What is new in Storage for Kubernetes 1.9

With the Holiday seasons upon us, many in the Kubernetes community have been hard at work preparing the delivery of one of the most significant set of changes brought to the Kubernetes storage subsystem. These changes are being released in alpha with some introducing minor updates while others are designed to fundamentally change the way storage is deployed within Kubernetes. This writeup highlights three of the major changes that have been released in version 1.9 of Kubernetes.

Raw Block Support

There are classes of applications, such as databases, that require direct access to raw block storage without the abstraction of a filesystem for peak performance and consistent IO. Starting with version 1.9, an alpha implementation will allow administrators and users to use existing Kubernetes storage semantics to specify and deploy raw block storage for consumption across a Kubernetes cluster. This feature has been implemented to be backward compatible with existing storage configurations while allowing support for PersistentVolume (PV), PersistentVolumeClaim (PVC), StorageClass, and Pods to provision block devices that can consume directly as raw block storage or alternatively with a filesystem.
Learn more about this feature here.

Additional Plugins with Volume Resize Capabilities


This feature has landed in version 1.9 as an alpha implementation. It is designed to allow an existing bound persistent volume to be resized after it has been provisioned. Resizing a volume has many use cases specifically when additional storage space is needed for a given workload. An administrator can simply edit an existing PVC and update the size of its storage claim. This release introduces several existing volume drivers which were updated to support volume resizes including Ceph and Cinder with additional drivers to show up in future releases.
Learn more about this feature here.

Local Storage with Topology-aware Volume Scheduling


Support for using local devices for persistent storage was introduced back in version 1.7. However, the Kubernetes scheduler did not use storage as a first class resource when determining the capability of a node for pod scheduling. Version 1.9 introduces topology-aware volume scheduling. This means a pod can influence where it is scheduled based on availability of local storage of a given node, rack locality, its geographical region, and so on. During pod scheduling, these volume constraints, along with other resource requests (such as memory, CPU, etc) and affinity rules, are taken into account to land the pod on the proper node if available. This refinement of local storage will make it possible to deploy new classes of applications continuing to make Kubernetes a great platform for stateful workloads.
Learn more about this feature here.

Container Storage Interface (CSI)


After months of discussing the Container Storage Interface (CSI) (here, here, and here), it has finally arrived! Version 1.9 of Kubernetes will bring with it an alpha version of CSI which was implemented in record time (given the Holiday season, this release period is shorter than others in the year). This herculean effort was made possible, partly, by the use of gRPC and its interface definition language as the lingua franca for specifying the API that makes up CSI. All new storage drivers must use CSI as a way to implement storage services for Kubernetes. Storage providers are able to develop their own storage drivers outside of the Kubernetes codebase, creating opportunities for fast-paced innovative storage solutions that have higher release frequencies than Kubernetes core.
Learn more about this feature here.

ScaleIO Updates


The {code} team contributed the native ScaleIO in-tree driver back in March, and has continually made slight improvements on each release. The Kubernetes 1.9 release (and back ported to 1.8.4) introduces the ability to decouple the base-64 encoded Secret for ScaleIO credentials into its own namespace while being referenced by StorageClass/PVC/PV in a default or different namespace. This prevents admins from having unauthorized access to ScaleIO secret information. Learn about this feature here. In addition, the kubelet can now be easily containerized since the dependency of having the drv_cfg binary for querying ScaleIO has been removed. This allows ScaleIO to introduce itself into more K8s distros that embrace complete containerization.
Learn about this feature here.

In Summary

Storage is really heating up in the Kubernetes space and the {code} team is excited to be a part of it. Plan on seeing a lot of momentum behind CSI in 2018 and future blog posts including one next week documenting how to turn on the CSI alpha gate and get started using CSI plugins!

