I have been seeing something for quite some time which has confused me considerably for over a year, perhaps one of you can help me understand.
Assumed: rpm queries are against _a_ database. Assumed: database queries against the same database, without changes to the data in the database, will return the same data.
Confusion: then why are some of the summaries reported by rpm different?
Each day I (cron.daily) run the following command rpm -qa \ --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-%{RELEASE}","%{ARCH}","% {SUMMARY}"\n' \ | sort -t" -k3 > ${OUTFILE}
Stuff the resulting ${OUTFILE} in an rcs file.
And some days the rcs file will show deltas such as the following (which was pulled from a rather recent set of flipflops):
--- mach.csv 2011/11/15 10:50:04 +++ mach.csv 2011/11/15 09:22:53
-"CentOS","bash","3.2","3.2-32.el5","i386","The GNU Bourne Again shell (bash) version 3.1." +"CentOS","bash","3.2","3.2-32.el5","i386","The GNU Bourne Again shell (bash) version 3.2"
-"CentOS","compat-db","4.2.52","4.2.52-5.1","i386","The Berkeley DB database library for CentOS 2.1 compatibility." +"CentOS","compat-db","4.2.52","4.2.52-5.1","i386","The Berkeley DB database library for Red Hat Linux 7.x compatibility."
-"CentOS","firefox","3.6.24","3.6.24-3.el5.centos","i386","Mozilla Firefox Web browser." +"CentOS","firefox","3.6.24","3.6.24-3.el5.centos","i386","Mozilla Firefox Web browser"
-"Adobe Systems Inc.","flash-plugin","11.1.102.55","11.1.102.55-release","i386","Adobe Flash Player 7.0" +"Adobe Systems Inc.","flash-plugin","11.1.102.55","11.1.102.55-release","i386","Adobe Flash Player 11.1"
-"CentOS","gdb","7.0.1","7.0.1-37.el5_7.1","i386","A GNU source-level debugger for C, C++, Java and other languages." +"CentOS","gdb","7.0.1","7.0.1-37.el5_7.1","i386","A GNU source-level debugger for C, C++, Java and other languages"
-"CentOS","gettext","0.17","0.17-1.el5","i386","GNU libraries and utilities for producing multi-lingual messages." +"CentOS","gettext","0.17","0.17-1.el5","i386","GNU libraries and utilities for producing multi-lingual messages"
-"CentOS","htmlview","4.0.0","4.0.0-2.el5","noarch","Tools for launching Preferred Applications" +"CentOS","htmlview","4.0.0","4.0.0-2.el5","noarch","Launcher of Preferred Web Browser"
rpm -q --last compat-db htmlview compat-db-4.2.52-5.1 Tue 17 Mar 2009 02:32:59 PM EDT htmlview-4.0.0-2.el5 Tue 17 Mar 2009 02:21:59 PM EDT i.e. these have been there, at this version, since machine install.
poking around in /var/lib/rpm/, with strings and grep, the only file I find "Adobe Flash Player " in is Packages and I only find the "Adobe Flash Player 11.1" variant of the string. Yet at the same time rpm -q \ --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-%{RELEASE}","%{ARCH}","% {SUMMARY}"\n' \ flash-plugin
displays: "Adobe Systems Inc.","flash-plugin","11.1.102.55","11.1.102.55-release","i386","Adobe Flash Player 7.0"
/var/tmp/rpm* is nonexistent.
Note: rpm does seem to be fairly consistent in its inconsistency, the above applications are a small sampling of a larger *set* (out of ~180 packages that show this) that seems to keep flip flopping from one SUMMARY to the other, *usually* in the day or two post a yum update (not sure if reboots also affect it). The summaries always bounce between the same text, i.e., htmlview has always used either "Launcher of Preferred Web Browser" or "Tools for launching Preferred Applications" and bash has always used either "The GNU Bourne Again shell (bash) version 3.1." or "The GNU Bourne Again shell (bash) version 3.2" (and a 3.1 version has never been installed on the system).
RHEL machines show the same flip flops with the same packages and same flopping summary texts.
So where is rpm getting the bad information from????? Can it be cleaned out? What would be lost by cleaning it out?
-----Original Message-----
Denniston, Todd A CIV NAVSURFWARCENDIV Crane wrote: Sent: Tuesday, 15 November, 2011 12:57 To: CentOS@centos.org Subject: [CentOS] How can rpm "%{SUMMARY}" not be consistent?
I have been seeing something for quite some time which has confused me considerably for over a year, perhaps one of you can help me understand.
Assumed: rpm queries are against _a_ database. Assumed: database queries against the same database, without changes to the data in the database, will return the same data.
Confusion: then why are some of the summaries reported by rpm different?
