v2.23.4 Armory Continuous Deployment Release (Spinnaker™ v1.23.6)

Release notes for Armory Continuous Deployment v2.23.4

2021/02/04 Release Notes

Note: If you’re experiencing production issues after upgrading Spinnaker, rollback to a previous working version and please report issues to http://go.armory.io/support.

Required Halyard or Operator version

To install, upgrade, or configure Armory 2.23.4, use one of the following tools:

  • Armory-extended Halyard 1.10 or later
  • Armory Operator 1.2.1 or later

Security

Armory scans the codebase as we develop and release software. Contact your Armory account representative for information about CVE scans for this release.

Breaking changes

Breaking changes are kept in this list for 3 minor versions from when the change is introduced. For example, a breaking change introduced in 2.21.0 appears in the list up to and including the 2.24.x releases. It would not appear on 2.25.x release notes.

Suffix no longer added to jobs created by Kubernetes Run Job stage

Spinnaker no longer automatically appends a unique suffix to the name of jobs created by the Kubernetes Run Job stage. Prior to this release, if you specified metadata.name: my-job, Spinnaker updates the name to my-job-[random-string] before deploying the job to Kubernetes. As of this release, the job’s name will be passed through to Kubernetes exactly as supplied.

To continue having a random suffix added to the job name, set the metadata.generateName field instead of metadata.name, which causes the Kubernetes API to append a random suffix to the name.

This change is particularly important for users who are using the preconfigured job stage for Kubernetes or are sharing job stages among different pipelines. In these cases, jobs often running concurrently, and it is important that each job have a unique name. In order to retain the previous behavior, manually update your Kubernetes job manifests to use the generateName field.

Previously, this behavior was opt-in.

Impact

As of Armory 2.22, this behavior is the default. Users can still opt out of the new behavior by setting kubernetes.jobs.append-suffix: true in clouddriver-local.yml. This causes Spinnaker to continue to append a suffix to the name of jobs as in prior releases.

The ability to opt out of the new behavior will be removed in Armory 2.23 (OSS 1.23). The above setting will have no effect, and Spinnaker will no longer append a suffix to job names. We recommended that 2.22 users note which jobs are using the old behavior and prepare to remove the setting before upgrading to Armory 2.23 in the future.

Introduced in: Armory 2.22

ManifestForceCacheRefreshTask removed from Orca

When you upgrade to 2.23.0 or later, you might encounter the following error:

2021-01-29 23:57:19.691 ERROR 1 --- [    scheduler-2] c.netflix.spinnaker.q.redis.RedisQueue   : Failed to read message 8f072714f1df6dbf3af93a4f4fe4cae2, requeuing...
com.fasterxml.jackson.databind.JsonMappingException: No task found for 'com.netflix.spinnaker.orca.clouddriver.tasks.manifest.ManifestForceCacheRefreshTask' (through reference chain: com.netflix.spinnaker.orca.q.RunTask["taskType"])

The ManifestForceCacheRefreshTask task is no longer a required task when deploying a manifest. In earlier releases, forcing the cache to refresh was part of the deployment process for manifests. Because of this change, if a task was running or retried before the upgrade, the error shows up in logs as an exception.

Workaround

Before starting, make sure that you have access to the Redis instance that Orca uses.

To resolve this issue, delete the message from the queue:

  1. Verify that there are pipeline execution failure messages that contain ManifestForceCacheRefreshTask:

    Redis

    hgetall orca.task.queue.messages
    

    The command returns information similar to the following:

    1) "93ac65e03399a4cfd3678e1355936ab2"
    2) "{\"kind\":\"runTask\",\"executionType\":\"PIPELINE\",\"executionId\":\"01EVFCCDG3Q2209E0Z1QTNC0FS\",   \"application\":\"armoryhellodeploy\",\"stageId\":\"01EVFCCDG3TJ7AFPYEJT1N8RDJ\",\"taskId\":\"5\",\"taskType\":\"com.netflix.spinnaker.   orca.clouddriver.tasks.manifest.ManifestForceCacheRefreshTask\",\"attributes\":[{\"kind\":\"attempts\",\"attempts\":1}],   \"ackTimeoutMs\":600000}"
    
  2. Delete the message(s):

    Redis

    hdel orca.task.queue.messages 
    

    The command returns information similar to the following:

    93ac65e03399a4cfd3678e1355936ab2
    (integer) 1
    

Zombie Executions

Starting in Spinnaker 2.23.0, ManifestForceCacheRefreshTask was removed, as Kubernetes manifest related stages now do live lookups. While upgrading to Spinnaker 2.23.0 or later, if there is a running pipeline that contains a Kubernetes manifest related stage, it becomes a zombie execution. This causes Orca, Spinnaker’s orchestration service, to fail to complete any Kubernetes manifest related stage in that pipeline.

Workarounds:

To resolve the issue, cancel any zombie executions. For information about how to cancel them, see the Orca Zombie Execution runbook.

Affected versions: 2.23.0 and later

Known issues

Bake failures

The Packer version included with Rosco disregards package overrides that use the -var-file= option. This may cause bakes to fail.

