-
Notifications
You must be signed in to change notification settings - Fork 678
Frequent failure in 110_encryption_test #1042
Comments
I tried raising the count of packets sent by As part of this exercise, at one point I had two ping commands - one that just printed out what it saw and one that was the original test. In every test I ran, either both succeeded or both failed. I also tried changing |
Looking at the logfiles from failed runs, we see that they fail to maintain UDP communication: Logfile from
Logfile from
Logfile from
|
I added some extra logging of heartbeats; we can see on this failed run that c2 never received any UDP heartbeats from c1: c1:
c2:
(note that the clocks on these two VMs were about 16 seconds out) |
Further investigation: I disabled encryption, but the intermittent failures continue. Packet-capture indicates that the UDP heartbeats are being sent to a high-numbered port, e.g. 54545, and then rejected.
for a working run, all UDP packets have source and destination ports of 6783, and
|
I wonder whether the issue is that the tests just blow away the weave
container instead of 'weave stop' ing. As a result the conntrack connection
state removal in the script does not happen. Which is there for a reason.
|
I tried adding this line before
but it still failed intermittently. |
Well, it failed on the 8th attempt, and then didn't fail again for the next 22 attempts, which represents an improvement. Perhaps that 1-in-30 failure was for a different reason. |
Interestingly, we were already calling |
@bboreham and @rade report seeing this in addition to myself. Root cause possibly the same as #1030.
The text was updated successfully, but these errors were encountered: