"Problem reporting" section of tech spec rather loose

[Adam Williamson]

I'm currently drafting some rough proposed release criteria for Server,
heavily based on the tech spec. I notice the "Problem reporting" section
of the spec is somewhat loose. It reads:

"Problem reporting

Problems and error conditions (e.g. kernel oopses, Selinux AVCs,
application crashes, OOM, disk errors) should all be reported in the
systemd journal.

Support for sending this information to a central place (like abrt does
for crashes today) is mandatory."

the word "must" appears multiple other times in the page, so we have the
old chestnut of whether we're using "must" and "should" with
strict
definitions, or as synonyms, or what. And then the second paragraph
throws in another candidate: "Support...is mandatory". What exactly is
"is mandatory" supposed to mean in this context? Such "support" must
be
*available* in Server? It must be *installed* (and enabled?) for the
system to count as a "Fedora Server system"? Both of these are possible
interpretations. What do we mean by "a central place"? Does a log server
qualify, or do we mean a *public*, *shared* repository, for communal
troubleshooting?

The definition of "problems and error conditions" is also both vague and
apparently wide. There is no precise definition - "e.g." means "for
example", i.e. it is not comprehensive. But just the things listed in
the "e.g." clause are quite broad:

"kernel oopses, Selinux AVCs, application crashes, OOM, disk errors"

abrt handles kernel oopses and crashes. setroubleshoot handles AVCs. We
do not to my knowledge have anything that submits OOM conditions or
"disk errors" to "a central place" in the style of abrt or
setroubleshoot. I'm not even sure that it makes any sense to send them
to some kind of public issue concentrator, given that these are as
likely to indicate local configuration issues or hardware problems as
they are to indicate bugs in Fedora Server.

So...do we feel this needs cleaning up / clarifying? Anyone who was
involved in drafting this bit remember what the intent was?

Thanks!
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net
http://www.happyassassin.net

[Miloslav Trmač]

2014-06-07 0:38 GMT+02:00 Adam Williamson <awilliam(a)redhat.com&gt;:

>  "Problem reporting
>
>  Problems and error conditions (e.g. kernel oopses, Selinux AVCs,
>  application crashes, OOM, disk errors) should all be reported in the
>  systemd journal.
>
>  Support for sending this information to a central place (like abrt does
>  for crashes today) is mandatory."
>
>  the word "must" appears multiple other times in the page, so we have the
>  old chestnut of whether we're using "must" and "should" with
> strict
>  definitions, or as synonyms, or what. And then the second paragraph
>  throws in another candidate: "Support...is mandatory". What exactly is
>  "is mandatory" supposed to mean in this context? 

IIRC this is different from the status quo mainly for *roles*: i.e. even if
the upstream daemon defaults to logging to its own private log files, the
role is expected to configure the daemon to log to journal/syslog instead
(or perhaps in addition).  This would in turn make it possible for a single
“forward logs to $host” configuration to apply to all roles.


>  "kernel oopses, Selinux AVCs, application crashes, OOM, disk
> errors"
>  
I think all of these end up in the journal/syslog already (except may be
the crashes, where kernel writing anything to dmesg is IIRC arch-specific).
    Mirek

[Kevin Fenzi]

On Fri, 06 Jun 2014 15:38:25 -0700
Adam Williamson <awilliam(a)redhat.com&gt; wrote:

>  I'm currently drafting some rough proposed release criteria for
>  Server, heavily based on the tech spec. I notice the "Problem
>  reporting" section of the spec is somewhat loose. It reads:
>  
>  "Problem reporting
>  
>  Problems and error conditions (e.g. kernel oopses, Selinux AVCs,
>  application crashes, OOM, disk errors) should all be reported in the
>  systemd journal.
>  
>  Support for sending this information to a central place (like abrt
>  does for crashes today) is mandatory."
>  
>  the word "must" appears multiple other times in the page, so we have
>  the old chestnut of whether we're using "must" and "should" with
>  strict definitions, or as synonyms, or what. 
I think we MUST clean the MUST and SHOULD's up there. 

>   And then the second
>  paragraph throws in another candidate: "Support...is mandatory". What
>  exactly is "is mandatory" supposed to mean in this context? Such
>  "support" must be *available* in Server? It must be *installed* (and
>  enabled?) for the system to count as a "Fedora Server system"? Both
>  of these are possible interpretations. What do we mean by "a central
>  place"? Does a log server qualify, or do we mean a *public*, *shared*
>  repository, for communal troubleshooting?
>  
>  The definition of "problems and error conditions" is also both vague
>  and apparently wide. There is no precise definition - "e.g." means
>  "for example", i.e. it is not comprehensive. But just the things
>  listed in the "e.g." clause are quite broad:
>  
>  "kernel oopses, Selinux AVCs, application crashes, OOM, disk errors"
>  
>  abrt handles kernel oopses and crashes. setroubleshoot handles AVCs.
>  We do not to my knowledge have anything that submits OOM conditions or
>  "disk errors" to "a central place" in the style of abrt or
>  setroubleshoot. I'm not even sure that it makes any sense to send them
>  to some kind of public issue concentrator, given that these are as
>  likely to indicate local configuration issues or hardware problems as
>  they are to indicate bugs in Fedora Server.
>  
>  So...do we feel this needs cleaning up / clarifying? Anyone who was
>  involved in drafting this bit remember what the intent was? 
I don't, but I think we should clean it up to be consistent on
must/should use. Also, I agree we don't need a central place for "e.g"
stuff. 

kevin

[Matthew Miller]

On Mon, Jun 09, 2014 at 08:31:45AM -0600, Kevin Fenzi wrote:
>  > the word "must" appears multiple other times in the
> page, so we have
>  > the old chestnut of whether we're using "must" and "should"
> with
>  > strict definitions, or as synonyms, or what.
>  I think we MUST clean the MUST and SHOULD's up there.  
http://www.ietf.org/rfc/rfc2119.txt


-- 
Matthew Miller
<mattdm(a)fedoraproject.org&gt;
Fedora Project Leader

[Adam Williamson]

On Mon, 2014-06-09 at 14:21 +0200, Miloslav Trmač wrote:
>  2014-06-07 0:38 GMT+02:00 Adam Williamson
> <awilliam(a)redhat.com&gt;:
>  
>  > "Problem reporting
>  >
>  > Problems and error conditions (e.g. kernel oopses, Selinux AVCs,
>  > application crashes, OOM, disk errors) should all be reported in the
>  > systemd journal.
>  >
>  > Support for sending this information to a central place (like abrt does
>  > for crashes today) is mandatory."
>  >
>  > the word "must" appears multiple other times in the page, so we have the
>  > old chestnut of whether we're using "must" and "should" with
> strict
>  > definitions, or as synonyms, or what. And then the second paragraph
>  > throws in another candidate: "Support...is mandatory". What exactly is
>  > "is mandatory" supposed to mean in this context?
>  
>  
>  IIRC this is different from the status quo mainly for *roles*: i.e. even if
>  the upstream daemon defaults to logging to its own private log files, the
>  role is expected to configure the daemon to log to journal/syslog instead
>  (or perhaps in addition).  This would in turn make it possible for a single
>  “forward logs to $host” configuration to apply to all roles. 
Other tech spec items explicitly indicate when they're scoped to roles,
so if that's intended here, it should probably be added.

>  
>  > "kernel oopses, Selinux AVCs, application crashes, OOM, disk errors"
>  >
>  I think all of these end up in the journal/syslog already (except may be
>  the crashes, where kernel writing anything to dmesg is IIRC arch-specific). 
They wind up in the journal, yes - but it's the 'support for sending
them to a central place' bit that's missing.
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net
http://www.happyassassin.net

[Stephen Gallagher]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 06/09/2014 10:33 AM, Matthew Miller wrote:
>  On Mon, Jun 09, 2014 at 08:31:45AM -0600, Kevin Fenzi wrote:
> >> the word "must" appears multiple other times in the page, so we
> >> have the old chestnut of whether we're using "must" and
> >> "should" with strict definitions, or as synonyms, or what.
> > I think we MUST clean the MUST and SHOULD's up there.
>  
>  http://www.ietf.org/rfc/rfc2119.txt
>  
>   
Yes, these are the definitions by which we MUST operate.
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1
> Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
>
> iEYEARECAAYFAlOZvUoACgkQeiVVYja6o6N1CQCfYmN/gA+hXManmJJVN7WBzApw
> nIsAniIUIq1oDaO6Inm9YoAP5scJKnjK
> =CMXA
> -----END PGP SIGNATURE-----