yahoo-eng-team team mailing list archive

[Ying Zuo <yinzuo@xxxxxxxxx>]

Ceilometer dashboard has been moved out of Horizon.

** Changed in: horizon
       Status: Confirmed => Invalid

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to OpenStack Dashboard (Horizon).
https://bugs.launchpad.net/bugs/1243694

Title:
  charting sum statistic is meaningless for cumulative meters

Status in OpenStack Dashboard (Horizon):
  Invalid

Bug description:
  The Value dropdown list on the Stats tab of the admin/Resource Usage
  panel allows the average, minimum, maximum or sum of all meters to be
  charted.

  However it doesn't really make sense to apply the sum aggregate to a
  cumulative meter (or even a gauge meter for that matter).

  As this panel allows the user to graph the sum of samples for any
  meter aggregated over a tenant, depending on the duration chosen, they
  would generally end up with a meaningless sawtooth pattern in the
  graph for cumulative meters.

  The exact shape seen depends on how many samples happen to fall into
  each period timeslice. Horizon arbitrarily divides the duration by 400
  to arrive at a period length, whereas ceilometer allows the metric
  gathering cadence to be configured via the
  /etc/ceilometer/pipeline.yaml.

  Given the typical time horizons that users tend to be interested in
  (last day, last week etc.),  the slices are very unlikely in general
  to match the pipeline interval, unless the user happened to configure
  this to a fairly improbable value (216s, 1512s etc.).

  So we end up seeing a sawtooth pattern that doesn't reflect peaks and
  troughs in actual usage, rather it simply follows from the number of
  samples that happen to fall into each arbitrarily sized period.

  I'll attach two screenshots illustrating this pattern for cumulative
  and gauge meters. In the first case it appears that a monotonically
  increasing value for CPU time is veering up & down (superimposed over
  an upward trend).

  Even more misleading is the second case, which seems to show regular
  churn in the numbers of instances in use by individual tenants
  (whereas in fact the number of instances spun up has been broadly
  static, and the apparent churn is just an illusion caused by a
  mismatch between the requested period and the actual metric cadence).

  Proposed solution: make the Value dropdown context-sensitive, such
  that the aggregate functions provided for the current meter always
  make sense for the meter type. In particular, exclude the Sum
  aggregate for all but delta meters.

To manage notifications about this bug go to:
https://bugs.launchpad.net/horizon/+bug/1243694/+subscriptions