No, for reference see here: https://github.com/openwrt/packages/commit/be848a000ef6b09da3385849c79fc7c200aeef0c#diff-cff4d2df69efe7bc5b0f9ececd37f5d2
Yes, I've also seen this on one of my 64MB/32MB routers with ~40K domains. The problem I found is not with average memory usage, but transient usage. Using many adblock domains bloats dnsmasq and, since dnsmasq forks to handle each TCP request, this seems to consume more than usual memory (even with sharing) sometimes. If already low on memory, it's then easy to trigger the OOM reaper e.g. try opkg update. One option is to switch to ipset based blocking e.g. banIP.
Would you say the problem is new? For me it seems like it is but maybe I was just on the limit of what it could handle.
@hnyman Sorry for offtopic. 
Just trying to bring some closure to the tangent... And no, none of the OOM / dnsmasq / opkg / etc behaviour is new, just that it's a question of a safe and reliable operating point. But compared to my borderline 40K domains, your 800K seems far beyond 
.
Thank you. The conclusion I make is that something in OpenWRT might use a little bit more memory now so I hit the limit. I do have about 220mb mem free, I thought that was enough...
I only have one dnsmasq instance up tho', I see no forks, unless they close after a while.
I realize this is the OpenWrt forum but still:
Firmware:[ DD-WRT v3.0-r37495M kongat (10/28/18)]
Time: 02:02:05 up 13 days, 18:24, load average: 0.84, 0.64, 0.37
Wireless Packet Info
Received (RX) 136342904 OK, no error
Transmitted (TX) 192206497 OK, no error
(I do realize a small amount of TX/RX errors is normal, but I was kinda surprised there were none at all).
For reference I use a R7000 and EA6300v2 as wireless media bridges (with Advanced Tomato and DD-WRT), plus various iphones, ipads, androids, usb wireless clients, laptops and desktop machines with internal WLAN cards and friends over as well with their mobile phones.
I'll stfu now 
2 Likes
That's pretty cool. I am keeping that Kong build just in case I start using this router for WiFi. Wonder if DD-WRT is stabler than OWRT when you enable their version of NAT acceleration (SFE?). My OWRT build can hang after 6 days of that enabled. Sometimes.
What ath10k driver does DDWRT use that is stable compared to the one in OpenWRT?
Idk about DDWRT but afaik there are no other drivers than the ones that exists in linux by default, you can toy with the official firmwares in openwrt and see if you find one that's works for you.
The non-CT driver/fw should be quite stable in openwrt now tho'. I'm currently not experiencing issues with CT.
Even with the non-ct driver I still see crashes in dmesg 
r9627: dnsmasq keeps starting even though I have the service disabled. Any ideas what could be causing it? What is strange is that I'm stopping it and disabling in rc.local but the process is still there after a reboot. There is no dnsmasq in /etc/rc.d, so something else must be causing it to start.
Luci is also very slow, it just sits there spinning. If I restart uhttpd it's fine for a while but then back to being painfully slow.
Anyone else seeing this problem? Seems to happen in Chrome, Firefox looks OK. I tried a clean Chrome profile and same problem. Not sure if it's related to this change:
Another problem:
I installed bind-dig opkg, but can't run dig, any ideas?
# dig
Error relocating /usr/lib/libdns.so.1207: ENGINE_get_id: symbol not found
Error relocating /usr/lib/libdns.so.1207: ENGINE_load_private_key: symbol not found
Error relocating /usr/lib/libdns.so.1207: ENGINE_load_builtin_engines: symbol not found
Error relocating /usr/lib/libdns.so.1207: ENGINE_set_default: symbol not found
Error relocating /usr/lib/libdns.so.1207: ENGINE_free: symbol not found
Error relocating /usr/lib/libdns.so.1207: ENGINE_load_public_key: symbol not found
Error relocating /usr/lib/libdns.so.1207: ENGINE_by_id: symbol not found
Could be openssl 1.1.1 misbehaving with the engine config option. Apparently some packages have been compiled with engine support dependency while openssl itself is not.
Thank you. Hopefully it gets fixed in future pkg builds.
Any ideas for the slow Luci UI in Chrome? Not only slow, many times request never finish, they just time out, rendering the UI unusable in Chrome.
Time for a solution update on the issues I've been facing:
- Trying to disbable dnsmasq service but it was still running: turns out that calling /etc/init.d/dnsmasq stop and disable from /etc/rc.local at every boot causes this, dnsmasq starts running even though it was previously disabled and stopped. Not sure whether it's a bug in the init scripts. I just have to remember to do it once after a firmware upgrade as I don't know how else I can automate it short of building my own firmware.
- Slow https UI/requests time out in Chrome only, but OK in Firefox, or plain http OK too: I was tired of accepting the self-signed TLS cert every time and didn't want to add it as an exception, so I decided to have my local CA and generate my own cert signed by it. I used this useful tool: https://github.com/FiloSottile/mkcert
Now no more warnings from browsers and to my surprise it also fixed the Chrome issues, though I didn't think it would. There must have been something about that self-signed cert generated by OpenWRT that Chrome didn't like.
hello,
i can't get nginx to start with this build.
did i miss something?
the service is not starting but not giving any error, so i tried manually and i got this:
root@RUTTO:/etc/nginx# /usr/sbin/nginx -c /etc/nginx/nginx.conf
Error relocating /usr/sbin/nginx: ENGINE_load_private_key: symbol not found
Error relocating /usr/sbin/nginx: ENGINE_set_default: symbol not found
Error relocating /usr/sbin/nginx: ENGINE_free: symbol not found
Error relocating /usr/sbin/nginx: ENGINE_by_id: symbol not foundthanks
edit to add: well, nginx with ssl, is it the problem?
openssl needs to be compiled with engine support
@ansuel
Do you see that as a hard new requirement for nginx, or does that only happen because some openssl in buildbot is now complied with engine support and nginx compliation detects it.
(There are some packages that currently erroneously force engine support to be enabled in openssl in buildbot, although engine support is not on by default. That likely leads also nginx to be compiled with engine support in buildbot, although that does not match the intended openssl defaults). See my analysis in
https://github.com/openwrt/packages/pull/8341#issuecomment-478322454
i have r7800 and as it doesn't have any hardware crypto i also disabled engine support for openssl
Got the same problem with nginx crashing on boot. By enabling engine support nginx starts to work again.
will check if nginx get compiled with engine support.
i suppose this is something i can only fix building the image by myself..
am i wrong?