Mahlon E. Smith <mahlon@laika.com> [Wed, 09 Nov 2011 16:06:49 -0800] rev 20
Removed tag 0.1
Mahlon E. Smith <mahlon@laika.com> [Wed, 09 Nov 2011 16:02:57 -0800] rev 19
Added tag 0.1 for changeset d4ce82194b64
Mahlon E. Smith <mahlon@laika.com> [Wed, 09 Nov 2011 15:54:37 -0800] rev 18
- 1st pass at documentation with the README.
- Attempt to detect trivial redirect loops.
- Loosen the request parser enough to just accept URLs, so you can
easily test rewrite behavior by just running volta and pasting URLs.
- Alter the whitelist "negative" rule syntax to more closely match
normal rules. Rename references to negative rules from "whitelist"
rules, to avoid confusion.
Mahlon E. Smith <mahlon@martini.nu> [Mon, 07 Nov 2011 10:43:09 -0800] rev 17
Retain the CDB struct between lookups, only freeing when reopening the
db. Big, big speed boost.
Mahlon E. Smith <mahlon@martini.nu> [Mon, 07 Nov 2011 10:43:06 -0800] rev 16
Ensure that all output is flushed immediately.
Fix free() bug for non-matching rules.
When creating the rules cdb, make the tmpfile in the cwd instead of /tmp.
Make regular expression matches case insensitive.
Mahlon E. Smith <mahlon@martini.nu> [Sat, 05 Nov 2011 12:52:29 -0700] rev 15
Add whitelisting rules, to negate other matches if they come first in
the ruleset. Remove the results set array completely, it's faster and
requires less memory to do the comparisons inline.
Mahlon E. Smith <mahlon@martini.nu> [Fri, 04 Nov 2011 20:34:28 -0700] rev 14
There isn't a fast way to look up ( value exists or null ) for every
piece of data that we're handed from squid -- a full table scan must be
performed for each request, regardless of any indexing in place. With
a decent number of rules, it's much, much slower than I anticipated,
making it fairly pointless to even do parsing in ragel for speed.
Trading some matching features (IPs, for now) for speed, and aiming at
just doing good host/path matching at the moment, using CDB for fast
host keying. Ripped out all the SQLite work, replacing it with a first
round of CDB lookups.
Added a ascii rulefile to cdb converter flag. Churning through the
requests again!
Also, fixed an off-by-one error with line extending.
Mahlon E. Smith <mahlon@laika.com> [Mon, 31 Oct 2011 17:17:07 -0700] rev 13
1st iteration of volta actually doing something. Process the request,
find the best matching rule, and rewrite the request.
Without the DB queries, volta was parsing over 750k requests a second.
Currently, it's down to 129.5 with 1161 rules in place. Yikes. I may
need to re-evaluate some design choices here.