The memory leak (we at least hope that it is the same one) has been reproduced - fix is coming soon.
21 Likes
Please take a look SUP-194457 , bugs with bgp vpn4 multihoming
1 Like
I bet in some issue related to contrack v4 or v6.
I've been looking for a way to collect the number of active connections (v4 and v6) for a long time.
I tried with scripts, but the CPU impact is very high when fetching the value.
/ip/firewall/connection/print count-only
1 Like
:put [/ip firewall/connection/tracking/get total-ip6-entries ]]
:put [/ip firewall/connection/tracking/get total-ip4-entries ]]
Those values do not seen to exist, but total-entries does.
(probably the sum of both?)
What's new in 7.20rc3 (2025-Sep-16 15:01):
...
*) firewall - added connection tracking "total-ip4-entries" and "total-ip6-entries" counters;
2 Likes
(is on Other changes since v7.19:)
Moved discussion on OpenWRT to Mikrotik fixes for OpenWRT in 7.19.6 and 7.20 - #12 by spippan from 7.19.6 topic. Not touching 7.20rc topic but if you want to discuss more then use the new topic please.
Why is this done?
I'm seeing a lot of blackhole routes now. How can I remove/filter or disable them?
What's new in 7.20rc4 (2025-Sep-23 14:38):
- bgp - fixed selection of received BGP VPN routes;
- bgp - introduced BGP instance configuration (note, downgrading to earlier versions without instance support may cause config issues) (additional fixes);
- console - fixed "special-login" (introduced in v7.20beta2);
- iot - added support for MQTT last will message;
- iot - fixed logic for unknown NetIDs;
- iot - fixed support for LoRa Alliance NetID list;
- lte - added passthrough support for RG650E-EU modem;
- sfp - fixed TX-Power monitor data output (introduced in v7.20beta9);
- system - fixed a potential memory leak when handling high-speed traffic (introduced in v7.20rc1);
1 Like
Could you please provide more details on this point?
- Related to conntrack?
- Related to IPv4, IPv6?
- Or perhaps other factors like IPFIX and hardware offload?
P.S.: I Upgraded 3 boxes of my own use... No special configs. Just home use.
5 minutes... Everything looks the same as in rc3 until now.
What does it mean?
The leak was introduced in 7.20rc1 with this change:
*) system - fixed stuck TCP transmit on virtual interfaces, leading to retransmits;
The memory leak was related to socket buffers and queuing under high traffic volumes. Not all devices experienced the leak, it seems to depend on the device CPU performance.
3 Likes
@EdPa
That one liner extended changelog explanation gives +100 to the communication.
7 Likes
We keep hoping that some time they find some time (e.g. an apprentice project) to implement a better changelog system as discussed before...
1 Like
I believe it is fix for buset1974 reported issue regarding bgp vpn4 multihoming. I do not know more specific details about that one, sorry.
I've run some tests which made my RB5009 run out of memory within a few attempts before: namely running several speedtest.net speedtests with multiple connections enabled. I've now run 5 and just by eyeballing the situation, the memory use seems to be stable. On the previous versions I'd be down to 100-200 MB of RAM at this point. So thanks for the fix!
3 Likes
Good news. I've run the same test that easily triggered a memory leak (iperf between two 1Gbps ports) continuously for two hours and memory usage remained stable.
4 Likes