Re: [sip-clf] A syslog approach to sip logging

["Rainer Gerhards" <[e-mail]>]

> -----Original Message-----
> From: [e-mail] [mailto:[e-mail]] On
> Behalf Of Cullen Jennings
> Sent: Thursday, February 04, 2010 12:39 AM
> To: David B Harrington
> Cc: 'SIP-CLF Mailing List'
> Subject: Re: [sip-clf] A syslog approach to sip logging
> 
> 
> One requirement that I suspect you will find fairly universal about
> transporting around SIP log like informations is that
> 
> 1) it is reliable

I need to first ask what level of reliability do you need? There are several
flavors of reliability and usually loss of message is acceptable at some
level. 

For example, let's assume the process writes out log messages to a text file.
Under (allmost?) all modern operating systems and by default, this means the
OS accepts the information to write, acks it, does NOT persist it to storage
and lets the application continue. The actual data block is usually written a
short while later. Obviously, this is not reliable: you can lose log data if
an unrecoverable i/o error happens or something else goes fatally wrong.

This can be solved by instructing the operating system to actually persist
the information to durable store before returning back from the API. You have
to pay a big performance toll for that. This is also a frequent question for
syslog data, and many operators do NOT sync and accept a small message loss
risk to save themselves from requiring a factor of 10 servers of what they
now need.

But even if writes are synchronous, how does the application react? For
example: what shall the application do if log data cannot be written? If you
really need reliable logging, the only choice is to shutdown the application
when it can no longer log. I know of very few systems that actually do that,
even though "reliability" is highly demanded. Here, the cost of shutting down
the application may be so high (or even fatal), that the limited risk of log
data loss is accepted.

There are a myriad of things when thinking about reliability. So I think it
is important to define the level of reliability that is required by the
solution and do that in detail. To the best of my knowledge, this is also
important for operators who are required by law to do "reliable" logging. If
they have a risk matrix, they can define where it is "impossible" (for
technical or financial reasons) to achieve full reliability and as of my
understanding this is information auditors are looking for.

Having said that, current RFC5425 provided good, but not perfect reliability.
A big problem is that RFC5425 does not provide an application-layer
backchannel and relies on TCP to achieve reliability. That means that when a
connection breaks, there is a window of uncertainty (sender's tcp send
buffers + current TCP window) for which messages may or may not be lost. This
is comparable to the unsynced file access paradigm.

RFC3195 provides a "fully reliable" protocol, but was not accepted by the
syslog community because it was a too-large departure from traditional
syslog. This is the main reason RFC5424/5425 do not specify backchannels. The
syslog WG will hopefully create a "RFC3195bis", which will be a bit simpler
than 3195 was and supports the new header formats. But this is "hopeware" and
not something you can put on a system today or some time this year).

> 
> 2) we can include complete SIP messages. These can get very large.
> (Magnus posted a 40k SDP to the mmusic awhile back, don't even ask how
> large MESSAGE messages get in the wild)

Size was a large problem with traditional syslog and the syslog community
still prefers small sizes. Syslog recommend "maximum message sizes", which
place an upper bound for the message on a given transport and implementation.
The minumum (maximum message size), so the max size that must be supported by
every implementation, is just 2k. The recommended (maximum message size) is
8k. Both are obviously too low. Thankfully, we managed to set no upper limit
per se on the log records. So a syslog receiver may be configured to use any
maximum message size an operator requires - but implementations are not
required to support that capability. IHE, for example, uses 32k for their
messages. Our Windows Event-Log to syslog conversion tool sometimes need to
emit messages in excess of 128K. 

So syslog can support that requirement, but some syslog implementations may
not support it. Also, an operator must specifically configure syslog
components for non-default sizes to make everything work. There is definitely
a failure potential in this process, and it is not trivial in a larger
network with different syslog users.