Affected versions: 2.22.2 and later

Lambda UI issue

There is a UI bug related to the caching agent that prevents Lambda functions from being displayed in the UI when there are no other clusters associated with the Application. In other words, in order for the function to show up in “Functions” tab, there needs to be a cluster (such as an AWS ASG/EC2 instance) deployed for that application.

Affected versions: 2.23.0 (1.23.0) - 2.26.2 Fixed version: 2.26.3

Failing health checks when using dynamic accounts

There is a known issue where the health checks for the Clouddriver pod fail when the following conditions are true:

  • Spinnaker is configured to use Spring Cloud Config Dynamic Accounts backed by Vault using a K/V Secrets Engine v2.
  • Spinnaker is configured to use Armory Vault Secrets.

The health check failure prevents Kubernetes from transitioning the Clouddriver pod to a ready and active state, which prevents Kubernetes from passing traffic to the Clouddriver pod.

Workaround

As an alternative to the default HTTP health check, use TCP probe.

If deploying with Halyard, add the following to clouddriver-local.yml:

kubernetes:
  useTcpProbe: true

If you use the Spinnaker Operator, include the following setting:

spec:
  spinnakerConfig:
    service-settings:
      clouddriver:
        kubernetes:
          useTcpProbe: true

Affected versions: 2.23.4, 2.23.5

Pipelines-as-Code fails unexpectedly when updating modules

The container for the Dinghy service that Pipelines-as-Code uses fails when updating pipelines using modules stored in GitHub. The error you encounter references a failure related to GitHub, such as one of the following:

422 Validation Failed [{Resource:CommitComment Field:body Code:custom Message:body is too long (maximum is 65536 characters)}]

or

422 No commit found for SHA: <SHA for a commit> []

This results in only some pipelines in your deployment getting updated when a module gets updated.

Workaround:

  1. Use the arm CLI to render the JSON for your dinghyfiles.
  2. Update pipelines manually using the UI.

Affected versions: 2.22.x, 2.23.x, 2.24.0 Fixed versions: 2.25.0

Fixed issues

  • Fixed an issue where Clouddriver fails to cache images that belong to the first account (alphabetically) for each region.

Highlighted updates

This release includes various improvements to security and performance.

Spinnaker Community Contributions

There have also been numerous enhancements, fixes, and features across all of Spinnaker’s other services. See the Spinnaker v1.23.6 changelog for details.

Detailed updates

Bill Of Materials (BOM)

Here’s the BOM for this version.

Expand
version: 2.23.4
timestamp: "2021-02-03 15:52:32"
services:
    clouddriver:
        commit: 7b2a33c8
        version: 2.23.36
    deck:
        commit: ccf47bbb
        version: 2.23.19
    dinghy:
        commit: 41fde564
        version: 2.23.9
    echo:
        commit: e7ef217c
        version: 2.23.14
    fiat:
        commit: 7124416c
        version: 2.23.11
    front50:
        commit: 95b3ff9c
        version: 2.23.14
    gate:
        commit: dee95e1f
        version: 2.23.13
    igor:
        commit: c932b693
        version: 2.23.13
    kayenta:
        commit: ac7147d0
        version: 2.23.17
    monitoring-daemon:
        version: 2.23.0
    monitoring-third-party:
        version: 2.23.0
    orca:
        commit: fe3d069a
        version: 2.23.21
    rosco:
        commit: 296e82dc
        version: 2.23.18
    terraformer:
        commit: 7710fd96
        version: 2.23.9
dependencies:
    redis:
        version: 2:2.8.4-2
artifactSources:
    dockerRegistry: docker.io/armory

Armory

Armory Gate - 2.23.5…2.23.13

  • chore(build): use armory BOM (#219)

Armory Igor - 2.23.5…2.23.13

  • chore(build): use armory commons bom (#177) (#179)

Armory Deck - 2.23.15…2.23.19

  • chore(deps): bump to new OSS 1.23.6 bom (#728)

Armory Clouddriver - 2.23.28…2.23.36

  • fix(build): Dependencies for Spring cloud config + Vault (#266)
  • fix(build): Remove explicit guava - get from armory-commons (#267)

Armory Fiat - 2.23.5…2.23.11

  • chore(build): use armory commons BOM (#156) (#158)

Armory Front50 - 2.23.6…2.23.14

  • chore(build): use armory commons BOM (#202) (#204)

Armory Kayenta - 2.23.8…2.23.17

  • chore(build): use armory commons BOM (#183) (#185)

Armory Rosco - 2.23.10…2.23.18

  • chore(build): use armory commons bom (#185) (#187)

Dinghy™ - 2.23.8…2.23.9

Armory Echo - 2.23.6…2.23.14

  • fix(build): use armory commons BOM (#279)

Armory Orca - 2.23.12…2.23.21

  • fix(build): use armory commons BOM as platform (#205)

Terraformer™ - 2.23.6…2.23.9


Last modified March 3, 2023: (22c29bf4)