在前一期山寨交流会上：桂新说他会将良好的构架和一套完整的监控系统作为两个非常重要的基础来抓，良好的架构一步到位这点很难做到（对于发展速度并不确定的小公司来说，架构过大也是一个成本上的风险），但对于后者：一套完整监控系统我是非常认同的。对于目前普遍缺乏测试的Web应用开发过程来说，监控几乎就是测试很重要的一部分；而系统监控本身，也可以帮助及时发现很多问题并量化优化的效果。而系统的扩展和调优的大部分技巧都可以从《构建可扩展的Web站点》(作者是Flickr的开发者)一书中看到。为了减少不必要的调优（盲目的调优是危害很大的），建议从这2章开始看起： 第8章《瓶颈》发现和第10章《统计数据，检测和报警》；有了这些瓶颈检查结果和统计监测/报警系统后，再有针对性的看其他章节做系统调优。

以下是我在博客大巴开发中的一些实践小结：
1 数据库相关
1.1 系统应用层面出问题我一般都习惯性的先去看数据库群的连接数统计：大部分应用最后的瓶颈都在于单表数据量过大后，数据存取慢导致的并发连接数过多的问题，
1.2 解决这个问题目前的主要策略就是分片(share nothing)，单个数据库连接数超过100，就想办法拆分并用多个daemon提供服务，数据的拆分也降低了单表的数据量；
1.3 对于需要全局访问的数据，通过另外冗余存储一份专门用于全局对比的操作用（会增加数据一致性开发量）；
1.4 slowquery一般是遇到问题才去看的，其实binlog转换成文本格式后，也可以用mysqldumpslow统计的，slowquery就是执行时间超过一定阈值（缺省是1秒）的binlog，更全面查询统计，则需要打开数据库的query log，binlog其实就是没有只读操作的querylog；

2 日志相关
2.1 对于应用来说：容忍过多的非致命错误容易让错误日志变得非常难读，从而对发现很多致命错误造成麻烦；在各种系统开发间隙，统计各种错误日志也是我非常乐于去做的一件事儿，从warning faital级的PHP错误信息到404错误日志；报警后的急救车固然重要，日常的体检及时发现潜在风险较高的漏洞或缺陷并修正都会在系统真正发生报警的时候为成功抢修节约大量时间；
2.2 经常出问题的环节要有错误日志，并且格式设计的比较便于sort和grep也是很重要的一方面；

3 需求削减：
3.1 并非所有用户都需要被满足： 发现spammer和盗链者，并设法降低资源流失的速度；
3.2 通过日志统计发现用户使用较少但对系统负载很高的功能；

此外：一个管理方面问题： 开发过程本身是日志是最少的，怎么解决？

For those of you who have been downloading the nightly builds or contributing code to 2.7, you’ve noticed how quickly features are being added, small layout changes are gradually being implemented, and the application is morphing before your very eyes. For the most part, the response has been extremely positive, but even the people who love 2.7 have been wondering what it’s going to end up looking like. Though 2.7 is still a work in progress, we’ve put together a set of wireframes to illustrate how we think it will all turn out, so you can take a look under the hood of the design process, so to speak.

The PDF attached to this post outlines the navigation model, header elements, and important screens such as the dashboard, the new post screen, and list screens for posts, comments, and media.

Some things to bear in mind if you’re not used to looking at wireframes:

1. These are a guide, not a dictate. Changes may be made by developers and designers as needed for technical, aesthetic and/or usability reasons. When you have a team of superfast developers like we do, sometimes wireframes can become out of date quickly. In the two hours since these wireframes were approved, for example, already there are a few things that have moved and a menu change or two. Tweaks will continue to be made over the next week or two before freeze. This is Alpha software, not Beta, and it’s not static. That’s part of what makes it exciting, that every time it’s updated there’s something new.

2. These are all black/grey/white. That’s because we have a designer hard at work on visual styles for the new admin panel, including color palette, fonts, graphic elements, etc. When we have a new look to show off, we will. For now, the wireframes are “lookless” on purpose.

3. Not every screen is wireframed. We focused on creating wireframes for those screens that are undergoing the most change. For screens retaining largely the same functionality and layout, we have not included wireframes. In some cases, we’ll be updating screens but haven’t decided how to do it yet, so those aren’t included either.

4. Some elements apply directly to wordpress.com or wordpress.org, so don’t be alarmed if you see something that doesn’t seem to apply (like multiple dashboards).

One of the things I love best about WordPress is the vibrant community full of talented developers and designers who care about the application and want it to be the best it can be. Despite the overwhelmingly positive response we’ve gotten when showing 2.7 at WordCamps and from the majority of the community, there will always be people who would prefer it to be structured another way, which is why we love plugins! The decisions that went into 2.7 were based on a combination of usability testing results from 2.5 and Crazyhorse (both including laser eye tracking, official report to be released soon, but slides from WordCamp SF available in meantime), community feedback, personal and professional opinions, and some thinking about where the next couple of versions will be going in terms of new features, so that we will have a design that scales to accommodate some the features we hope to incorporate in the future.

So, I hope you enjoy getting an inside look at how we’ve been organizing our thoughts around 2.7, and that when the community feedback starts flowing everyone remembers that we all want the same thing: the best WordPress possible.

Although we haven’t announced any new members in a while, the Percona team has continued to grow steadily behind the scenes. Our hiring policy is to have a few months’ provisional period to ensure the absolute highest quality of service and consistency of results provided to our clients. Today I’m proud to officially welcome Ewen Fortune and Piotr Biel, who have been working with us for quite a few months.

Ewen Fortune joins us from Spain, where he lives with his family. Ewen is actually a native English speaker, but is fluent in written and spoken Spanish, of course. Ewen has been a consultant and contractor before joining Percona, and has many years of experience in Oracle, MySQL, Unix systems administration, programming, and networking. Ewen is an Oracle Database 10g Administrator Certified Professional (OCP), a Certified MySQL 5.0 DBA (CMDBA), and a Certified MySQL 5.1 Cluster DBA (CMCDBA). Like all of our consultants, he also has some hard-to-find skills, such as experience with CFEngine. Ewen has been working with our largest clients, helping them make their data centers more reliable, higher performance, and easier to manage. Other recent accomplishments that come to mind include helping clients move to new data centers and performance audits.

Piotr Biel lives in Poland, quite close to our other Polish team member (Maciek Dobrzanski). In his past as a system architect he helped build two of Poland’s largest internet portals serving billions of pageviews a month. It’s impressive to hear him talk about the techniques he used to manage thousands of servers as easily as most of us manage our email, but it’s even better to watch him do it! Like Ewen, Piotr has extensive knowledge of other technologies such as Unix (IBM AIX Certified Expert, experienced SUN Solaris administrator), enterprise-level networking, advanced virtualization techniques with DLPAR/LPAR and Oracle DBA including RAC clusters, and has been helping our customers move to new data centers, set up replication, eliminate performance and security problems, and much more.

Anything I write about these two gentlemen will surely miss a lot of what they’ve been doing — deep, complex tasks such as patching our MMM high availability toolkit, debugging obscure networking problems, and recovering data from InnoDB tables that are so badly damaged that InnoDB’s own recovery capabilities are overwhelmed. So I’ll stop giving examples of how much they can do!

Ewen and Piotr, a very warm welcome to the Percona team from all of us, and we are so glad to have you with us! We all look forward to the months and years ahead. Go team!

Entry posted by Baron Schwartz | 4 comments

Add to: | | | |