Skip to content

[JENKINS-1246] Trend graph still shows very long build after delete #13885

Closed

Description

@jenkins-infra-bot

hudson ver 1.176
seamonkey 1.1.7 on winxp

Problem:
my builds typically take about, say 10 minutes.
one build (for whatever problem) takes 10 hours, which in turn makes the
"trend graph" looking nasty.

  • deleted that one long build
  • it's then no longer on the lists visible (neither on the "build history" on
    the left of the page, nor on the list below the graph.
  • BUT: it is still shown on the graph.

Note that I assigned P5 .....


Originally reported by gerhard6, imported from: Trend graph still shows very long build after delete
  • status: Closed
  • priority: Trivial
  • component(s): core
  • resolution: Fixed
  • resolved: 2008-03-04T17:43:31+00:00
  • votes: 0
  • watchers: 0
  • imported: 2025-11-24
Raw content of original issue

hudson ver 1.176 seamonkey 1.1.7 on winxp

Problem: my builds typically take about, say 10 minutes. one build (for whatever problem) takes 10 hours, which in turn makes the "trend graph" looking nasty.

  • deleted that one long build
  • it's then no longer on the lists visible (neither on the "build history" on the left of the page, nor on the list below the graph.
  • BUT: it is still shown on the graph.

Note that I assigned P5 .....

  • environment: Platform: All, OS: All

Activity

  1. jenkins-infra-bot commented on Feb 2, 2008

    @jenkins-infra-bot
    Author
    kohsuke:
    • Original comment link
    • Raw content of original comment:

      On a related topic, Hudson should probably choose the Y-axis value range more wisely, so that those out-of-proportion values wouldn't mess up the entire graph.

    On a related topic, Hudson should probably choose the Y-axis value range more
    wisely, so that those out-of-proportion values wouldn't mess up the entire graph.

  2. jenkins-infra-bot commented on Feb 2, 2008

    @jenkins-infra-bot
    Author
    kohsuke:

    For example, Chebyshev's theorem
    <http://en.wikipedia.org/wiki/Chebyshev%27s_inequality> states that 96% of the
    data would be within 5-stddev range from the average, so maybe we can choose the
    maximum Y-value as "min( maximum data value, average + 5xStdDev)"

  3. jenkins-infra-bot commented on Mar 4, 2008

    @jenkins-infra-bot
    Author
    kohsuke:
    • Original comment link
    • Raw content of original comment:

      Implemented in 1.188.

    Implemented in 1.188.

  4. removed
    bugFor changelog: Minor bug. Will be listed after features
    on Nov 25, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions