Re: [openxdas-devel] Discussion about changes to specification
Brought to you by:
dsandersorem,
jcalcote
From: Wayne H. <wh...@no...> - 2007-04-19 18:02:16
|
Although having comma separated <idname>=3D<idvalue> pairs in the = correlation id field is more flexible, it adds complexity when parsing or = analyzing the data. I suspect one string value for the correlation id will = be fine for most of the applications. In a database's term, every record = has only one primary key that serves to correlate the records within = tables, even though there may be other unique keys that identifies the = same record. Application can create a compound correlation id with the = xdas naming format or store other unique keys in the EVT section for this = type of scenario. They can even stores the <idname>=3D<idvalue> pairs. My = point is not to impose any rule on this field.=20 In the database logger, this will make querying the data tables easier. = For example, if I want to find a series of actions of a perticular = workflow request, I can just do "select * from table_name where correlation= _id =3D '<request_id_value>'. A free form, <idname>=3D<idvalue> pairs in the correlation if field will = make this type of query harder to write. Wayne Hsu Senior Software Engineer, eXtend Engineering (703) 967-4980 Novell, Inc., the leading provider of Net Business Solutions. www.novell.com >>> "David Corlette" <dco...@no...> 04/19/07 12:38 PM >>> Hi Wayne, Yes, that's fine, but mostly I'm curious what kind of event interrelationsh= ips people expect to see. This in part is because on the backend Sentinel = would split the data off into separate fields for easy filtering, most = likely. >>> On Thu, Apr 19, 2007 at 10:01 PM, Wayne Hsu wrote:=20 > Are we confined to these names for the Correlation only? I thought the = field=20 > is extensible, therefore, instrumenation application can use whatever = names=20 > that make sense within its domain, be it sessionid or auid, as long as = it=20 > follows the name=3Dvalue format. >=20 > Wayne Hsu > Senior Software Engineer, > eXtend Engineering > (703) 967-4980 > Novell, Inc., the leading provider of Net Business Solutions. > www.novell.com >=20 >=20 >>>> "David Corlette" <dco...@no...> 04/13/07 7:43 PM >>> > John, >=20 > I agree with what you are saying, exactly. Correlation has a slightly=20= > broader meaning in the context of Sentinel, but the use of that word = here -=20 > e.g. to help us correlate between several related events - is perfect. >=20 > As for the field format, that also works for me. >=20 > So far we seem to have come up with: > sessionid > processid > threadid > requestid >=20 > I notice that the LAF (SuSE Linux audit framework) uses "auid" - but=20 > perhaps this could be considered a "sessionid" or "requestid"? > Anyone else care to comment? >=20 >>>> On Sat, Apr 14, 2007 at 12:58 AM, in message=20 > <461...@no...>, > John Calcote wrote:=20 >> Okay, I see now.=20 >> =20 >> You called it instanceid, but in reality, it's a form of general = purpose=20 >> correlation id, right? I think the concept of correlation id is = generic=20 >> enough (regardless of the domain) to put it into the HDR section.=20 >> Furthermore, why not define some standardized correlation id types, = and=20 > apply=20 >> them within a single field in the HDR section?=20 >> =20 >> For instance, you called it instanceid, but that could mean anything. = Why=20 >> not have a field in the HDR section called <correlation> whose format = is comma=20 >=20 >> separated <idname>=3D<idvalue> pairs. For instance: >> =20 >> sessionid=3D096v432lkna9er8h42okn,processid=3D12345,threadid=3D27 >> =20 >> That way, we can add id types as we see the need - the field is = extensible,=20 >> like the event data field, but it is specialized enough, and globally=20= >> important enough to warrant its own field, and is generic enough in = nature=20 > to=20 >> events in general to warrant a place in the HDR section. >> =20 >> From an API perspective, header information is VERY controlled - most = of it=20 >> is self generated. That is, not specified by the caller, but rather by = the=20 >> XDAS client library. This is not always true, but where it's not true - = as=20 > in=20 >> the cases of (eg) event number and outcome code - the client API = provides=20 >> well-defined functions to set those values and ensures that they = conform to=20 >> the XDAS standard. We would want to provide similar additions to the = API to=20 >> set correlation information, and we would want to auto-generate some of = the=20 >> more obvious id values such as process and thread. The auto-generated = values=20 >=20 >> could be policy controlled by configuration file entries. >> =20 >> Whereas event data is very free form and user-defined/specified, = correlation=20 >=20 >> information is very controlled. >> =20 >> I believe correlation information is important enough to warrant this = sort=20 >> of management. What do you think? >> =20 >> John >>=20 >>>>> On 4/12/2007 at 7:04 PM, in message <461...@no...>= , David=20 > Corlette <dco...@no...> wrote: >>=20 >> Hi John, >>=20 >> I debated about this - I think the Instance ID will be wildly useful = outside=20 >=20 >> the EVT structure (NVP) as a way of connecting multiple events together = into=20 >=20 >> a stream of related events. So I *absolutely* think that it needs to = have=20 >> its own field, and I think it can be domain-specific as to the meaning = of=20 > the=20 >> ID (NAM, for instance, creates some sort of connection ID which is = opaque,=20 >> but would serve this purpose). >>=20 >> As for putting it in EVT, I don't actually care - HDR seemed to be = more=20 >> information about the structure of this event, not actually any event = data. =20 >=20 >> But I could see pulling it up to that level as well. >>=20 >>=20 >>> May I ask why you've decided to add this value here? Event Data has a = very=20 >>> well-defined format: comma separated name=3Dvalue pairs. You could = have=20 >>> specified that instanceid be added as a well-known name/value pair. = Better=20 >>> still, if it's supposed to be a process id, then simply call it=20 >>> processid=3Dxxx, as a well known name/value pair. The standard doesn't = define=20 >>> any well-known name/value pairs today. If it's something that is = really=20 >> useful=20 >>> in all cases, perhaps we should consider adding it to the HDR section. >>=20 >>=20 >>=20 >=20 >=20 > -------------------------------------------------------------------------= > Take Surveys. Earn Cash. Influence the Future of IT > Join SourceForge.net's Techsay panel and you'll get the chance to share = your > opinions on IT & business topics through brief surveys-and earn cash > http://www.techsay.com/default.php?page=3Djoin.php&p=3Dsourceforge&CID=3D= DEVDEV > _______________________________________________ > openxdas-devel mailing list > ope...@li... > https://lists.sourceforge.net/lists/listinfo/openxdas-devel ------------------------------------------------------------------------- This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/ _______________________________________________ openxdas-devel mailing list ope...@li... https://lists.sourceforge.net/lists/listinfo/openxdas-devel |