Binary compatability and application dependicies


I see Oracle are echoing what many customers have been saying about Linux patches in this article over on the register.

"Speaking at the Software 2005 conference in Santa Clara, California, on Tuesday, Phillips told an audience of 1,000 chief executive officers: "[Customers] are bogged down by the dependencies between applications. Add a patch in Linux and five other things break. People ask: 'Why can't you tell me what the dependencies are?'"

This is exactly why binary compatability and interface stability are important. If you can't add patches with the confidence that your production systems will stay up, and that your applications will stay running on them, you can only be nervous. And it is also why Sun invests a huge amount of time into testing patches.

Comments:

How much effort is put into testing third party (e.g., Oracle, Veritas) software when new patches are released? We have been burned several times with patches (high load when starting databases on Solaris 8, and mpstat/vmstat reporting incorrect load averages) breaking existing functionality. Any thoughts on how these get through the pit/QE process? You guys definitely do a solid job with patch releases, and things usually (2.5.1 had some issues) work after a new recommended bundle is applied. - Matty

Posted by Matty on April 27, 2005 at 10:19 PM IST #

Oracle and Veritas are tested quite extensively, I've asked one of the guys in the patch group to blog exactly what they test. <p /> I will need to look into the two patches you mention above before I can comment on them.

Posted by fintanr on April 27, 2005 at 10:57 PM IST #

I think the issue around mpstat etc. was with bug 4978228 introduced in 108528-29. If I remember correctly this was due to a change in a kernel structure that veritas was using. The problem seemed to be with the reporting tools (mpstat,vmstat,uptime) rather than an actual performance problem, io wait was indicated as 100% even though this was not really the case. Our Oracle tests that ran on veritas passed, had the issue caused veritas or oracle to not work then the issue would have been caught at our system test stage and the patch would not have been released. I'll try and expand a bit on the document Fintan mentioned in my blog over the next few days.

Posted by Albert White on April 28, 2005 at 06:44 AM IST #

When is Oracle 10G going to be included in the Solaris 10 patch regression test procedure? John groenveld@acm.org

Posted by John Groenveld on June 07, 2005 at 07:57 PM IST #

Post a Comment:
  • HTML Syntax: NOT allowed
About

fintanr

Search

Archives
« April 2014
MonTueWedThuFriSatSun
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
    
       
Today