[geeks] RANT (ftp)

[velociraptor]

On 10/25/06, Bryan Fullerton <fehwalker at gmail.com> wrote:
> Yes, suits love the pictures. And they really love the pictures of
> stats - abstraction of an abstraction, reducing any significant
> reality impact.

People who can't wrap their heads around the numbers themselves,
and/or create their own pictures to interpret the data.

> I would have given a kidney to switch to Urchin back when I had to
> support a WebTrends Enterprise install processing *log files* for a
> 1M+ view/day web site (there is no logic in this, don't try to find
> it).
>
> Of course, Urchin is now a Google Analytics product. I'm not
> optimistic that this is progress.

They've delayed the release of the new version over 12 months, so,
yeah, def. not progress.  They are still trying to convince people
it's feasible to direct their "analytic" traffic through google
itself.  Umm, not really.

I was using Urchin on a site with about 3M+ hits/week, with the
abomination that is SunOneWeb server (ok, it works, sorta, but can we
just use Apache, *please*?).

The guy that originally installed it had zippo experience with it (and
probably getting paid 2x what I was an hour--*puke* consultants).  He
just a$$umed the pretty clickety-click web interface would work, was
too lazy to write his own scripts to pull over logs, etc.  Long story
short, Urchin barfs on URLs longer than a certain number of characters
(in our case 3k chars--don't ask--I didn't test to find out what the
real limit was).

I ended up reworking the entire log processing set up he'd built to
work entirely from local logs that I pulled with cron b/c Urchin also
blocks *all* processing if one log analyzer process hangs when you use
*it's* scheduler.  It will do concurrent processing fine when run from
a shell (modulo disk access delays).  But (isn't there always), if you
run it from the command line, it doesn't "remember" which logs it's
already processed, so you have to be very careful not to re-run logs.

2 months of my life lost to YACCOTS product, thanks to
trouble-shooting, hand-holding marketing idiots, juggling logs to do a
clean cut-over, etc., etc. ad naseum, and fighting with the so-called
hosting service that hadn't bothered to include logs in backups.  Web
logs are arguably the most important data on the servers in a
corporate site's DMZ when you are talking marketing driving
sales--second only to the customer db itself.  So of course the
hosting company was only backing up the static portion of the web site
that could easily be sync'd from staging. <boggle> [0]

=Nadine=

[0] Same "enterprise" hosting facility claimed to have "expertise" in
Veritas Cluster, yet installed Oracle on VCS with the *APPLICATION* on
the disks shared between the nodes because it was "easier to patch
this way".  Uh, yes, it's easier for YOU to patch that way, but it
means WE--you know, the *customers*--have a total outage when you
patch instead of a 3-4 minute cut-over, or an outage only as long as
it takes to apply the database changes required by patching Oracle.  I
swear their "expert", who barely spoke English, still didn't get what
the issue was when I explained, in terms reserved for a fresh n00b sys
admin, what the problem was.

p.s. Could someone pass me the pitcher of margarita's--it's Friday, right?