userland core files.

During the week I collected several escalations that required analysis of userland core files. In both cases I was supplied the core file and a showrev -p and asked for a quick analysis. So my choices were to either..

  • Point our jumpstart configuration scripts at the showrev-p output and install a machine so that I have all the right libraries in the right place. Machines are an expensive resource and it seems a waste to book and occupy a whole machine just for this.

  • Use the shared library checksums in the core file to extract the correct libraries from our patch database and get dbx to use those. I haven't written the scripts to do this yet so it would take a while.
  • Ask the customer for some more files from his machine. The binary that generated the core and then all the shared libraries listed in the output of pldd core. With those I can build a dbxrc that maps the loads of the libraries to use the customer's ones. Once I have these I can debug the core on any machine where dbx is available.

In solaris 10 coreadm gives you much better control over what parts of the faulting processes address space end up in the core file. The default will include the shared library segments so core files will be bigger but debugging them wont need all the other files.



Comments:

Sorry Tim. I should have finished my scripts for doing the same thing could have saved you some work. See http://blogs.sun.com/roller/page/chrisg/20041102#core_files_and_friday_madness

Posted by Chris Gerhard on April 30, 2005 at 02:57 AM GMT+00:00 #

Post a Comment:
Comments are closed for this entry.
About

timatworkhomeandinbetween

Search

Archives
« July 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
31
   
       
Today
News

No bookmarks in folder

Blogroll

No bookmarks in folder