Each day I (cron.daily) run the following command rpm -qa \ --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-%{RELEASE}","%{ARCH}","%{SUMMARY}"\n' \ | sort -t" -k3 > ${OUTFILE}
Stuff the resulting ${OUTFILE} in an rcs file.
And some days the rcs file will show deltas such as the following (which was pulled from a rather recent set of flipflops):
--- mach.csv 2011/11/15 10:50:04 +++ mach.csv 2011/11/15 09:22:53
-"CentOS","bash","3.2","3.2-32.el5","i386","The GNU Bourne Again shell (bash) version 3.1." +"CentOS","bash","3.2","3.2-32.el5","i386","The GNU Bourne Again shell (bash) version 3.2"
<snip>
Using the double quote (") as a delimiter, the third key would be a comma (,) (see below for an example)
"A","B","C","D","E",... 1"2"3"4"5"6"7"8"9"A"B...
And since your sort command is sorting only on the fifth key, all the commas are already in order (unless there is a blank line in there somewhere). The order of the re cords within a sorted group may or may not be guaranteed to change.
Change your sort order from '-k3' to '-k2 -k4 -k6 -k8 -k10 -k12' to sort by Vendor, Name, Version, Version-Release, Arch, Summary and see if that helps.
Cheers!
Simba Engineering
-----Original Message----- From: centos-bounces@centos.org [mailto:centos-bounces@centos.org] On Behalf Of Phoenix, Merka Sent: Tuesday, November 15, 2011 18:48 To: CentOS mailing list Subject: Re: [CentOS] How can rpm "%{SUMMARY}" not be consistent?
-----Original Message-----
Denniston, Todd A CIV NAVSURFWARCENDIV Crane wrote: Sent: Tuesday, 15 November, 2011 12:57 To: CentOS@centos.org Subject: [CentOS] How can rpm "%{SUMMARY}" not be consistent?
<SNIP>
rpm -qa \ --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-
%{RELEASE}","%{ARCH}","%{SUMMARY}"\n' \
| sort -t" -k3 > ${OUTFILE}
Stuff the resulting ${OUTFILE} in an rcs file.
And some days the rcs file will show deltas such as the following
(which
was pulled from a rather recent set of flipflops):
--- mach.csv 2011/11/15 10:50:04 +++ mach.csv 2011/11/15 09:22:53
-"CentOS","bash","3.2","3.2-32.el5","i386","The GNU Bourne Again
shell
(bash) version 3.1." +"CentOS","bash","3.2","3.2-32.el5","i386","The GNU Bourne Again
shell
(bash) version 3.2"
<snip>
Using the double quote (") as a delimiter, the third key would be a comma (,) (see below for an example)
"A","B","C","D","E",... 1"2"3"4"5"6"7"8"9"A"B...
And since your sort command is sorting only on the fifth key, all the commas are already in order (unless there is a blank line in there somewhere). The order of the re cords within a sorted group may or may not be guaranteed to change.
Change your sort order from '-k3' to '-k2 -k4 -k6 -k8 -k10 -k12' to sort by Vendor, Name, Version, Version-Release, Arch, Summary and see if that helps.
Although you may be correct on the need to sort the NAME field using -k4 instead of -k3 ***, the question was not about the sort order, but about the *content* of the SUMMARY fields being different between multiple runs against the same database information for the same package, i.e. notice the difference in the summary fields for the same bash package above.
*** and this seems to explain a different but MUCH less annoying confusion.
Thanks for that bit of help.
On Tue, Nov 15, 2011 at 12:56 PM, Denniston, Todd A CIV NAVSURFWARCENDIV Crane todd.denniston@navy.mil wrote:
I have been seeing something for quite some time which has confused me considerably for over a year, perhaps one of you can help me understand.
Assumed: rpm queries are against _a_ database. Assumed: database queries against the same database, without changes to the data in the database, will return the same data.
Confusion: then why are some of the summaries reported by rpm different?
Each day I (cron.daily) run the following command rpm -qa \ --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-%{RELEASE}","%{ARCH}","% {SUMMARY}"\n' \ | sort -t" -k3 > ${OUTFILE}
Stuff the resulting ${OUTFILE} in an rcs file.
And some days the rcs file will show deltas such as the following (which was pulled from a rather recent set of flipflops):
--- mach.csv 2011/11/15 10:50:04 +++ mach.csv 2011/11/15 09:22:53
-"CentOS","bash","3.2","3.2-32.el5","i386","The GNU Bourne Again shell (bash) version 3.1." +"CentOS","bash","3.2","3.2-32.el5","i386","The GNU Bourne Again shell (bash) version 3.2"
What you are seeing is indeed odd. I see 'version 3.1' but not '3.2' anywhere on the Summary line of bash. What is your kernel by the way? uname -mr ?
