Re: Re: regexless parsing, again?
"Mordechai T. Abzug" <[email protected]> Wed, 19 Sep 2007 20:59:11 -0400
| Newsgroups | gmane.comp.sysutils.loganalysis |
|---|---|
| Message-ID | <[email protected]> |
On Tue, Sep 18, 2007 at 12:51:13PM -0400, Marcus J. Ranum wrote: > >> There have been some amazing advances in hardware to do PCRE. > That'd the the "hardware turbo-charged lipstick on a pig" option. Even parsing is lipstick on a pig. The correct solution is to redesign all log systems so log events are structured, and come with machine-readable catalogs, a la SNMP trap. [Yes, SNMP sucks in general, but the SNMP community dodged most log analysis problems 10+ years ago, such that SNMP analysis solutions have been ahead of syslog analysis solutions for many years, IME.] Too bad unstructured syslog isn't going to go away any time soon. Any attempt to analyze unstructured data is "lipstick on a pig" compared to importing vendor log catalogs in a machine-readable, vendor-neutral format (such as *spit* ASN.1 *spit*). Whether you use regexes, parsing, hybrid multi-layer techniques, or a rigidly structured format with event catalogs, it's all a trade-off between log analysis system design complexity, log analysis system performance, log analysis user burden, log compatibility with "legacy" log systems, and comfort of the developers generating the logs. [Cue a bunch of starry-eyed innocents talking up CEE, CEF, and other "standards" that are supposed to make unstructured logs go away Real Soon Now.] - Morty