Container Engine

  1. Support notification for job while job is fail

    Kubernetes Job might run failed. I think it could add a feature to send notification while the job is failed

    3 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
  2. Improve eviction to enable debugging and/or soft eviction thresholds

    Currently my 1.9 cluster evicts pods due to memory pressure and I'm unable to confirm why. Stackdriver Metrics don't seem to capture it (perhaps due to the 60s sampling interval).

    Currently the kubelet is configured for 'hard' eviction:
    --eviction-hard=memory.available<100Mi,nodefs.available<10%,nodefs.inodesFree<5%

    It'd be nice if GKE would have a way to debug eviction - e.g. dumping pod memory usage to the logs, writing a memory dump, etc.. or at least allow for soft eviction so that this can be done by some custom tool.

    4 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
  3. Allow customisation / reduction of system resource reservations

    Currently GKE is extremely inefficient for small clusters (dev, test, early start-up work, hackathons, etc). For example a node with 4GB of RAM only has 3GB marked as "allocatable". In addition system pods have high resource reservations (e.g. kube-dns takes 260 mCPU & ~100 MB memory, calico-typha 200 mCPU) which no doubt makes a lot of sense for many cases, but not so much for low-load environments.. Customisation of these would be great. I've tried editing resource yaml directly but changes get reverted.

    12 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →
  4. Monitor the number of nodes in a node pool via Stack Driver

    There does not seem to be a metric in Stack Driver for how many nodes a node pool has.
    With autoscaling enabled, this would be an interesting metric to show in dashboards.

    1 vote
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
  5. Support changing scopes on GKE cluster without rebuilding

    I would like to be able to change the scopes of a GKE cluster without downtime from the console or command line.

    There doesn't currently appear to be a way to alter the scopes of a GKE cluster without downtime, other than doing a bunch of complex work involving disabling autoscaling, cordoning, draining etc.

    Currently, it is possible to change the scopes of an instance inside a cluster without taking it down. However, if the node is killed, it will be replaced by one that has the main cluster scopes.

    11 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
  6. Automatically reassign the reserved IPs to nodes after the cluster node upgrades

    After each cluster node upgrade I need to not forget to manually reassign the static IPs to the new nodes. Would be great to do it automatically

    3 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
  7. Hybrid Public/Private Clusters

    Given the (now-Beta) support for private clusters, it would be great to be able to support hybrid cluster environments, where some node pools only allocate a private IP, and other node pools can allocate a public IP and be accessible directly.

    So far as I can tell, this is sort of possible today, by editing the instances that GKE creates when managing a node pool. I don't see any reason why simply adding a public IP to an instance should break anything (indeed it doesn't seem to on manual changes), so hopefully this is just as simple as exposing the…

    1 vote
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
  8. Make it so that GCEPersistentDisk or another Persistent Volume type is ReadWriteMany

    Coming from Azure there was AzureFile which supports ReadWriteMany. It seems ridiculous that I should have to roll my own NFS server to get this capability on GKE.

    51 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →
  9. google-fluentd Logging Agent - Debian 9 (stretch) support

    Please add support to Debian 9 (stretch) for google-fluentd logging agent.

    At the moment we can't upgrade our k8s pods:
    https://stackoverflow.com/questions/47838777/google-fluentd-error-on-debian-stretch-google-logging-agent

    Thank you

    3 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →
  10. Create a way to update scopes on existing container node-pools.

    There is no good way to currently update scopes. There are going to be times where scopes are originally set incorrectly or needs change. Addressing this would be useful.

    1 vote
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
  11. Can we stop, the container engine clusters when not in use and restart it when needed ?

    We have created a cluster for development purpose using n1-standard-8 machines. We generally used it in day time for development purpose. Is there any way we can stop the cluster during night time or on weekends ?

    4 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →
  12. Create Actual Working Documentation on How to Get Jenkins CI/CD working for GKE!

