Has anyone else been experiencing severe networking issues after installing Istio onto a cluster? We are seeing helm deploys (of other services) hang and fail as well as the kube dashboard runs extremely slow. Kubectl commands still are quick (sub 200ms via -v=6).
This is after a vanilla Istio 0.8 install onto a fresh AKS (w/ RBAC enabled), have attempted various regions (East/West/Central US) and they all exhibit the same issue.
It's currently unusable for us, but maybe I'm just missing something, I haven't found any tickets about the same issue? I believe this began happening about a week ago (6/27) which is around the time Microsoft made some changes to their AKS clusters, but we started using RBAC at the same time so it could be any combination of the two.
Experiencing the same issue with istio v1.0, I think it's because of the continuously crashing galley instance in v1.0, but I did not experience such slowness with istio v0.8. Did you come up with a solution? For me it started to slow down when I deployed all my the applications.
The galley issue in v1.0 is solved by the way, it will be merged into the 1.0 release hopefully very soon. (https://github.com/istio/istio/pull/7605)
Related
I currently have a deployment of AI platform-pipelines v0.2.5 running. I saw that 8 days ago, a new version v0.5.1 got added to the container registry. There have been a lot of changes, fixes, etc.. between these versions and I would like to update my current deployment. Is there an easy way of doing so, without losing my experiments, pipeline runs, etc..?
Can somebody please help me? I configure three masters following the guide of digital ocean and I'm trying to access the Mesos and Marathon interface using my host only adapter address, but it just says the site cant be reached and refused to connect
Thanks for asking your question. First of all, the document you are referencing from Digital Ocean is from 2014 and while DC/OS can run on Ubuntu it is not a supported operating system for this product. There are also concerning suggestions in this article that I would avoid (such as only have a 2 Mesos Master quorum which it looks like you already noticed). Lastly, the company itself is called Mesosphere and the product is called DC/OS :)
With all that out of the way, since 2014 much has changed in DC/OS and the Digital Ocean document is obsolete. You no longer have to manually configure Zookeeper, cluster quorum size, Marathon, Mesos, edit host files on agents, or any of the other items this document references for a production cluster. All of that is taken care of in a YAML configuration file called config.yaml if you attempt an "advanced" install.
You should have more success with connecting to the UIs (DC/OS, Marathon, and Mesos) by attempting an installation method that is up-to-date, on a tested and supported OS, and using the latest versions of DC/OS provided by Mesosphere. This will remove the difficulties you are seeing from the obsolete documentation.
https://docs.mesosphere.com/1.11/installing/oss/
https://docs.mesosphere.com/1.11/installing/oss/custom/system-requirements/
https://docs.mesosphere.com/1.11/installing/oss/custom/configuration/configuration-parameters/
https://docs.mesosphere.com/version-policy/
Hope this helps and you see success with DC/OS!
***Edit, if you want to avoid DC/OS, use the Mesos advanced course https://open.mesosphere.com/advanced-course/
I'm trying to trace services in Openstack Mitaka using osprofiler, but i'm having some issues. It seems it's not possible to trace nova service in Mitaka using osprofiler (correct me if i'm wrong). So i was thinking of using Zipkin.
Can anyone tell me if Zipkin integrates with openstack mitaka?
OpenStack does not have Zipkin as an inbuilt tracer. Hence OSProfiler was adopted as a standard project for tracing in OpenStack.
As far as i can see from the documentation, Nova should have OSProfiler support for Mitaka. Although i have not used OSProfiler with Mitaka, I have worked with OSProfiler with Newton and subsequent releases.
You can post the issue that you are facing so that it will be easier to debug.
I installed Cloudfoundry (approximately version v220) on OpenStack and I want to work with private Docker images on Cloudfoundry.
I would like to run docker registry/repository (Doc|Github) server on Cloudfoundry.
I found tutorials on how to install it directly on a machine/VM (1|2|3).
Is there something to be said against running it on Cloudfoundry?
How do I install it?
Is Diego or something like that already providing the registry/repository service?
I thought Diego was part of Cloudfoundry but reading the CF release notes it looks like I have to install Diego separately is that correct (see "Recommended Diego Version")?
It is possible to run private Docker images on Cloudfoundry and there is a CF-specific registry you can use. In order to do that, there are a number of extra steps that you will need to undertake.
To answer your last question first, we must tease apart what exactly is meant by "Diego is a part of Cloud Foundry". Cloud Foundry is deployed using BOSH, which among other things has a concept of a release. A release is in essence a versioned collection of source code, configuration, dependencies, etc. that your system needs to run. I would recommend reading the BOSH docs to gain more of an understanding as to exactly what BOSH is.
Historically, Cloud Foundry has been made up of a single BOSH release, cf-release, but that is no longer the case. Diego itself is deployed as a separate release, diego-release, and that is what is being referred to in the cf-release release notes. To ensure compatibility, each release of cf-release publishes which release of diego-release is being run alongside.
Diego does support an internal docker registry that can run private docker images, but in order to do so, you must deploy another BOSH release and configure it correctly. That bosh release is the diego-docker-cache-release, the README should hopefully help in getting you started. This cf-dev post by the current Diego PM might also be helpful in setting it up. If you run into any problems or issues, I would recommend posting to the cf-dev mailing lists as the CF community and developers maintain a closer watch on that communication channel.
I'm redoing a badly built web application that my company uses in python/django (after deciding it was the best tool for the job).
I don't have much time to spend on development, which means I have even less time to get it deployed, and since its resource intensive and will be used by a lot of people concurrently, I'd like to be able to take advantage of all the tools that AWS offers, such as RDS, ElastiCache, CloudWatch, and especially any auto scaling tools.
I've seen Heroku and liked it, but I would prefer to use AWS, and the price seems quite high.
I don't mind getting my hands dirty as long as it doesn't take half the development time setting up deployment.
I'm looking for something we can use, whether it be a service or AMI so that we can deploy automatically from our repository, without spending days configuring it and figuring out how to get it working, and without drastically increasing the price to host our app.
As you want something quick and simple, maybe consider RightScale's ServerTemplates to get you up and running quickly. RightScale have a free developer account. There are a few Django ServerTemplates and they are all priced for "All Users", so they'll work with the free developer account.
That will get you a base application stack quickly.
Next, I'd look into using fabric (similar to capistrano) and/or github post-commit hooks to automate deployment of your application.
Once you're happy with that and have more time on your hands you could look at adding all the other stuff you want to use (ElastiCache, etc).
Heroku runs on AWS: http://devcenter.heroku.com/articles/external-services
So, you can use AWS services from Heroku as any EC2 instance. If really wanting it, set Heroku for hard-to-setup services and some little AWS EC2 instance for I-do-myself services.
To automate the deployment you can use a 3rd party tool like capistrano or http://nudow.com. Capistrano will do a lot of the deployment but you have to host it yourself and you have to do the deployment in a specific way for it to work correctly (such as using the same keys everywhere, etc). Nudow.com is easier to setup and is hosted. It will deploy to your existing infrastructure and will do stuff like versioning. Also it has a lot of tools to do things like minimizing javascript/css and uploading to cloudfront.