Have you cleared yum cache? Not just running a 'yum clean all' but emptying the /var/cache/yum directory ?
Akemi
On Wed, 16 Nov 2011, Akemi Yagi wrote:
What you are seeing is indeed odd. I see 'version 3.1' but not '3.2' anywhere on the Summary line of bash. What is your kernel by the way? uname -mr ?
Have you cleared yum cache? Not just running a 'yum clean all' but emptying the /var/cache/yum directory ?
Why would yum cache have any bearing on what rpm reported?
jh
On Wed, Nov 16, 2011 at 6:56 AM, John Hodrien J.H.Hodrien@leeds.ac.uk wrote:
On Wed, 16 Nov 2011, Akemi Yagi wrote:
What you are seeing is indeed odd. I see 'version 3.1' but not '3.2' anywhere on the Summary line of bash. What is your kernel by the way? uname -mr ?
Have you cleared yum cache? Not just running a 'yum clean all' but emptying the /var/cache/yum directory ?
Why would yum cache have any bearing on what rpm reported?
In my attempts to reproduce what you are seeing, I used 'yum info' a few times for the packages that were not on my systems. But in your case (pure rpm operations) yum cache will not be relevant. By the way I looked at both CentOS 5 and 6 but did not see any inconsistency. And the reason why I asked about the kernel version was because it was not clear which version/release of CentOS you are running. Sorry for the noise. I will shut up now.
Akemi
-----Original Message----- From: centos-bounces@centos.org [mailto:centos-bounces@centos.org] On Behalf Of Akemi Yagi Sent: Wednesday, November 16, 2011 11:20 To: CentOS mailing list Subject: Re: [CentOS] How can rpm "%{SUMMARY}" not be consistent?
On Wed, Nov 16, 2011 at 6:56 AM, John Hodrien
wrote:
On Wed, 16 Nov 2011, Akemi Yagi wrote:
What you are seeing is indeed odd. I see 'version 3.1' but not
'3.2'
anywhere on the Summary line of bash. What is your kernel by the
way?
I would have said you also seeing 'version 3.1.' is one of the very odd things, but then I check the bash rpm in a repo and it has 'version 3.1.' in the 3.2-32.el5 rpm.
uname -mr ?
Have you cleared yum cache? Not just running a 'yum clean all' but emptying the /var/cache/yum directory ?
Why would yum cache have any bearing on what rpm reported?
In my attempts to reproduce what you are seeing, I used 'yum info' a
It takes days/weeks of collecting the data via cron.daily and (I think) having a few updates/installs happen between some of the runs. On the boxes where I see it more, I often run the data collection script immediately following updates. I almost think there is some kind of rpm housekeeping that gets done on a daily basis that could affect it, but I can't figure out what it would be, because the rpm script in cron.daily only dumps data (similar to what mine dumps) to /var/log/rpmpkgs... it does not issue any rpm clean up commands. as I understand anacron, each of the scripts should finish before anacron starts the next, so there should not be any DB contention, between the rpm script and mine, I would think.
At one time (in the mists of history, probably around RHEL 1|2) I thought there was a daily rpm cleanup task, but I can't find it on Cent 5 systems.
few times for the packages that were not on my systems. But in your case (pure rpm operations) yum cache will not be relevant.
That is my thought too. But I don't know where rpm could be getting the different info.
By the way I looked at both CentOS 5 and 6 but did not see any inconsistency.
I am on CentOS 5.
And the reason why I asked about the kernel version was because it was not clear which version/release of CentOS you are running. Sorry for the noise. I will shut up now.
Denniston, Todd A CIV NAVSURFWARCENDIV Crane wrote:
At one time (in the mists of history, probably around RHEL 1|2) I thought there was a daily rpm cleanup task, but I can't find it on Cent 5 systems.
there is something in /etc/rc.sysinit, so it would happen on reboot: $ grep rpm /etc/rc.sysinit rm -f /var/lib/rpm/__db* &> /dev/null
-----Original Message----- From: Denniston, Todd A CIV NAVSURFWARCENDIV Crane Sent: Wednesday, November 16, 2011 16:37 To: CentOS mailing list Subject: RE: [CentOS] How can rpm "%{SUMMARY}" not be consistent?
-----Original Message----- From: centos-bounces@centos.org [mailto:centos-bounces@centos.org]
On
Behalf Of Akemi Yagi Sent: Wednesday, November 16, 2011 11:20 To: CentOS mailing list Subject: Re: [CentOS] How can rpm "%{SUMMARY}" not be consistent?
