- From: Brian Behlendorf <brian@organic.com>
- Date: Wed, 28 Jun 1995 20:01:10 -0700 (PDT)
- To: Andrew Payne <payne@openmarket.com>
- Cc: www-talk@www10.w3.org
On Wed, 28 Jun 1995, Andrew Payne wrote: > >AN342sQ StartRequestTime: 804531212.825 > >AN342sQ Method: GET > >AN342sQ URI: /foo.html > >AN342sQ HTTP-level: HTTP/1.0 > >AN342sQ Accept: text/html, image/gif, */* > >AN342sQ Referrer: http://host.com/bar.html > > >Thoughts? We're busy plugging holes in Apache 0.7 right now, so we > >haven't started building TOO much yet. > > Do you think that folks need to have the log items for each request > interleaved with other items? I'm thinking here about the extremes of load > (4,000,000 hits/day --> ~50 hits/sec --> 750 log entries/sec, assuming 15 > entries per request.) Four million hits/day is a lot now, but it won't be > much in a year. Will this scale? Remains to be seen. This info is *not* meant to be written to a flat file, it's meant to be passed via pipes or IPC or sockets to another machine or a stream within the same daemon itself - it simply describes an API. It's also very flexible - if speed/load was an issue, then the server could selectively decide not to send info down the pipe - info like HTTP-leveol, Accept: headers, progress reports, etc. The logging tools could be written to deal with that lack of information gracefully. > We've actually started work in the other direction, where the server might > accumulate the logging information for some user-specified period of time > (several seconds, typically), and dump the information in a single I/O > operation. You can still do all of the fancy stuff about displaying > response times, etc., but they'll be a few seconds behind the curve. That would seem to be implementable as a linked object code module using this API. Brian --=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-- brian@organic.com brian@hyperreal.com http://www.[hyperreal,organic].com/
Received on Wednesday, 28 June 1995 23:01:11 UTC