From Newsgroup: news.software.nntp
Il 21/07/2026 23:45, Ivo Gandolfo ha scritto:
Hi all,
Hi all again,
a short update following the Postfilter-NG announcement.
The current release candidate is:
Postfilter-NG 2026.07.5-rc3
Repository and releases:
https://github.com/infybofh/postfilter-ng
According to GitHub traffic statistics, the repository has currently
received:
33 clones
27 unique cloners
116 page views
I know that clone statistics may also include automated systems, but the project has clearly been downloaded and inspected by more than just one
or two people.
The slightly less encouraging part is that I have received almost few
feedback :-)
Meanwhile, Postfilter-NG has been running on a real public INN reader,
and I collected a complete day of verbose logs for debug
Results from 1172 submitted articles:
Accepted by Postfilter-NG: 1146 (77.8%)
Rejected by Postfilter-NG: 26 (22.2%)
Accepted by Postfilter but later
rejected by Cleanfeed-NG: 5 (6.9%)
Finally accepted by INN: 1141 (70.8%)
Postfilter-NG rejection reasons:
7 forbidden crossposts
4 multipost detections
3 cumulative group quota exceeded
4 cumulative ip quota exceeded
2 permanently closed group for posting (too many abuse)
1 oversized text article
2 binary on text-only server/group
3 bad MIME/type encoding
No crashes, configuration fallbacks, SQLite errors, SQLITE_BUSY events,
stack traces or internal failures were observed.
Overall processing times:
Median: 285 ms
Average: 466 ms
95th percentile: 988 ms
Maximum: 1.010 s
Accepted articles:
Median: 304 ms
Average: 525 ms
95th percentile: 785 ms
Maximum: 1.010 s
Rejected articles:
Median: 228 ms
Average: 260 ms
95th percentile: 513 ms
Maximum: 695 ms
Median timing of the main checks:
style: 46.2 ms
attachments: 0.34 ms
binary: 0.03 ms
user database: 0.01 ms
Tor lookup: 47.9 ms
RBL lookup: 18.0 ms
banlist: 0.94 ms
badwords: 0.25 ms
rate limiting: 0.96 ms
custom checks: 0.01 ms
The slowest occasional operations were external DNS lookups,
particularly URIBL queries (if the post contain a lot of url). The MIME attachment scanner itself remained below approximately 1 ms in this
workload.
SQLite rate counters also worked correctly across multiple independent
nnrpd processes. Multipost and cumulative posting limits were enforced
using shared persistent state, not only per-process memory.
Based on these results, rc3 also includes:
- configuration-generation deduplication and retention;
- a one-second total DNS lookup budget;
- a 2700 ms processing budget with fail-closed behaviour;
- additional timing information for configuration, profiles, headers,
database recording and finalization;
- improved Tor-marking logs;
- further installer privilege-drop hardening.
The current results are encouraging, but my own server is still only one installation. My goal it's to reach 95th percentile < 300ms to analize
the post and accept/reject.
I would therefore very much appreciate feedback from anyone who has
cloned, installed, reviewed or even only tested the configuration.
Useful feedback does not need to be a detailed technical report. Even something as simple as:
INN version:
Operating system:
Perl version:
Postfilter-NG mode: audit / enforce
Installed successfully: yes / no
False positives observed: yes / no
General result: works / does not work / still testing
would be genuinely useful.
Reports about unusual MIME messages, PGP or S/MIME posts, binary
hierarchies, custom readers.conf configurations, performance, false
positives, migration from the original Postfilter, or installer problems
are especially welcome.
Issues can be opened directly on GitHub:
https://github.com/infybofh/postfilter-ng/issues
So, to the people behind those clones: please do not be shy. :-)
A one-line rCLinstalled and working on INN x.yrCY is already valuable. A bug report is even more valuable. Silence is the only benchmark I cannot
analyse.
Thank you in advance.
Sincerely
--
Ivo Gandolfo
bofh.team
--- Synchronet 3.22a-Linux NewsLink 1.2