycle, raw)
return
# We are going to try to track every event pair that comes
# through, but that's not as easy as it seems since we don't
# have a unique key for each request (request_id won't work
# since the call could come multiple times via a retry loop).
# So, we're just going to look for Timing objects that have
# start_raw but no end_raw. This could give incorrect data
# when/if we get two overlapping foo.start calls (which
# *shouldn't* happen).
start = step == 'start'
timing = None
timings = STACKDB.find_timings(name=name, lifecycle=lifecycle)
if not start:
for t in timings:
try:
if t.end_raw == None and t.start_raw != None:
timing = t
break
except models.RawData.DoesNotExist:
# Our raw data was removed.
pass
if timing is None:
timing = STACKDB.create_timing(name=name, lifecycle=lifecycle)
if start:
timing.start_raw = raw
timing.start_when = raw.when
# Erase all the other fields which may have been set
# the first time this operation was performed.
# For example, a resize that was done 3 times:
# We'll only record the last one, but track that 3 were done.
timing.end_raw = None
timing.end_when = None
timing.diff_when = None
timing.diff_ms = 0
else:
timing.end_raw = raw
timing.end_when = raw.when
# We could have missed start so watch out ...
if timing.start_when:
timing.diff = timing.end_when - timing.start_when
# Looks like a valid pair ...
update_kpi(timing, raw)
STACKDB.save(timing)
INSTANCE_EVENT = {
'create_start': 'compute.instance.create.start',
'create_end': 'compute.instance.create.end',
'resize_prep_start': 'compute.instance.resize.prep.start',
'resize_prep_end': 'compute.instance.resize.prep.end',
'resize_revert_start': 'compute.instance.resize.revert.start',
'resize_revert_end': 'compute.instance.resize.revert.end',
'resize_finish_end': 'compute.instance.finish_resize.end',
'delete_end': 'compute.instance.delete.end',
'exists': 'compute.instance.exists',
}
def _process_usage_for_new_launch(raw):
notif = json.loads(raw.json)
payload = notif[1]['payload']
values = {}
values['instance'] = payload['instance_id']
values['request_id'] = notif[1]['_context_request_id']
(usage, new) = STACKDB.get_or_create_instance_usage(**values)
if raw.event == INSTANCE_EVENT['create_start']:
usage.instance_type_id = payload['instance_type_id']
STACKDB.save(usage)
def _process_usage_for_updates(raw):
notif = json.loads(raw.json)
payload = notif[1]['payload']
instance_id = payload['instance_id']
request_id = notif[1]['_context_request_id']
(usage, new) = STACKDB.get_or_create_instance_usage(instance=instance_id,
request_id=request_id)
if raw.event in [INSTANCE_EVENT['create_end'],
INSTANCE_EVENT['resize_finish_end'],
INSTANCE_EVENT['resize_revert_end']]:
usage.launched_at = str_time_to_unix(payload['launched_at'])
if raw.event == INSTANCE_EVENT['resize_revert_end']:
usage.instance_type_id = payload['instance_type_id']
elif raw.event == INSTANCE_EVENT['resize_prep_end']:
usage.instance_type_id = payload['new_instance_type_id']
STACKDB.save(usage)
def _process_delete(raw):
notif = json.loads(raw.json)
payload = notif[1]['payload']
instance_id = payload['instance_id']
launched_at = str_time_to_unix(payload['launched_at'])
deleted_at = str_time_to_unix(payload['deleted_at'])
values = {
'instance': instance_id,
'launched_at': launched_at,
'deleted_at': deleted_at,
'raw': raw
}
delete = STACKDB.create_instance_delete(**values)
STACKDB.save(delete)
def _process_exists(raw):
notif = json.loads(raw.json)
payload = notif[1]['payload']
instance_id = payload['instance_id']
launched_at = str_time_to_unix(payload['launched_at'])
launched_range = (launched_at, launched_at+1)
usage = STACKDB.get_instance_usage(instance=instance_id,
launched_at__range=launched_range)
delete = STACKDB.get_instance_delete(instance=instance_id,
launched_at__range=launched_range)
values = {}
values['message_id'] = notif[1]['message_id']
values['instance'] = instance_id
values['launched_at'] = launched_at
values['instance_type_id'] = payload['instance_type_id']
if usage:
values['usage'] = usage
if delete:
values['delete'] = delete
values['raw'] = raw
deleted_at = payload.get('deleted_at')
if deleted_at and deleted_at != '':
deleted_at = str_time_to_unix(deleted_at)
values['deleted_at'] = deleted_at
exists = STACKDB.create_instance_exists(**values)
STACKDB.save(exists)
USAGE_PROCESS_MAPPING = {
INSTANCE_EVENT['create_start']: _process_usage_for_new_launch,
INSTANCE_EVENT['resize_prep_start']: _process_usage_for_new_launch,
INSTANCE_EVENT['resize_revert_start']: _process_usage_for_new_launch,
INSTANCE_EVENT['create_end']: _process_usage_for_updates,
INSTANCE_EVENT['resize_prep_end']: _process_usage_for_updates,
INSTANCE_EVENT['resize_finish_end']: _process_usage_for_updates,
INSTANCE_EVENT['resize_revert_end']: _process_usage_for_updates,
INSTANCE_EVENT['delete_end']: _process_delete,
INSTANCE_EVENT['exists']: _process_exists,
}
def aggregate_usage(raw):
if not raw.instance:
return
if raw.event in USAGE_PROCESS_MAPPING:
USAGE_PROCESS_MAPPING[raw.event](raw)
def str_time_to_unix(when):
if 'T' in when:
try:
# Old way of doing it
when = datetime.datetime.strptime(when, "%Y-%m-%dT%H:%M:%S.%f")
except ValueError:
try:
# Old way of doing it, no millis
when = datetime.datetime.strptime(when, "%Y-%m-%dT%H:%M:%S")
except Exception, e:
print "BAD DATE: ", e
else:
try:
when = datetime.datetime.strptime(when, "%Y-%m-%d %H:%M:%S.%f")
except ValueError:
try:
when = datetime.datetime.strptime(when, "%Y-%m-%d %H:%M:%S")
except Exception, e:
print "BAD DATE: ", e
return dt.dt_to_decimal(when)
def process_raw_data(deployment, args, json_args):
"""This is called directly by the worker to add the event to the db."""
