It's almost like some part responsible for this just collapses. This is the interesting scenario - you can see ESP coming in on the docker host (not container) while pinging, but it will not send ESP back (or originate if you try to originate only). Ex: in this case where it just happened, as soon as I rebooted the leader, it started working again. I have noticed a few times that the issue seems to be surrounding the leader node. Once it hits this state, I do open it from all the sides, and test to see which way exactly it fails. As soon as I stop getting "replies", that indicates a problem.
If I get both ICMP echo + reply, the overlay works. I have one container on each node (when I test), but I've noticed that since I first test with just ICMP (easiest), it won't matter which way I am initializing it. I can start monitoring this more carefully. I have seen it as soon as an hour, and the "average" (if you can even call it that) seems to be a few hours to less than a couple of days. For example, on the latest cluster it worked for a few days (given no containers on it) before it hit this status again. Systems are located in 3 different regions, on 3 different public IP seems "random". We have changed it around to be only virtual and only physical - same results. ID: GVD4:VFPH:ELAN:X2CK:CLFZ:MFDC:C5LT:RLTU:DWKE:KDKY:HT6M:BAC2Īdditional environment details (AWS, VirtualBox, physical, etc.):Įnvironment is between physical and virtual systems. Network: bridge host macvlan null overlayĬontainerd version: 03e5862ec0d8d3b3f750e19fca3ee367e13c090e
0 kommentar(er)
