Thorsten,

This looks like very time-consuming work. Thanks for putting in the 
effort! See comments below.

Thorsten Frueauf wrote:
> Hi everyone,
> 
> the following is about my current work to manually determine package
> dependencies for the 11 agents, which are part of Colorado-I.
> 
> I already got some positive feedback by Detlef Ulherr for 
> SUNWscPostgreSQL, SUNWsctomcat and SUNWscmys, and from Neil Garhtwaite 
> for SUNWscdhc and SUNWscsmb - thanks a lot!
> 
> I specifically seek review for
> 
> SUNWscapc (HA Apache)
> SUNWscdns (HA DNS)
> SUNWsckrb5 (HA Kerberos)
> SUNWscnfs (HA NFS)
> SUNWscs1as (HA SunOne Appserver / Glassfish)
> 
> since I am not the agent expert for them, I would appreciate if
> the owners could have a look to verify and sanity check with the
> dependencies I found and determined.
> 
> Here is what I did:
> 
> Step 1) Going through all source files of the individual agents, I did
>         record the commands getting invoked by the code, the
>         corresponding packages for those commands, and which agent did
>         really use them, marked with a "*" in the correspinding row.
>         I did also record which files I did analyze.
>         The result is in a StarCalc sheet, available at
> 
> http://opensolaris.org/os/project/colorado/files/Colorado-I-IPS-agent-dependencies-20090210.ods
>  
> 
> 
> Step 2) The StarCalc sheet can get saved as .csv file, where the
>         separator is "," and content of fields is kept in double quotes.
>         The result is available at
> 
> http://opensolaris.org/os/project/colorado/files/Colorado-I-IPS-agent-dependencies-20090210.cvs.txt
>  
> 
> 
> Step 3) Have a script which uses awk to extract the package list by
>         looking at the "*" recorded for each package.
>         This then results into a sorted list of uniq packages.
>         The script I used is available at
> 
> http://opensolaris.org/os/project/colorado/files/map-agent-package-dependency.ksh.txt
>  
> 
> 
> Step 4) Verify the list shown by the script of Step 3) and insert the
>         missing entries into the usr/src/ipsdefs/<package>/depend_static
>         file. I did leave out entries already recorded within depend_auto.
> 
> The result can be reviewd within the following webrev:
> http://cr.opensolaris.org/~frueauf/colorado-1-ips-agent-dependencies/
> 
> Please note the following:
> 
> - I did bump the year to 2009 within the common copyright file. Guess the
>   same needs to be done within the core gate (was only recently done within
>   the pkgdefs area, but not within ipsdefs). Is it ok to have it part of 
> this
>   review/putback? I am also happy to leave it out, if required.
> 

It's fine with me to leave it.

> - I did add the dependency to the group package 
> ha-cluster-framework-minimal
>   to all packages. That way if the user just installs the agent package,
>   it should pull in all required packages in order to configure a working
>   cluster.

Great!

> 
> - The following files contain questions, where I seek input, before I
>   finalize and send out the webrev to ha-clusters-discuss:
> 
>   * SUNWscapc/depend_static
>   * SUNWscdns/depend_static
>   * SUNWsckrb5/depend_static
>   * SUNWscnfs/depend_static
>   * SUNWscs1as/depend_static
> 
>   a) for some I am not sure to record the OpenSolaris package
>      dependency for the application - specifically I don't know if it
>      is a good idea to depend on just one version.
>      I would be fine with nfs and krb5, but I doubt if we want that for
>      dns and apache - although we have hardcoded paths. At least in the
>      case of apache we offer choice, and maybe users do not want both
>      versions.

I agree that it doesn't seem quite right to pull in all available 
versions of the applications. For example, for Apache, I'm likely to 
want Apache 2 only, and wouldn't want an older version cluttering up my 
file system.

On the other hand, the only other option seems to be to depend on 
neither version, and require the user to install the application 
herself. That also doesn't seem quite right.

I guess if I had to make a decision I'd say to include the dependencies. 
  At this point, I'd rather err on the side of pulling in something 
extra rather than making the user to additional work.

> 
>   b) for the non-GDS based agents my auto conversion recorded the 
> dependencies
>      from the pkgdefs common depend file. I would tend to delete those 
> entries,
>      but would like to get feedback how reasonable they are.
> 

IMHO: Get rid of them.

Thanks,
Nick

Reply via email to