[Babel-users] Race condition in babeld?

[Juliusz Chroboczek]

Hi,

The author of the attached bug report wishes to remain anonymous.  I'll
puzzle over it later, in the meantime I'm sharing it with the list (with
his permission).

-- Juliusz

I am a bit new to babel but I am having an odd problem that looks like
a race condition in babel.  I am running babel-1.3.0 under Linux.

When I am at the edge of wifi range of even a 2 link (point to point)
mesh sometimes I permanently lose connection from NodeA to NodeB. When
I monitor the kernel route table on Node A I see that the host route is
permanently banned (!H). If I manually delete the route then I instantly
regain a connection.

I trying to determine the catalyst for the problem I have discovered the
following timing causes it to happen.  It looks like the internal babel
route status and the kernel routing table can get out of sync when Node
B regains comms while NodeA is in the process of banning the route.

I have 3 scenarios. I have found 2 Work and 1 does not

Scenario 1 expected Good behavior: T0: Node A loses wifi connection to
Node B ~8seconds T1: Node A host route gets marked as banned ~8seconds
T2: Node A host route gets deleted from route table ~ arbitrary time
later T3: Node B comes back into range ~1-2 seconds T4: route added and
traffic passes


Scenario 2 expected Good behavior: T0: Node A loses wifi connection to
Node B ~8 seconds T1: Node A host route gets marked as banned ~5seconds
T2: Node B comes back into range ~1-2 seconds T3: route added and
traffic passes

Scenario 3 unexpected bad behavior: T0: Node A loses wifi connection to
Node B ~8 seconds T1: Node A host route gets marked as banned T2: Node
B comes back into range banned host route never gets deleted.