A summary and reflection on the server shelves

Due to the company's recent increase in business volume, the progress of various projects, so that the original saturated server resources become tense. Bloggers are very anxious ah, so he did a series of resource usage statistics and analysis, and then submitted a procurement plan.

Title "A summary of the server shelves and reflection," The server procurement plan is finally approved. Bloggers are happy to wait for the arrival of this batch of servers, and design the relevant shelves and automation solutions, and finally carried out specific implementation operations. Although there are some problems in the implementation process, but are not affect the main process of small problems, there are many in the process of ignoring the points and not designed to the point, these are worthy of hindsight and reflection. After all, this is only a one-time expansion, and this situation will happen many times later. For the future better, for the motherland More prosperous 650) this.width=650; "src=" Http://img.baidu.com/hi/jx2/j_0013.gif "alt=" J_0013.gif "/> Write a blog post on the server shelves of things!

Assi Bar, no nonsense, the following into the article to the point.

Preparatory work

Before the arrival of the server, bloggers in the busy current server resources usage statistics and analysis, write the server procurement plan, procurement issues and the computer room new cabinet contract signed.

Usage statistics and analysis aspects of server resources

Because the company this card is not very strict, so we collected the current server CPU, memory and disk related data indicators to reflect the current server resources are more nervous ... It's too tight to do it ...

Server procurement planning aspects

First, the current architecture to adjust the local and expected resources to use the growth of the estimate to calculate the amount of resources required for this procurement, and then based on the amount of resources converted into the relevant server configuration, the end of the server hardware gap.

Procurement aspects

as the size of the company is not very large, but also an internet company, there is no way to go public bidding. Mainly for the company's several common suppliers to make the relevant inquiry, and then against the structure of the comparison, ultimately, of course, the choice of a cost-effective (must be the cheapest) one.

Computer room aspects

The contact person of the computer room was notified in advance to discuss the demand of the new cabinet. Because it is old customers, in terms of price also have nothing good to talk about. Walk the process, the contract signed, the computer room is ahead of the new cabinet ready. Old cabinet to the new cabinet between the network cable also in advance to explain the room for our laying finished.

After the server arrives, due to the lack of both on-line and offline environment resources. Therefore, the server is divided into two batches, a batch to the online room, a batch of the company's internal computer room. To the online room of this batch, to advance and the machine room confessed to sign and shelves operation. The computer room will help us record the SN (Serial number) number and MAC address of the server, which is normally affixed to the server backplane.

Because this want to put the server out of the process, so that the entire process as far as possible automation, so I and a colleague went to the computer room on-site assistance shelves. The egg hurts early, went to the server before delivery are still not demolished ... To assist the computer room personnel to remove the server, to get the room inside the human flesh put server, human registration SN and Mac for the CMDB storage operations.

Start implementation

The following is the beginning of the automatic installation operation!

I first deployed a set of kickstart on a single server (using cobbler is also possible, cobbler just a layer of kickstart). Since the server on this shelf is a batch of virtualized servers and a batch of storage servers, I have written 2 KS files for both cases.

New server to think of the PXE machine this step, first do the following several steps:

1. Press F2 to enter system Setup

2. Select the integrated RAID Controller xxxx Configuration utility tab under Device setting to make the appropriate level of RAID array according to your server type

3, under the Idrac setting (that is, the Dell Remote Control Settings), set up the remote control IP and the associated NIC interface, etc.

4, in the system BIOS under the default server boot sequence. The default boot order of the server is important, and it must be set to boot from the hard disk, preventing PXE from installing repeatedly from PXE after the system restart.

Once the above steps are set up, we can set up the server via Dell's Remote Control page (the Idrac WEB) to enter PXE boot only on the next reboot.

In this process of automatic installation, I put the server installed, system initialization, application initialization, virtualization deployment, such as unpacking into two major steps.

The first big step: Server Installed

In this step, mainly through the KS file settings to automatically answer the server installation process of various settings, such as root password, time zone, partition situation and so on.

In this step, one of the key points is how to make sure that the hostname and IP of the server you are loading are pre-set.

This is how I do it, by matching the SN number of the server with the default hostname and IP address, and then entering into the CMDB. In the post-installation script of KS, I put a command by getting the SN number of the system (using the DMIDECODE-T 1 command), and then through the SN number the CMDB server gets the hostname and IP address that corresponds to the current SN number, and then makes further setup operations.

Second largest step: system initialization, application initialization and virtualization deployment

In this step, the process operation of the entire step is accomplished primarily by writing a standalone initialization tool. Since this step occurs after the server system installation operation, we will set the automatic acquisition of the initialization tool in the post-installation script of the system installed. Executes after the server restarts, and completes the entire step operation.

The script mainly contains:

System-level initialization (firewall, SELinux, kernel parameters, account permissions, server trust, and related security settings, etc.)

Initialization of related applications (automatically added to Zabbix monitoring, puppet, and LDAP)

Virtualization initialization (automatic deployment of virtualization, automatic generation of virtual machines by template machine According to the settings in the CMDB)

Throughout the automation process, the above aspects are now taken into account. Through the room provided by the SN and MAC address, the information into the CMDB and then after the server installation through the CMDB to obtain relevant metadata, complete the specified operation, and completely release the enterprise operations. However, I think it's just that the hard part is passed on to the head of the machine room operation, like putting the server in the cabinet, copying the SN and Mac, setting up RAID and Idrac, or need someone to do it. But who let us pay? 650) this.width=650; "src=" Http://img.baidu.com/hi/jx2/j_0059.gif "alt=" j_0059.gif "/> emphasized that I do not have a black computer room operation of the meaning.

Summary of issues

As I talked above, there seems to be no problem. Here is the time to summarize the problem.

The problem is as follows:

1, in the procurement plan, forgot to purchase cable, resulting in the day before the machine room hurried contact procurement

2, to the computer room when the work tools are not with the (personal charging line forgot to bring)

3, the server information in the CMDB, personal reasons less input 2 field information, resulting in 2 servers installed after the host name and IP

4, does not need to install the package to do local Yum source causes the installation process is slow, mainly Zabbix and puppet ...

5. Because of omission. System disk size is 102400G, mistakenly written 1024000, resulting in 2 servers have been re-installed ...

6, in the system initialization, forget to add the SSH port changes, resulting in the server after the installation, another script to run the batch from the job.

7, the KS file was written, did not use the server to do the test, led to run on the Dell server, reported a network card selection problem, finally Google, added the relevant parameters to solve the problem

Assi, it's so sad ... The whole process, the problem encountered is quite a lot. Summary down, no technical difficulties and barriers, the main problem is not comprehensive in the plan, the implementation of the time is not careful enough to lead. Alas, it is time to reflect on a wave ...

At present, the above problems have been amended by me. This batch of servers is still out of the line, and it will be easy to get it done next week.

OK, this article is here, relevant implementation documents, I will be in these two days to write down to share to everyone.

This article is from the "not only Linux" blog, so be sure to keep this source http://nolinux.blog.51cto.com/4824967/1661437

A summary and reflection on the server shelves