https://issues.apache.org/SpamAssassin/show_bug.cgi?id=5590

Quanah Gibson-Mount <qua...@zimbra.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |qua...@zimbra.com

--- Comment #47 from Quanah Gibson-Mount <qua...@zimbra.com> ---
This remains an issue with 3.4.0:

May  1 19:12:53 edge01 amavis[30846]: (30846-02) TIMING-SA [total 2274 ms, cpu
1962 ms] - parse: 2.7 (0.1%), extract_message_metadata: 18 (0.8%),
get_uri_detail_list: 17 (0.7%), tests_pri_-1000: 2.9 (0.1%), tests_pri_-950:
0.97 (0.0%), tests_pri_-900: 1.25 (0.1%), tests_pri_-400: 43 (1.9%),
check_bayes: 42 (1.8%), b_tokenize: 20 (0.9%), b_tok_get_all: 10 (0.4%),
b_comp_prob: 9 (0.4%), b_tok_touch_all: 0.22 (0.0%), b_finish: 0.65 (0.0%),
tests_pri_0: 2109 (92.7%), check_spf: 0.24 (0.0%), check_razor2: 156 (6.8%),
check_pyzor: 154 (6.8%), tests_pri_500: 2.5 (0.1%), learn: 80 (3.5%), b_learn:
69 (3.0%), b_count_change: 31 (1.3%), get_report: 0.69 (0.0%)

Total SA time was just over 2 seconds, and 92.7% (2109ms) was due to test_pri_0

Generally, what I see is:

May  1 19:24:12 edge01 amavis[30886]: (30886-13) TIMING-SA [total 423 ms, cpu
169 ms] - parse: 1.53 (0.4%), extract_message_metadata: 2.3 (0.5%),
get_uri_detail_list: 3.1 (0.7%), tests_pri_-1000: 3.8 (0.9%), tests_pri_-950:
1.48 (0.4%), tests_pri_-900: 1.68 (0.4%), tests_pri_-400: 17 (3.9%),
check_bayes: 15 (3.6%), b_tokenize: 4.7 (1.1%), b_tok_get_all: 4.3 (1.0%),
b_comp_prob: 3.7 (0.9%), b_tok_touch_all: 0.11 (0.0%), b_finish: 0.73 (0.2%),
tests_pri_0: 351 (83.1%), check_spf: 0.63 (0.1%), check_razor2: 143 (33.8%),
check_pyzor: 144 (34.0%), tests_pri_500: 2.8 (0.7%), learn: 33 (7.9%), b_learn:
29 (6.9%), b_count_change: 12 (2.7%), get_report: 0.49 (0.1%)

Even still, we can see that test_pri_0 is the majority of the time (83%,
351ms/423ms).

I.e., even for "fast" emails, test_pri_0 is the bottleneck.

-- 
You are receiving this mail because:
You are the assignee for the bug.

Reply via email to