On Wed, Nov 16, 2011 at 6:56 AM, John Hodrien
wrote:
On Wed, 16 Nov 2011, Akemi Yagi wrote:
What you are seeing is indeed odd. I see 'version 3.1' but not
'3.2'
anywhere on the Summary line of bash.
I would have said you also seeing 'version 3.1.' is one of the very
odd
things, but then I check the bash rpm in a repo and it has 'version 3.1.' in the 3.2-32.el5 rpm.
I think I know some of what is causing what is going on now... After a little testing with a format string of
'"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-%{RELEASE}","%{ARCH}","% {INSTALLTIME:date}"\n' I found I got different results based on the LANG environment setting, which led me to try it with the SUMMARY field too.
###start with a normally configured terminal $ cd /to/your/CentOS/mirror/ $ locale |grep LANG LANG=en_US.UTF-8 $ rpm -q \ --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-%{RELEASE}","%{ARCH}","% {SUMMARY}"\n' \ -p 5.7/os/i386/CentOS/bash-3.2-32.el5.i386.rpm "CentOS","bash","3.2","3.2-32.el5","i386","The GNU Bourne Again shell (bash) version 3.1."
#now change LANG to what I see from cron $ export LANG= $ locale |grep LANG LANG= $ rpm -q \ --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-%{RELEASE}","%{ARCH}","% {SUMMARY}"\n' \ -p 5.7/os/i386/CentOS/bash-3.2-32.el5.i386.rpm "CentOS","bash","3.2","3.2-32.el5","i386","The GNU Bourne Again shell (bash) version 3.2"
Why does a number in a text string change based on LANG?
####and for added fun... look at an agg package from epel $ export LANG=en_US.UTF-8 $ rpm -q --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-%{RELEASE}","%{ARCH}","% {SUMMARY}"\n' agg "Fedora Project","agg","2.5","2.5-9.el5","i386","Anti-Grain Geometry" $ export LANG= $ rpm -q --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-%{RELEASE}","%{ARCH}","% {SUMMARY}"\n' agg "Fedora Project","agg","2.5","2.5-9.el5","i386","Anti-Grain Geometry graphical rendering engine"
Why does the string length change this dramatically based on LANG?
Well I guess I at least know what I need to do to the script to have it be consistent.
I might even get around to filing a RH bug, because this does not seem to me to be a correct behavior to me. does anyone here see these changes as correct behavior (and why)?
-----Original Message----- From: Denniston, Todd A CIV NAVSURFWARCENDIV Crane Sent: Tuesday, November 29, 2011 10:54 To: 'CentOS mailing list' Subject: RE: [CentOS] How can rpm "%{SUMMARY}" not be consistent?
####and for added fun... look at an agg package from epel $ export LANG=en_US.UTF-8 $ rpm -q --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}- %{RELEASE}","%{ARCH}","%{SUMMARY}"\n' agg "Fedora Project","agg","2.5","2.5-9.el5","i386","Anti-Grain Geometry" $ export LANG= $ rpm -q --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}- %{RELEASE}","%{ARCH}","%{SUMMARY}"\n' agg "Fedora Project","agg","2.5","2.5-9.el5","i386","Anti-Grain Geometry graphical rendering engine"
Why does the string length change this dramatically based on LANG?
###And even more fun... try compat-db $export LANG= ;rpm -q --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-%{RELEASE}","%{ARCH}","% {SUMMARY}"\n' compat-db "CentOS","compat-db","4.2.52","4.2.52-5.1","i386","The Berkeley DB database library for Red Hat Linux 7.x compatibility." $ export LANG=en_US.UTF-8 ;rpm -q --qf '"%{VENDOR}","%{NAME}","%{VERSION}","%{VERSION}-%{RELEASE}","%{ARCH}","% {SUMMARY}"\n' compat-db "CentOS","compat-db","4.2.52","4.2.52-5.1","i386","The Berkeley DB database library for CentOS 2.1 compatibility."
Yep that seems a strange one.
On Tue, Nov 29, 2011 at 7:53 AM, Denniston, Todd A CIV NAVSURFWARCENDIV Crane todd.denniston@navy.mil wrote:
Why does a number in a text string change based on LANG?
There's a separate dictionary of translated text strings (called a catalog) for each language. Those translations are looked up and substituted for the default text based on the locale or language settings in the environment. If the number is stored in the catalog as part of the string, you could see this effect.
It appears that either the catalogs have not been updated along with the packages that make reference to them (which would be a bug), or the updated catalogs have not been installed because they are part of a "language pack" that is not listed as a dependency of all the packages that refer to it (which likely is a deliberate optimization).
-
Akemi Yagi -
Bart Schaefer -
Denniston, Todd A CIV NAVSURFWARCENDIV Crane -
John Hodrien -
Nicolas Thierry-Mieg -
Phoenix, Merka