https://issues.apache.org/SpamAssassin/show_bug.cgi?id=6956
Bug ID: 6956
Summary: Bayes/Redis/Lua buglet and speedup
Product: Spamassassin
Version: SVN Trunk (Latest Devel Version)
Hardware: All
OS: All
Status: NEW
Severity: minor
Priority: P2
Component: Libraries
Assignee: [email protected]
Reporter: [email protected]
While reviewing code in BayesStore/Redis.pm I noticed one bug
and one speedup opportunity there.
The bug is in multi_hincrby Lua code, which test for s and h
being nonzero in order to skip some unnecessary redis calls:
if s ~= 0 then
The problem is in the fact that all arguments are passed from
redis to Lua as strings (not numbers), but the relational
operators in Lua do not perform type coertion, so any string
is always different from a numeric zero.
Luckily no real harm is done, it's just that redundant calls
to increment counts by a zero are not skipped, wasting some
(little) time.
The jackpot optimization opportunity lurks in the Lua
procedure used by tok_get_all. Remembering the inefficient
implementation of parsing multivalued result in a CPAN module
Redis, changing the result as returned by the multi_hmget_script
Lua code from a multivalued result (one spam/ham pair for
each bayes token) into a single string of concatenated pairs,
to be later split in perl on spaces, offers a nice boost.
On our production server, comparing a median time to execute
tok_get_all for an hour's worth of mail, before and after
the change, dropped the median time from 10 ms to 6 ms.
A noteworthy speedup for such a seemingly insignificant change
in passing back results.
Taking into account that Lua virtual machine is a register-based
machine (and checking the compiled code), also took the opportunity
to use local variables where they make sense, as accessing them
(e.g. in a loop) is cheaper than accessing globals.
Also avoided one unnecessary call to keys() in Plugin/Bayes.pm
and avoided unnecessary entering/exiting a block in a map() call
there. Probably insignificant, but anyway...
--
You are receiving this mail because:
You are the assignee for the bug.
