Re: sql server 面试题（5） - 未名空间MITBBS历史存档

国际科技财经博客移民网络热点娱乐民生时事公众号

Redian新闻

>未名空间

>Database - 数据库

Re: sql server 面试题（5）

Re: sql server 面试题（5）# Database - 数据库

z*y2008-10-30 07:10

1 楼

今天刚刚出的incident, 拿出来跟大家讨论一下。
3:00am EST today dba team got alert that one of our main applications failed
in London. It was 8:00am in London and it's peak hour. The DBA in London
is not familiar with clustering so they ask help from headquarter.
Now it's 7:10am in office. As dba, you need get it solved asap. your
director, CFO, CIO,CTO are all stand behind you watching you typing.
The database was in two nodes A/A clustering environment. Before the
incident each node has one sql server instance

m*02008-10-30 07:10

2 楼

it's in production environment, should try to fix the problem first, then
look for answers later, in off-hour maybe.
on the side note, if you don't have enough money, don't use cluster, don't
try to save money by using a/a. cluster is for company who have enough
money to do cluster with a/p.
1) fail back to that server if you can, although it's running on 2nd code,
but performance is bad, not only because memory, they will fight for cpu
and most importantly, they will fight for i/o, which is

【在 z***y 的大作中提到】

: 今天刚刚出的incident, 拿出来跟大家讨论一下。
: 3:00am EST today dba team got alert that one of our main applications failed
: in London. It was 8:00am in London and it's peak hour. The DBA in London
: is not familiar with clustering so they ask help from headquarter.
: Now it's 7:10am in office. As dba, you need get it solved asap. your
: director, CFO, CIO,CTO are all stand behind you watching you typing.
: The database was in two nodes A/A clustering environment. Before the
: incident each node has one sql server instance

S*k2008-10-30 07:10

3 楼

Hahahaha
Here is my thought:
Try to bring the failed node back.
Try to make more available VM in the working node.
If cannot solve in 30 minutes, say sorry to people standing around and
suggest to buy a new server as soon as possible.
Close the door and kick the ass who set up the A/A cluster.

z*y2008-10-30 07:10

4 楼

Good one!
Any configuration setting you need to do down the road? Say it's Q4 and we
are run out of money or resources to do A/P...

t
,
sql

【在 m***0 的大作中提到】

: it's in production environment, should try to fix the problem first, then
: look for answers later, in off-hour maybe.
: on the side note, if you don't have enough money, don't use cluster, don't
: try to save money by using a/a. cluster is for company who have enough
: money to do cluster with a/p.
: 1) fail back to that server if you can, although it's running on 2nd code,
: but performance is bad, not only because memory, they will fight for cpu
: and most importantly, they will fight for i/o, which is

m*02008-10-30 07:10

5 楼

if you don't have money now, buy more memory then, or set up a WS server at
least, and have log shipping running. these two things won't cost you that
much, but you have more backup options now.

we

【在 z***y 的大作中提到】

: Good one!
: Any configuration setting you need to do down the road? Say it's Q4 and we
: are run out of money or resources to do A/P...
:
: t
: ,
: sql

z*y2008-10-30 07:10

6 楼

好！
在实际工作中要学会根据情况做折中。 best practice 是一回事，valid practice又
是一回事。实际中，大家都是能省就省。越是大公司，越抠门。 Q4每个部门的预算
都花了差不离了.木得钱乐！
前面有人提到reboot then fail back, 这是重点。
以后呢：
第一要分析问什么会fail over.
第二要调整sql server 的 max memory settings.
第三要预算买内存，或者买设备。有的时候内存比设备还贵。比如HP的ＤＬ６ｘｘ，
２４ＧRam 要20k。必要时要拽上几个PM一起帮你make a case.
mjd回答的很完整，个人看法呵。

p*l2008-10-30 07:10

7 楼

Good solution.
I usually say "You get what you paid for.".

t
,
sql

【在 m***0 的大作中提到】

w*e2008-10-30 07:10

8 楼

我们公司的A/A好着呢, 没什么问题.....
Last week, two instance ran on one node without any problem.....
倒是这Monday, 分开run出了事, 然后reboot 再fail back.....又没事了....
我觉得case by case吧

t
,
sql

【在 m***0 的大作中提到】