未加星标

Store flaw? Naw! The hyper-converged vendor and the 'bug'-bash

字体大小 | |
[系统(linux) 所属分类 系统(linux) | 发布者 店小二03 | 时间 2016 | 作者 红领巾 ] 0人收藏点击收藏

Store flaw? Naw! The hyper-converged vendor and the 'bug'-bash

CommentA UK consultancy claims it had such a poor experience with Maxta's hyper-converged infrastructure software that it asked for its money back. But the vendor has told a very different story.

Nick Chapman is the managing director and principle consultant at CloudFlux, which was founded in the UK in 2016 to provide consultation on virtualisation, storage, networking, cloud and more. He blogs that his "experience with Maxta was without doubt a spectacular failure."

His says hisMaxta experience started in March 2015 with a remote-access to a proof-of-concept (POC) lab and checking out a hybrid flash/disk set-up with Maxta's SW. He wrote that "read performance was excellent, and whilst the write performance was unspectacular it was good enough to want to see Maxta running on our own kit," even though "write performance was a little unpredictable at times."Maxta said the issue had been fixed in an upcoming release.

August 2015 saw CloudFlux getting a pre-release version of the upcoming Maxta SW to checkout on its own HW. Chapman claims he "uncovered some issues with the new Maxta release, the climax of which was the catastrophic failure of the Maxta cluster, and the loss of all the [test] data."

Maxta "recommended that we re-test the version we had tested back in March on our own hardware. Their assertion being that the old version was 'Rock-Solid' and that the beefier specification of our nodes compared to those in the Maxta POC lab would smooth off the write performance unpredictability."

Chapman "tested the old version on our hardware, and it was very impressive. Read performance was blistering and write performance was much improved." And the write unpredictability seen before wasn't evident. He found that deduplication wasn't actually supported "despite being listed as a key product feature on Maxta's website."

CloudFlux could go ahead without that and, due to a temporary price advantage, decided to go ahead "with it already installed on the hardware that would be the start of our new cluster, we started storage vmotioning VMs over to Maxta."

Just after 200 VMs were moved over and with another one being moved over "the Maxta datastore went into read-only mode without warning. All the VM's went down and Maxta's US-based 24/7 support were not contactable by any means."

They rebooted to bring the Maxta datastore online, and for a couple of weeks it ran, periodically going into read-only mode, while Maxta engineering worked on the problem.

Eventually Chapman said he found that a "LSI storage controller in one of the cluster nodes was periodically resetting due to a bad firmware version." When this happened, he claimed, the Maxta software "was unable to flush the write cache to disk quickly enough in a node failure scenario, causing the filesystem to become read-only."

They updated the controller and increased write disk cache flushing. Maxta said "their installer had failed to setup metadata caching correctly." A new software release was ready and Maxta recommended CloudFlux upgraded to it, which was done, even though dedupe support was still absent.

After 20 VMs had been moved onto the cluster with the new software release "Maxta controller VMs starting randomly rebooting, seemingly due to resource exhaustion. This resulted in IO performance and latency becoming erratic in the extreme. At times it would take the entire Maxta filesystem offline, other times it would be limited to just a subset of VMs. (There were other issues too, albeit less critical.)"

He claims Maxta tried to fix the issues over the next month but failed, while CloudFlux endured "a barely usable storage platform".

Feeling that we had exhausted all our options, in March 2016 we pulled the plug, and asked Maxta for our money back. In our opinion, the product was clearly not fit for purpose, as the myriad of issues we experienced over the previous 6 months had proved. Maxta declined, arguing that all software has bugs, and that they would be fixed in time.

'Grossly misleading account'

Maxta's view of the affair is somewhat different and we reproduce it in full:

The recent blog titled “Maxta A Cautionary Tale of Hyperconverged Storage Gone Wrong” is grossly misleading and factually inaccurate.

The facts are as follows:

The customer contacted Maxta to evaluate our software. Maxta provided access to our then generally available release. The customer tested the software and determined it did not meet their requirements. They then disengaged to pursue other solutions.

Several months later, the customer contacted Maxta requesting to partner with us to deliver a solution that would meet their requirements. We offered to demonstrate an early version of a future release to determine potential fit. The customer stated the future version was exactly what they needed. Maxta stated the version was not ready, however, the customer insisted on testing it.

Maxta furnished a beta version for testing with the explicit caveat that the software had known issues and could not be used in production.

Subsequently, without Maxta awareness the customer purchased a software license via a channel partner. When queried by Maxta about the reason for the unexpected purchase, the customer informed us they had deployed the beta version of the software to production.

At no time did the customer deploy a generally available release of the software.

Despite the surprise the customer had deployed beta software to production, Maxta made all possible efforts to address the issues reported which were primarily related to third-party storage controller firmware.

While the author states they could not reach Maxta support, after checking our records we have not been able to substantiate this. Maxta has global 7x24 support that is reachable via telephone and email.

The customer requested a refund from Maxta beyond the time frame and terms of the license agreement. As the customer did not purchase the license from Maxta the request for refund could not be approved. Maxta referred the customer to their channel partner. To date, Maxta has not been contacted by the reseller regarding this matter.

Maxta says it offers a free licence to MxSP software that may be used to fully test the functionality, performance and suitability for purpose. Interested parties can register for the free software at www.maxta.com .

Comment

There seems to have been a communication breakdown here and the goodwill on both sides that must have been there originally has boiled away in the heat of the dispute.

Trying to get to a reliable point of view we're struck by this Maxta point: "At no time did the customer deploy a generally available release of the software."

Beta software is, well beta software. And dealing via a channel partner with a startup in the US is a situation that could bring attendant communications problems.

Clearly the software did not work and, with it being beta software, that's not that surprising, particularly with a hyper-converged system where the software has to do an awful lot and co-ordinate the operations of many pieces of hardware and code subsystems. There's little room for error there.

The moral would seem to be clear: don't use beta code in production.

本文系统(linux)相关术语:linux系统 鸟哥的linux私房菜 linux命令大全 linux操作系统

分页:12
转载请注明
本文标题:Store flaw? Naw! The hyper-converged vendor and the 'bug'-bash
本站链接:http://www.codesec.net/view/484560.html
分享请点击:


1.凡CodeSecTeam转载的文章,均出自其它媒体或其他官网介绍,目的在于传递更多的信息,并不代表本站赞同其观点和其真实性负责;
2.转载的文章仅代表原创作者观点,与本站无关。其原创性以及文中陈述文字和内容未经本站证实,本站对该文以及其中全部或者部分内容、文字的真实性、完整性、及时性,不作出任何保证或承若;
3.如本站转载稿涉及版权等问题,请作者及时联系本站,我们会及时处理。
登录后可拥有收藏文章、关注作者等权限...
技术大类 技术大类 | 系统(linux) | 评论(0) | 阅读(26)