summaryrefslogtreecommitdiff
path: root/README.md
diff options
context:
space:
mode:
authorjfesler <jfesler@gigo.com>2020-06-28 21:08:45 -0700
committerjfesler <jfesler@gigo.com>2020-06-28 21:08:45 -0700
commitf14f6334fda7b78191c9ab23d2bdef2bf5597860 (patch)
tree2ead612f59cbb5ac2225ed88de1bc86907f1e8be /README.md
parentbf4b3fe21e71cdbf8e8bdad77ec1c427490b53c4 (diff)
downloadmtu1280d-master.tar.gz
mtu1280d-master.zip
experimental code to watchdog and reset nfqueueHEADmaster
Diffstat (limited to 'README.md')
-rw-r--r--README.md21
1 files changed, 21 insertions, 0 deletions
diff --git a/README.md b/README.md
index d8c9b1c..2598334 100644
--- a/README.md
+++ b/README.md
@@ -41,6 +41,27 @@ The `preferred_lft 0` is important to mark the address as a deprecated address.
This means only use the address for incoming connections; not for outgoing.
+UBUNTU 18 NOTES ON NFQUEUE HANGS
+--------------------------------
+
+We're seeing reports of the daemon wedging. So far, my observations
+on my own ubuntu 18 system are that the recv() calls against the
+iptables nfqueue hang.
+
+The master branch (not pushed to the rsync server) specifically
+adds in a watchdog function; after a configurable numbrer of seconds,
+it will disconnect the nfqueue and reattach. If it does this
+too many times, it will abort.
+
+You can tune this with these options:
+
+ -w 60 - How long we can go without seeing a packet
+ -W 1440 - How many times we can reset the socket without seeing a pocket
+
+For most of you, I'm monitoring your web sites. At minimum I should
+be hitting your mirror once every 30 minutes; somehow you should
+see and accept traffic in the time above (1 day!).
+
REQUIREMENTS
------------