db.reset_queries()
routing_key, body = args
record = None
handler = HANDLERS.get(routing_key, None)
if handler:
values = handler(routing_key, body)
if not values:
return record
values['deployment'] = deployment
try:
when = body['timestamp']
except KeyError:
when = body['_context_timestamp'] # Old way of doing it
try:
try:
when = datetime.datetime.strptime(when, "%Y-%m-%d %H:%M:%S.%f")
except ValueError:
# Old way of doing it
when = datetime.datetime.strptime(when, "%Y-%m-%dT%H:%M:%S.%f")
except Exception, e:
pass
values['when'] = dt.dt_to_decimal(when)
values['routing_key'] = routing_key
values['json'] = json_args
record = STACKDB.create_rawdata(**values)
STACKDB.save(record)
aggregate_lifecycle(record)
aggregate_usage(record)
return record
def _post_process_raw_data(rows, highlight=None):
for row in rows:
if "error" in row.routing_key:
row.is_error = True
if highlight and row.id == int(highlight):
row.highlight = True
row.fwhen = dt.dt_from_decimal(row.when)
def _default_context(request, deployment_id=0):
deployment = None
if 'deployment' in request.session:
d = request.session['deployment']
if d.id == deployment_id:
deployment = d
if not deployment and deployment_id:
try:
deployment = models.Deployment.objects.get(id=deployment_id)
request.session['deployment'] = deployment
except models.Deployment.DoesNotExist:
pass
context = dict(utc=datetime.datetime.utcnow(),
deployment=deployment,
deployment_id=deployment_id)
return context
def welcome(request):
deployments = models.Deployment.objects.all().order_by('name')
context = _default_context(request)
context['deployments'] = deployments
return render_to_response('welcome.html', context)
def home(request, deployment_id):
context = _default_context(request, deployment_id)
return render_to_response('index.html', context)
def details(request, deployment_id, column, row_id):
deployment_id = int(deployment_id)
c = _default_context(request, deployment_id)
row = models.RawData.objects.get(pk=row_id)
value = getattr(row, column)
rows = models.RawData.objects.select_related()
if deployment_id:
rows = rows.filter(deployment=deployment_id)
if column != 'when':
rows = rows.filter(**{column:value})
else:
when = dt.dt_from_decimal(value)
from_time = when - datetime.timedelta(minutes=1)
to_time = when + datetime.timedelta(minutes=1)
from_time_dec = dt.dt_to_decimal(from_time)
to_time_dec = dt.dt_to_decimal(to_time)
rows = rows.filter(when__range=(from_time_dec, to_time_dec))
rows = rows.order_by('-when')[:200]
_post_process_raw_data(rows, highlight=row_id)
c['rows'] = rows
c['allow_expansion'] = True
c['show_absolute_time'] = True
return render_to_response('rows.html', c)
def expand(request, deployment_id, row_id):
c = _default_context(request, deployment_id)
row = models.RawData.objects.get(pk=row_id)
payload = json.loads(row.json)
pp = pprint.PrettyPrinter()
c['payload'] = pp.pformat(payload)
return render_to_response('expand.html', c)
def latest_raw(request, deployment_id):
"""This is the 2sec ticker that updates the Recent Activity box."""
deployment_id = int(deployment_id)
c = _default_context(request, deployment_id)
then = datetime.datetime.utcnow() - datetime.timedelta(hours=1)
thend = dt.dt_to_decimal(then)
query = models.RawData.objects.select_related().filter(when__gt=thend)
if deployment_id > 0:
query = query.filter(deployment=deployment_id)
rows = query.order_by('-when')[:20]
_post_process_raw_data(rows)
c['rows'] = rows
return render_to_response('host_status.html', c)
def search(request, deployment_id):
c = _default_context(request, deployment_id)
column = request.POST.get('field', None)
value = request.POST.get('value', None)
rows = None
if column != None and value != None:
rows = models.RawData.objects.select_related()
if deployment_id:
row = rows.filter(deployment=deployment_id)
rows = rows.filter(**{column:value}). \
order_by('-when')[:22]
_post_process_raw_data(rows)
c['rows'] = rows
c['allow_expansion'] = True
c['show_absolute_time'] = True
return render_to_response('rows.html', c)