Rainer

> 
> Can you say a bit more about how to use syslog to achieve these.
> 
> 
> On Feb 2, 2010, at 11:53 AM, David B Harrington wrote:
> 
> > Hi,
> >
> > Operators often use syslog to carry Apache CLF log data.  Syslog, in
> > practice, is primarily used for tunneling Apache CLF format. This
> > seems to be attractive for operators, because of already existing
> > infrastructure and syslog knowledge.
> > I suggest it makes sense to develop a sipclf format that can be
> > carried within an IETF syslog message. In general, that means
> > information in an ascii format dumped into a local file that can be
> > parsed with tools like grep, and that can later be secured, filtered,
> > transported, aggregated, and correlated using existing
> infrastructure.
> >
> >
> > Multiple use cases raised by contributors in this WG lead to
> different
> > requirements. Some want to pay attention to the data dumped by one
> > server; some want to follow traffic flows through the network; some
> > want to filter on standardized fields; some want to aggregate and
> > correlate log information.
> >
> > It is not enough to figure out how to dump data on a single system;
> > that data will need to be compatible with infrastructure used to
> > provide secure transport, filtering, correlation, etc. Operators
> > already have existing infrastructures designed for long-term
> archiving
> > of (potentially enormous) logging information, and the goal to
> > correlate log records, including data-mining.
> >
> > Syslog is already widely deployed, is well understood by operators,
> > and the IETF syslog WG has standardized many aspects of security and
> > transport, such as (D)TLS-secured transport, support for large
> > messages, optional digitally signed logging for law enforcement and
> > for message stream integrity checking, etc.
> >
> > IETF syslog standardizes a number of parameters useful for
> > correlation, such as
> > facility (specific applications), severity classification, timestamp,
> > hostname, the name of the application sending the message (often
> > syslogd), process ID, and message ID that are in the syslog header in
> > ascii format. These were designed to be compatible with ITU logging
> > standards, and the ALARM-MIB, to provide easier correlation of events
> > across different event reporting mechanisms. Also to improve
> > correlation, work has been done to translate syslog messages into
> SNMP
> > traps, and SNMP traps into syslog messages.
> >
> > The IETF syslog standard also provides structured data elements. SDEs
> > are
> > designed to supplement the human-readable text with
> > application-parseable data fields (also encoded in 7-bit ascii),
> which
> > makes it easier for applications, such as security management
> systems,
> > to extract and correlate the data across vendor implementations, and
> > across nodes in a network.
> >
> > The IETF syslog standard already defines some SDEs that would likely
> > be useful for the problems sip clf is trying to resolve, like
> > precisiely tracking sequence inside a networked system: a
> > high-precision timestamp, the quality of the time source for a given
> > system, time zone accuracy, whether a node is synched with a network
> > time source, the origin of a log entry (useful after aggregation and
> > relay), the ip address at time of logging, an enterprise identifier,
> > the software that generated the message (i.e., the application that
> > asked syslogd to send the message), the software version, a sequence
> > number to provide sequence as seen by a single node, the sysUpTime of
> > a co-resident SNMP system, and the language used within the
> > human-readable MSG. The IETF syslog standard also contains some
> > recommendations on intelligently dropping messages if the log volume
> > becomes overwhelming.
> >
> > The syslog WG deliberately did not standardize the content of the
> > human-readable message field. The WG standardized the header, and has
> > provided SDEs to standardize certain aspects of the information where
> > consensus can be reached. Having both (potentially non-standardized)
> > human-readable data, and standardized human-and-machine-readable
> > structured data in the same message addresses a wide range of use
> > cases, and gives the human more information to work with to interpret
> > an event.
> >
> > I propose that the WG reach consensus on specific fields of data that
> > would be good to standardize, such as those defined in the problem
> > statement doc, and define them as syslog SDEs (which, remember, are
> > text fields so they would be greppable and printable and diff-able
> and
> > human-readable). Structured data elements would better support
> > application-parsing of the data, such as for training IDS/IPS anomaly
> > engines.
> >
> > There are only a few restrictions placed on the content of a MSG
> field
> > in a syslog message. According to the problem-statement document,
> > there already exist a number of proprietary sip clf formats. Well, if
> > those are in a format that can fit within the MSG field within an
> IETF
> > standard syslog message, then that proprietary data can also be
> > carried in the syslog message. Any vendor-specific log-parsing tools
> > would continue to work with the extracted MSG field, and they could
> be
> > supplemented by tools that can parse the standardized SDE
> information.
> > The IETF syslog standard also supports vendor-specific SDEs for
> > extensibility of structured data.
> >
> > In a similar manner to the dual stack approach for IPv4/IPv6
> > transition, implementers could choose to drop specific fields from
> > their proprietary formats as consensus on useful SDEs is reached, and
> > their tools are adapted to use the standardized header and SDE
> > information.
> >
> > This approach would work with the WG goal to constrain its focus to
> > the "useful  information" and not need to reinvent solutions such as
> a
> > data modeling language, character sets, delimiters, secure transport,
> > log integrity checking, log filtering, log aggregation and
> correlation
> > issues, and so on.
> >
> > I do not see much benefit from designing a whole new ascii file
> format
> > that no existing tools support (except generic text-handling tools
> > like grep), and operators would need to learn in addition to the
> > semantics in the information model.
> >
> > I recommend the WG focus on specific and actually existing problem
> > cases, and build the semantic "information model" incrementally. Then
> > use the existing standard syslog format to provide an example data
> > model, which inherits the benefits of an existing widely-deployed
> > infrastructure for logging.
> >
> > David Harrington
> > [e-mail]
> > [e-mail]
> > [e-mail]
> >
> >
> >
> > _______________________________________________
> > sip-clf mailing list
> > [e-mail]
> > https://www.ietf.org/mailman/listinfo/sip-clf
> 
> For corporate legal information go to:
> http://www.cisco.com/web/about/doing_business/legal/cri/index.html
> 
> 
> 
> _______________________________________________
> sip-clf mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/sip-clf