Akregator causing hundreds of TIME WAIT to same IP? DNS?

[chconnor]

Hi -- for a few days now I've been experiencing intermittent network dropouts that last a minute or so.

When it happens, the router interface (dd-wrt on an ancient linksys) shows that the IP Filter Maximum Ports value (1024) is being reached. Inspection of the IP connections table during these periods shows many hundreds of identical TIME_WAIT state TCP connections.

The IPs reverse DNS to google (1e100.net). Since I use 75.75.75.75 and 75.75.76.76, I'm guessing this is all DNS related? Example IPs: 172.217.3.211 and 216.58.195.83

I have traced the immediate cause to Akregator checking its feeds (I have ~103 total feeds, 621 unread items). I haven't changed anything recently in Akregator -- the same collection of feeds has worked fine with this computer/router for years. I did recently upgrade to 18.10 from 18.04, but that was a couple weeks ago and I've only noticed the issue in the last couple of days. Might be that it started with the upgrade, though.

Does anyone know if Akregator has changed anything in how it handles lookups recently?

[jlittle]

Hi -- for a few days now I've been experiencing intermittent network dropouts that last a minute or so.

I have had similar dropouts for months, irregularly, usually not much of a problem but occasionally a nuisance. Disconnecting. wait 10 s, reconnect, usually gets me along till the next dropout.

When it happens, the router interface (dd-wrt on an ancient linksys) shows that the IP Filter Maximum Ports value (1024) is being reached. Inspection of the IP connections table during these periods shows many hundreds of identical TIME_WAIT state TCP connections.

That's greek to me. I have a telco supplied, locked down, router and no connection table visible. Is it possible to see that state from Kubuntu?

I have traced the immediate cause to Akregator checking its feeds...

Is it possible to briefly explain how you know that? I was surprised to find that akregator is running on my system as I've never used it, and it just has the feeds of a standard install.

I had been thinking that my trouble is a driver issue. I've installed the latest driver for my Intel NIC, "e1000e", and this has lessened the duration of the dropouts.

[chconnor]

I guess I'd be a little surprised if it was exactly the same issue causing your problems... I'm using a very old router. I have a feeling something modern wouldn't struggle like this. But I have had issues since 18.10, hence my suspicions.

That's greek to me. I have a telco supplied, locked down, router and no connection table visible. Is it possible to see that state from Kubuntu?

No, it's on the router side; you'd need some kind of access to a status page or something. The Kubuntu side (for me anyway) doesn't show many TIME_WAIT connections. Perhaps the socket state on the Kubuntu side gets turned over faster, or I have a faster expiration time, or maybe the expiration from the IP tables is slower than what netsat shows on the Kubuntu side for open sockets. I'm definitely not a protocol expert, but I have seen this kind of thing a lot in back-end servers I've worked with that handle lots of connections. It's something that sysadmins deal with all the time: OS resource exhaustion in high concurrency situations. It's just usually not an issue with a home user running Akregator. :-)

Given that I have 103 feeds, it's weird that a refresh would hit DNS 1000 times (assuming that's what is happening.)

Is it possible to briefly explain how you know that? I was surprised to find that akregator is running on my system as I've never used it, and it just has the feeds of a standard install.

Huh -- it's autostarting or something? I wonder if Akregator -> Settings -> Configure Akregator -> General -> "Show tray icon" causes it to do that. (I run it manually.)

Anyway: I believe to be the case because if Akregator is not running the issue doesn't happen and the current open connections shown on the router hovers around 20 to 80. As soon as I open Akregator it climbs into the many hundreds and the numerous TIME_WAIT's all to the same (or similar) IPs show up. Akregator is set to refresh the feeds every 10 minutes, and that's about how often I see the issue crop up. Triggering a refresh of the feeds can drive the count to 1024, causing the dropout until the TIME_WAITs expire. I can also watch (via netstat) on the Kubuntu side the Akregator process opening connections to those IPs.

If I were you I'd start by quitting Akregator and seeing if the issue goes away, but again, it'd be lucky (though very interesting) if it was exactly the same issue I'm having.

You could also install and run iftop (via sudo) if you want an easy way to watch what your local connections are doing. Or as super user

while (true); do { clear; netstat -antp; sleep 1; } done

(have to install netstat first i think)

[chconnor]

I filed a bug here: https://bugs.kde.org/show_bug.cgi?id=401850

Changed my router config to use 4096 entries in the IP table and it doesn't freeze up any more. Akregator caused about 1200 of those to be used up on a refresh, though.