Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Reimplement self-healing using internal statistics #102

Merged
merged 8 commits into from
Jun 1, 2021
Merged

Reimplement self-healing using internal statistics #102

merged 8 commits into from
Jun 1, 2021

Commits on May 27, 2021

  1. Fix isRevisionPublishedConditionReconciled func 

    Need to use the value of the var here, not the pointer, as the pointer
changes in each iteration of the loop. Add a test to catch this.
    @roivaz
    roivaz committed May 27, 2021
    Configuration menu
    Copy the full SHA
    d7687ea View commit details
    Browse the repository at this point in the history

  2. Do not check nil list around range loop 

    range is already safe to use with a nil list.
    @roivaz
    roivaz committed May 27, 2021
    Configuration menu
    Copy the full SHA
    46b40fd View commit details
    Browse the repository at this point in the history

  3. Implement a mechanism to report stats internally 

    An in-memory key-value store has been added so the xdss server can report
statistics that can the be then consumed from the controllers. The
mechanism to detect failing resources has been reimplemented using this
store in conjunction with the new ability of detecting which pod is actually
reporting an ACK/NACK.
    @roivaz
    roivaz committed May 27, 2021
    Configuration menu
    Copy the full SHA
    24e5b59 View commit details
    Browse the repository at this point in the history

  4. Reimplement self-healing using internal stats 

    The implementation consist in the EnvoyConfigRevision tainting itself
when the internal stats show that the percentage of pods reporting
failures is higher that a given threshold. Right now:

* A pod is considered "in failure" when it has reported at least one
  NACK for a given resource version, for at least one resource type.
* The percentage of pods required to trigger the taint of a revision is
  right now hardcoded to 100%.
    @roivaz
    roivaz committed May 27, 2021
    Configuration menu
    Copy the full SHA
    1b17645 View commit details
    Browse the repository at this point in the history

  5. Implement a backoff strategy for failure retries 

    When a NACK is received by the discovery service, it immediately retries
the the same response. This usually does not help because unless
something changes in the resource, the client is likely to reject the
resource again and again. This leads to a storm of retries that elevates
CPU comsumption in the clients, even triggering HPA if enabled.
This commit implements a backoff strategy for retries to avoid this
situation.
    @roivaz
    roivaz committed May 27, 2021
    Configuration menu
    Copy the full SHA
    86d1b6d View commit details
    Browse the repository at this point in the history

  6. Remove irrelevant log line from secrets controller

    @roivaz
    roivaz committed May 27, 2021
    Configuration menu
    Copy the full SHA
    87eccc1 View commit details
    Browse the repository at this point in the history

  7. Set requeue time to 60 seconds for published ecr

    @roivaz
    roivaz committed May 27, 2021
    Configuration menu
    Copy the full SHA
    4ff2f13 View commit details
    Browse the repository at this point in the history

  8. Fix typos 

    Co-authored-by: Sergio López <41513123+slopezz@users.noreply.github.com>
    @roivaz @slopezz
    roivaz and slopezz committed May 27, 2021
    Configuration menu
    Copy the full SHA
    bdd15f7 View commit details
    Browse the repository at this point in the history