syslog-ng/syslog-ng

syslog-ng is an enhanced log daemon, supporting a wide range of input and output methods: syslog, unstructured text, queueing, SQL & NoSQL.

Balazs Scheidler

@bazsi

$(context-lookup) iterates both the original messages and the synthetic one added on top, meaning house.html would be repeated.

I agree this is more complicated than it should be.

The solution to your immediate problem is to cut off the final element from the list returned by $(context-lookup), e.g. something like $(list-slice 0:-1 $(context-lookup ('$attachment' ne '') ${attachment}))

I'll check that expression in a minute.

Balazs Scheidler

@bazsi

The longer term solution is to make this a bit more intuitive. I was thinking on making $(context-lookup) ignore the last element of the context. The problem with this is that it is an incompatible change, one that is difficult to communicate and fix. The other problem with this approach is that only grouping-by() (and db-parser()) aggregation is what adds this extra, synthetic message in the context and there are other cases which don't have a synthetic message generated (e.g. simple matching rules in db-parser()).

Balazs Scheidler

@bazsi

Another solution would be to have an argument to $(context-lookup) and $(context-values) to ignore the synthetic message, which would at least make the expression a bit easier to come up with and read.

Balazs Scheidler

@bazsi

A 3rd solution is to simply document this better.

arteta22000

@arteta22000

Thanks @bazsi for the explanation.

I have another question. With these ugly cisco ironport logs. There are 3 different ids, MID is the main id. In the raw log example above, we see that a session can start with "ICID" ( New SMTP ICID 219986669 ...) and then another log does the mapping (ICID <> MID , Start MID 198090335 . ICID 219986669 ...).
Can I aggregate the logs with ICID and DCID and then aggregate with MID. I don't know if I'm clear?

arteta22000

@arteta22000

I repost an example of a log session:

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: New SMTP ICID 219986669 interface responsibility (8.16.8.16) address 17.39.24.2 reverse dns host cuis.com verified no

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: ICID 219986669 RELAY SG UNKNOWNLIST match sbrs[0.0:7.0] SBRS 5.1

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: Start MID 198090335 ICID 219986669

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: MID 198090335 ICID 219986669 RID 0 To: fcvez@yaho.com

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: MID 198090335 ready 12710 bytes from tgamb@hotil.com

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: MID 198090335 ICID 219986669 RID 0 To: fcvez@yaho.com

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: MID 198090335 Subject 'Which add raise.'

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: MID 198090335 Message-ID 'tgamb@hotil.com'

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: MID 198090335 antivirus negative

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: MID 198090335 attachment 'try.odt'

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: MID 198090335 attachment 'house.html'

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: Delivery start DCID 177209764 MID 198090335 to RID [0]

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: MID 198090335 RID [0] Response '2.6.0 22633284F46CFCC33@mray.com [InternalId=4230] Queued mail for delivery'

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: MID 198090335 using engine: CASE spam positive

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: MID 198090335 rewritten to MID 204028130 by add-footer filter 'Footer Stamping'

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: Message done DCID 158099468 MID 198090335 to RID [0] ['([168.xxxx.137])\r\n by hester.org with ESMTP; 11 Aug 1976 17:03:41 +0200)']

<22>Mar 18 10:21:29 server01/8.8.8.8 serv-mail1: Info: Message finished MID 198090335 done

Balazs Scheidler

@bazsi

This is possible to do in db-parser() but that's pretty arcane to use with its XML based format. @faxm0dem created a puppet wrapper for it and yaml based generation that is a lot easier to use.

The keyword to connect two independent correlation state is called the <create-context> action. Here's a bit of documentation for that:

Example: Creating a new context from an action

In syslog-ng OSE version 3.8 and newer, you can create a new context as an action. For details, see Element: create-context.

The following example creates a new context whenever the rule matches. The new context receives 1000 as ID, and program as scope, and the content set in the <message> element of the <create-context> element.

<rule provider='test' id='12' class='violation'>
  <patterns>
    <pattern>simple-message-with-action-to-create-context</pattern>
  </patterns>
  <actions>
    <action trigger='match'>
      <create-context context-id='1000' context-timeout='60' context-scope='program'>
        <message inherit-properties='context'>
          <values>
            <value name='MESSAGE'>context message</value>
          </values>
        </message>
      </create-context>
    </action>
  </actions>
</rule>

In the case of db-parser() you are not limited to simply stuffing messages into a "context" that grouping-by() does, rather you have individual rules that can act based on the input.

This is a powerful feature but one that is pretty difficult to use. It honours your efforts with performance though.

What you would need to do basically is the following:

1) match the beginning of the session and correlate on ${CID}

Balazs Scheidler

@bazsi

2) match the "Start CID ... MID ..." message and on that create a context, this will create a NEW correlation context with a key that is present in the current message (e.g. ${MID}), in this case you can generate a synthetic message that becomes part of your new correlation state.

3) as you are matching the rest of the session (e.g. which contain ${MID}), you correlate against ${MID} and you will find a very-first synthetic message in this context. The one that you created as a result of the first point, e.g. the CID based correlation. Here you would need to represent all CID based data as a single message. With syslog-ng lists, you could encapsulate all CID based messages into a single syslog-ng list. e.g $(context-values $MSG) as a name-value pair. but if you have name-value pairs to extract, that might be a better solution.

hope this helps.

arteta22000

@arteta22000

Thank you, it seems complex for me. I preferred the solution of aggregating cid/mid/dcid and then aggregating the whole.
Currently; we do this with a python script and redis, but we have performance problems.

But I can't do it with grouping-by ? with several grouping-by subsequent

Balazs Scheidler

@bazsi

you are right. that could work.

I didn't think of that.

arteta22000

@arteta22000

How do I do this? I can't fowardthe aggregation from one grouping-by to another.

Balazs Scheidler

@bazsi

if you performed grouping-by()s separately: 1) one context CID stuff (and the Start CID/MID message), 2) one for the MID based messages plus the Start CID/MID message), then both contexts would have both MID/CID information. Then a 2nd layer of grouping-by() could correlate the aggregation of the 1st and the 2nd.

exactly.

good idea on using two. I was just too focused on doing it in one step.

arteta22000

@arteta22000

But how to send the aggregation from one grouping-by to another? I have the feeling that "grouping-by" works by default with the "message" macro?

example:

parser groupingby {
grouping-by(
key("${cid_id}")
scope("HOST")
value("event.aggregate" "ok")
value("MESSAGE" "Session completed >> ${attachment} cid_id:${cid_id} info1: ${info1}}")
inherit-mode("context")
)
timeout(10)
inject-mode("pass-through")
);
};

parser groupingby2 {
grouping-by(
key("${mid_id}")
scope("HOST")
trigger( "${.classifier.rule_id}" eq "regex_finished_status" )
having( "${finished_status}" eq "done" )
aggregate(
value("event.aggregate" "ok")
value("MESSAGE" "Session completed >> ${attachment} mid_id:${mid_id} cip_icid: ${cip_icid} ironport_interface_name: ${ironport_interface_name} cid_id:${cid_id} info1: ${info1}}")
inherit-mode("context")
)
timeout(10)
inject-mode("pass-through")
);
};

arteta22000

@arteta22000

With "inject-mode("internal")" in the first grouping-by ?

Balazs Scheidler

@bazsi

inject-mode(internal) would generate the message as if it was coming from the internal() source

I think inject-mode(passthrough) is better, that way the message is generated as if coming from grouping-by() itself.

arteta22000

@arteta22000

I'm going to do some tests but I'm a bit lost,)

My first "grouping-by" will generate a macro (that I configure; in the value parameter) with the aggregation (example: mid :xxx cid:xxx source_smtp: xxxx).

How to forward this information (cid:xxx source_smtp: xxxx) to the other grouping-by (after) so that it aggregates this information with the central mid

Balazs Scheidler

@bazsi

just connect the two grouping-by()s in a log path, e.g.

log { source(whatever); parser { grouping-by(FIRST...); grouping-by(SECOND...); }; ... };

you can also define it as a top-level parser block and then reference it:

parser p_ironport {
    grouping-by(FIRST...);
   grouping-by(SECOND...);
};

log { source(whatever); parser(p_ironport); ... };

any element can be concatenated to another one, each would be processed in order.

NOTE: in the first example I used braces to indicate that the parsers are defined in-line, within the log statement. In the 2nd example, I used parenthesis which references a parser that was declared earlier.

so parser { in-line-parser-expression }; or parser(namedparserblock);

arteta22000

@arteta22000

I tried, and I got 2 aggregated logs in output. I need to put "value("MESSAGE")" in the first grouping-by block?

The second grouping-by will see the message aggregated using the message macro ?

Balazs Scheidler

@bazsi

I don't really understand the question. I'd try something like this (not syntax checked, just from the top-of-my-head):


log {
    source(whatever);
   parser { db-parser(file('ironport.xml')); };
    log {
        filter { <match only iron port logs...>; };

        if ('${CID}' ne '') {
            parser { grouping-by(<CID aggregation options> inject-mode(passthrough)); };
        };
        if ('${MID}' ne '') {
            parser { grouping-by(<MID aggregation options> inject-mode(passthrough)); };
        };
        if (<match aggregated message from either CID or MID based aggregation>) {
            parser { grouping-by(<MID + CID result aggregation> inject-mode(passthrough)); };
        };
       filter { <match only aggregated logs emitted by the last grouping-by()>); };
    };
    destination(whatever);
};

Balazs Scheidler

@bazsi

the messages flow through the sequence one by one.
1) The source() statement in the front pulls in any messages received by the specific source.
2) the db-parser() right next would extract ironport name-value pairs from the messages
3) the embedded log {} statement just encapsulates a set of operations
4) the first filter embedded in the log {} statement would drop anything but ironport (I am not sore if the db-parser() field extraction sets a specific field or not, but that's certainly doable based on some criteria)
5) the first if() checks if $CID is set, if it is, a grouping-by() parser is executed, which should in turn emit an aggregated result of all messages where CID was set.
6) the second if() checks if $MID is set, similarly to the first, it would need to aggregate the $MID related messages into a result
7) the last if() would match messages that either the first or the second grouping-by() emitted and runs a 3rd grouping-by()
8) this third grouping-by() is our end result
9) the filter as the last statement within our log {} would drop anything but the aggregated result
10) the destination only receives the output of the entire log {} statement, e.g. the result of the 3rd aggregation

Balazs Scheidler

@bazsi

now as I think of it, the sample can even be improved so that the MID based grouping-by (e.g. the 2nd aggregation) does not receive the output of the first one. The two aggregation both get a copy of the incoming data, and their combined output is fed through the 3rd aggregation

but I leave you to it, this is just improvement in performance, functionally it should work just like that.

Where communities thrive

People

Repo info

Activity