I recently upgraded from SA 2.64 to 3.1.0 and one thing I've noticed is bayes
operations are inordinately slower than they used to be. Especially expiry runs,
which often take in excess of 15 minutes to complete.
I've tried changing the lock_method to flock, but things are still slow.
Even just doing a --dump magic seems to take a strangely long time:
[EMAIL PROTECTED] root]# time sa-learn --dump magic
0.000 0 3 0 non-token data: bayes db version
0.000 0 780736 0 non-token data: nspam
0.000 0 91958 0 non-token data: nham
0.000 0 178712 0 non-token data: ntokens
0.000 0 1132316197 0 non-token data: oldest atime
0.000 0 1132597905 0 non-token data: newest atime
0.000 0 1132597952 0 non-token data: last journal sync atime
0.000 0 1132592420 0 non-token data: last expiry atime
0.000 0 212693 0 non-token data: last expire atime delta
0.000 0 63586 0 non-token data: last expire reduction
count
real 0m10.568s
user 0m9.410s
sys 0m0.370s
[EMAIL PROTECTED] root]# time sa-learn --dump magic
<snip>
real 0m11.954s
user 0m5.270s
sys 0m0.160s
Notes:
I do have a larger bayes token count than default, 250k, but under 2.64 I was
running even more with 300k tokens. I tried switching to using 250 to see if I
could speed up expiry a bit.
bayes_expiry_max_db_size 250000
#bayes_expiry_max_db_size 300000
The box all this is running on is a Intel celeron 2ghz with 512mb of ram.
The box doesn't look to be resource starved. Swap usage is very mild and there's
more free ram than swap used. Load averages and CPU utilizations all seem quite
reasonable.
13:48:39 up 42 days, 6:57, 1 user, load average: 0.15, 0.50, 0.56
81 processes: 79 sleeping, 2 running, 0 zombie, 0 stopped
CPU states: 12.8% user 2.3% system 0.0% nice 0.0% iowait 84.7% idle
Mem: 514544k av, 442556k used, 71988k free, 0k shrd, 50976k buff
120436k active, 299564k inactive
Swap: 1052216k av, 42312k used, 1009904k free 85552k cached
In the past 24 hours mrtg reports load average statistics of:
Max Load: 1.2
Average Load: 0.3
Current Load: 0.5
And CPU utilization:
Max CPU Utilization (total): 61.0 percent
Average CPU Utilization (total): 16.0 percent
Current CPU Utilization (total): 19.0 percent
And ram:
Max RAM used including cache: 485.0 M RAM Used
Average RAM used including cache: 438.0 M RAM Used
Current RAM used including cache: 419.0 M RAM Used
Max RAM used excluding cache: 308.0 M RAM Used
Average RAM used excluding cache: 265.0 M RAM Used
Current RAM used excluding cache: 265.0 M RAM Used
Anyone have any suggestions for tuning bayes on a simple single-box single
sitewide-bayes setup (using MailScanner with no bayes_path statements, so all
bayes is in the homedir of the user specified by MailScanner's "Run As User"
option.)
