mightyhorst / fabfi Goto Github PK

Add option for configuring admin wlan in fabfi setup for headnode (wrt160nl)

Running squid on the headnode causes it to reboot frequently.  Suspect 
insufficient memory to start a new process, so cron jobs are bouncing the 
device.  Disabling squid fixes.  Maybe need swap?

we get in invalid argument error when formatting the disk during startup, 
sometimes it hangs, sometimes it doesn't.  it's possible that if you're 
inserting the drive during the setup you need to be sure to wait until it's 
fully recognised by the OS

I'm not sure Exactly how to reproduce this, but here's the setup I'm seeing a 
problem with:

I have a headnode with a wireless WAN.  The IP and DNS settings for this node 
are static.  

When there are multiple headnodes on the net, OLSR completely overwrites 
/var/resolv.conf.auto with the remote DNS servers.  The EXPECTED behavior is 
that olsr then runs dnsmasq and the locally defined DNS should be appended.  
This doesn't work.  

The problem is fixed in real time by running /etc/init.d/network restart 

maybe this should run every time OLSR updates DNS?  Seems like doing surgery 
with a sledgehammer.

If a Picostation is on and loses its route to the net (radius server?) it will 
begin blinking the topmost LED and will not associate clients until the route 
returns.  Assume this is a symptom of chilli, Since we don't have any resources 
that would e available when there's no route to the radius server, we're ok 
with this issue for now.

to reproduce:
1. start up the device with no route to net OR
2. start up with route to net and then remove that route

when route not available, LED will blink and it will not associate clients.  

cache size on squid is statically set to 5000GB, for larger flash drives' it 
would be good to be able to select this dynamically

Configure headnode and map.mesh is defined in /etc/config/dhcp as nodeIP not 
cloudIP.  Could be old config file?  was unable to find the location of the 
problem in setup script or source files.  

Can be manually fixed by editing config (i think)

the setup script sets the wlan to 1 in the chilli config for pico2M when 
selecting Client AP mode.  It should be wlan0

What steps will reproduce the problem?
1. set up a headnode where DNS for master.mesh and radius.mesh are not the same
2. set up a second node with chilli running

You will see the splash page but the login will not complete 

Note: this is not entirely verified.  Could have been a reboot / dns caching 
issue.

As expected the check-uplink script incorrectly handles the case where the net 
is available but ping fails.  This can be corrected by adding:


echo "`date` trying wget as a backup" >> $LG
wget -s http://master.mesh
if [ $? -eq 0 ]
then
echo "`date` w-got master.mesh" >> $LG
set_server
exit 0
fi

but the olsrd_dyn_gw plugin had to be nastily hacked to check the local 
address, which breaks truly dynamic gateways

This is not a bug , rather a feature i am asking about.

I just started working for some small project of mine.
I was able to do some olsr testing with some 20 EUR routers with openwrt.

Since your platform is so well maintained i was loking into it. 

I tried to build image for a different Ar 7xx chipsets and after compiling your 
trunk i realized that the images generated are limited to Ubiquti .

There are several other router running openwrt nicely they are cheap and are 
based on atheros chipset, why is that you cannot build for different platforms 
. It would be nice for the community if one could build for every 7xx platform 
that openwrt supports.

For instance I am playing with TPlink 741ND . It works well but i cannot use 
fabfi even the chipsets are close to nanaostation.

The only reason i was not using picostation is that the regulation here does 
not allow more than 100mW RF output end even we are developing country no one 
likes to mess up with the allready crowded spectrum.

Besides that this router si small works fine with openwrt (75 mbps N 20 Mhz) 
and is dirt cheap 20 EUR / 25 USD .

Is there any script to build different platforms ? Did anybody try that .

If there is such project i am willing to test , i have serial port hooked to my 
router so I am relatively safe.

Thank you. 

What steps will reproduce the problem?
1. try to create a headnode, configuring squid for the first time
2. umount will always fail (ok) but only sometimes hang (not ok)

you have to restart the setup script and then it generally works

Entering decimal GPS coordinates (ex -.75, 36.5) causes updates to fail on 
server side.  all decimal GPS coordinates needs to be entered as 0.xxx

When a headnode is configured with STA on the radio, the adhocmesh entry is 
left without an interface.  This may break VLAN config, and should be 
investigated.

even if it doesn't break VLAN, extra sections should be removed from network 
file

Backfire seems to have cleaned up time syncing so that some of our manual hacks 
with rdate may no longer be needed, though they don't seem to be interfering 
now.  Should investigate what the minimum config is.

openwrt doesn't build nagios-plugins out of the box in 25475 because it doesn't 
show correctly in menuconfig.  You have to manually add libpq and 
nagios-plugins in the .config file.  Note version has changed:
https://dev.openwrt.org/changeset/17923/packages/net/nagios-plugins/Makefile

It seems that selecting coreutils-su fixes.  

Added as a bug here because it seems like there are all of a sudden a lot of 
dependencies.  Not sure in nagios-pplugins will work in this version

builds from r244 will have the new version

With links running in the red zone (< -75dBm), STA will frequently disconnect 
and not come back.  running wifi on the AP seems to fix, but a permanent fix is 
needed.

Suggested test: switch to ADHOC

To Reproduce:
1. browse to the network map using the aliased url map.mesh

The browser will pop up a message about the map needing a new key, but it still 
seems to work.  

some might call this a feature, but if you're logged in as a LAN client, you 
can't directly access anything on 10.0.0.0/8 (am assuming default GW is 
excepted from this since the system works...

map.mesh should be configured to the cloudIP by the startup script, but is 
configured to the local IP instead.  

Not sure how to reproduce, but sometimes the date sits in 1970 forever on Mt. 
View HN.  must test with new build of headnode

while developing failover functionality for headnodes, I noticed that when the 
default route switches (maybe just when the interface switches), chilli locks 
up and needs to be restarted

currently, nodes with gateways do not advertise HNA for the networks they're 
connected to. it's a good idea for this to not be automatic in the case of most 
local networks, but if a node has a global IP, it's better for traffic to 
traverse the mesh than to go to the public internet.

A useful enhancement would be to have gateways automatically advertise the 
public networks they use as HNA.