    Create Actual Working Documentation on How to Get Jenkins CI/CD working for GKE! This documentation doesn't have enough troubleshooting information to reliably bring up a working CI/CD pipeline with Jenkins on GKE: https://cloud.google.com/solutions/continuous-delivery-jenkins-container-engine#deployingacanary_release
    The Jenkins setup on GKE relies too much on features that are available in Jenkins only when installed on GKE, it's not generic enough. We should be able to deploy Jenkins anywhere and it should be able to deploy to GKE without requiring a "magic" feature which is only available in Jenkins if installed on GKE.

    1 vote
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
  13. Support IPv6

    Support IPv6 in a way that we can have IPv6 addresses for external services in Kubernetes. Maybe all the way down into the container, but external facing would be something at least. It's 2017 :) Thanks!

    59 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    2 comments  ·  Flag idea as inappropriate…  ·  Admin →
  14. For multi-AZ clusters, failure to create a new node due to no capacity in one AZ should retry on another

    Just ended up in a situation where my cluster was trying to autoscale but couldn't create a new node in zones/asia-east1-c because it didn't have resources available - this caused autoscaling to just give up.

    However, my cluster spans asia-east1-b and asia-east1-c, and by manually creating a new GCE machine in c I was able to make it work.

    It'd be good if this was automatic.

    1 vote
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
  15. Offer a "Storage as a Service" (STaaS) solution for Kubernetes and GKE

    Right now everyone has to build their own storage service for GKE and Kubernetes, which is not only painful but goes against the cloud principle where everything is served as a service.

    It would be awesome if devops guys wouldn't have to build their own PV solutions for Kubernetes PVCs.

    88 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    6 comments  ·  Flag idea as inappropriate…  ·  Admin →
  16. Firewall rules for Kubernetes Master

    It would be interesting and will increase the safety of our cluster if we could add firewall rules to the master for the dashboard and API requests.

    Currently firewall rules only affect computing nodes for containers, not the master.

    41 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →
  17. Prevent Kubernetes API Outage During Master Upgrade

    Right now, the Kubernetes API becomes unavailable while a master upgrade is happening. This could become very problematic as infrastructure-level components become Kubernetes-native (i.e. they pull from that API to configure themselves), as this means that if one of those components were to fail during the master upgrade and need to be restarted, they will be unable to configure themselves. For example, a component that consumes endpoints from the API for service discovery could be in trouble if it were to fail and be restarted during a master restart.

    20 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
  18. Add Context Name Flag For `gcloud container clusters get-credentials`

    Right now, when you run gcloud container clusters get-credentials, gcloud will produce a user spec, a cluster spec, and a context spec in ~/.kube/config, all with long generated names. For the user spec and the cluster spec, that's great: the kubernetes user will mostly be reading these but not typing them, so the important criteria should be uniqueness and that the name make clear where the thing is from. For the context, spec, however, long generated names are a very poor choice: these exist so the user can quickly switch between configuration spaces by typing, for example, kubectl

    27 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
  19. Container Registry Namespacing

    Currently the Container Registry WebUI in the Developer Console only displays Images pushed if they reside at the top level of the storage tree. The proposal it to allow the WebUI to also display images is they are organized in a namespacing like structure.

    Where a push to {project}/web/{image}, {project}/platform/{image}, {project}/ops/{image} properly pushes into the repository the Developer Console displays as if there are not images there.

    5 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →
  20. Container engine should accept valid subnet for container address range

    When provisioning container engine, entering any valid subnet into the container address range causes the build to immediately fail on a regex failure. It appears that the decorative /14 UI element never makes it into the data and cluster.clusteripv4cidr, failing to include the /14.

    Further, the cli command suggested by the UI appears to not accept the --containeripv4cidr parameter in any form, with our without the /14 and seems the --containeripv4cidr argument itself is invalid.

    The 10/8 subnet collides with the far side of a VPN connection and, as we'd already adopted (fortunately) the…

    2 votes
    Vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
← Previous 1
  • Don't see your idea?

Container Engine

Categories

Feedback and Knowledge Base