留言

張貼留言

這個網誌中的熱門文章

Json概述以及python對json的相關操作

圖片
什麼是json： JSON(JavaScript Object Notation) 是一種輕量級的數據交換格式。易於人閱讀和編寫。同時也易於機器解析和生成。它基於JavaScript Programming Language, Standard ECMA-262 3rd Edition - December 1999的一個子集。JSON採用完全獨立於語言的文本格式，但是也使用了類似於C語言家族的習慣（包括C, C++, C#, Java, JavaScript, Perl, Python等）。這些特性使JSON成為理想的數據交換語言。 JSON建構於兩種結構： “名稱/值”對的集合（A collection of name/value pairs）。不同的語言中，它被理解為對象（object），紀錄（record），結構（struct），字典（dictionary），哈希表（hash table），有鍵列表（keyed list），或者關聯數組（associative array）。 值的有序列表（An ordered list of values）。在大部分語言中，它被理解為數組（array）。 這些都是常見的數據結構。事實上大部分現代計算機語言都以某種形式支持它們。這使得一種數據格式在同樣基於這些結構的編程語言之間交換成為可能。 jso官方說 ​​明參見：http://json.org/ Python操作json的標準api庫參考：http://docs.python.org/library/json.html 對簡單數據類型的encoding 和decoding： 使用簡單的json.dumps方法對簡單數據類型進行編碼，例如： import json obj = [[ 1 , 2 , 3 ], 123 , 123.123 , 'abc' ,{ 'key1' :( 1 , 2 , 3 ), 'key2' :( 4 , 5 , 6 )}] encodedjson = json.dumps(obj) print repr (obj) print encodedjson 輸出： [[1, 2, 3], 123, 123.123, 'abc', {'ke...
閱讀完整內容

Docker容器日誌查看與清理

1. 問題 docker容器日誌導致主機磁盤空間滿了。 docker logs -f container_name 劈裡啪啦一大堆，很佔用空間，不用的日誌可以清理掉了。  2. 解決方法 2.1 找出Docker容器日誌 在linux上，容器日誌一般存放在 /var/lib/docker/containers/container_id/ 下面，查看各個日誌文件大小的腳本docker_log_size.sh，內容如下： #!/bin/sh echo "======== docker containers logs file size ========" logs=$(find /var/lib/docker/containers/ -name *-json.log) for log in $logs do ls -lh $log done # chmod +x docker_log_size.sh # ./docker_log_size.sh 2.2 清理Docker容器日誌（治標） 如果docker容器正在運行，那麼使用 rm -rf 方式刪除日誌後，通過 df -h 會發現磁盤空間並沒有釋放。 原因是在Linux或者Unix系統中，通過 rm -rf 或者文件管理器刪除文件，將會從文件系統的目錄結構上解除鏈接（unlink）。 如果文件是被打開的（有一個進程正在使用），那麼進程將仍然可以讀取該文件，磁盤空間也一直被佔用。 正確姿勢是 cat /dev/null > *-json.log ，當然你也可以通過 rm -rf 刪除後重啟docker。 接下來，提供一個日誌清理腳本 clean_docker_log.sh ，內容如下： #!/bin/sh echo "======== start clean docker containers logs ========" logs=$(find /var/lib/docker/containers/ -name *-json.log) for log in $logs d...
閱讀完整內容

利用 Keepalived 提供 VIP

Keepalived 是一種基於 VRRP 協定實現的高可靠 Web 服務方案，用於防止單點故障問題。因此一個 Web 服務運作至少會擁有兩台伺服器執行 Keepalived，一台作為 master，一台作為 backup，並提供一個虛擬 IP（VIP），master 會定期發送特定訊息給 backup 伺服器，當 backup 沒收到 master 訊息時，表示 master 已故障，這時候 backup 會接管 VIP，繼續提供服務，來確保服務的高可靠性。 VRRP VRRP（Virtual Router Redundancy Protocol，虛擬路由器備援協定），是一個提供備援路由器來解決單點故障問題的協定，該協定有兩個重要概念： * VRRP 路由器與虛擬路由器：VRRP 路由器是表示運作 VRRP 的路由器，是一個實體裝置，而虛擬路由器是指由 VRRP 建立的邏輯路由器。一組 VRRP 路由器協同運作，並一起構成一台虛擬路由器，該虛擬路由對外提供一個唯一固定的 IP 與 MAC 位址的邏輯路由器。 * 主控制路由器（master）與備援路由器（backup）：主要是在一組 VRRP 中的兩種互斥角色。一個 VRRP 群組中只能擁有一台是 master，但可以有多個 backup 路由器。 VRRP 協定使用選擇策略從路由器群組挑選一台作為 master 來負責 ARP 與轉送 IP 封包，群組中其他路由器則作為 backup 的角色處理等待狀態。當由於某種原因造成 master 故障時，backup 會在幾秒內成為 master 繼續提供服務，該階段不用改變任何 IP 與 MAC 位址。 Keepalived 節點配置 IP ADDRESS ROLE 172.16.1.101 vip 172.16.1.102 master 172.16.1.103 backup 安裝與設定 這 ubuntu 14.04 LTS Server 中已經內建了 Keepalived 可以透過 apt-get 來安裝： $ sudo apt-get install -y keepalived 也可以透過 source code 進行安裝，流程如下： $ sudo apt-get install build-essential libssl-dev $ wge...
閱讀完整內容