Description of problem: Test system has IPv6 address from auto-configuration, there is a DNS AAAA record for LC, but LC itself can't be reached over IPv6. Every new request (log upload, etc.) from restraint to LC now has ~60 second delay before it falls back to IPv4: ... [pid 5283] 04:07:47 connect(7, {sa_family=AF_INET6, sin6_port=htons(8000), inet_pton(AF_INET6, "2620:52:0:aa1:216:3eff:fe60:cd5c", &sin6_addr), sin6_flowinfo=0, sin6_scope_id=0}, 28) = -1 EINPROGRESS (Operation now in progress) [pid 5283] 04:07:47 eventfd2(0, EFD_CLOEXEC|EFD_NONBLOCK) = 8 [pid 5283] 04:07:47 write(3, "\0\0\0\0\0\0\0\1", 8) = 8 [pid 5283] 04:07:47 write(3, "\0\0\0\0\0\0\0\1", 8) = 8 [pid 5283] 04:07:47 poll([{fd=3, events=POLLIN}, {fd=4, events=POLLIN}, {fd=5, events=POLLIN}, {fd=7, events=POLLOUT}, {fd=8, events=POLLIN}], 5, 60000) = 1 ([{fd=3, revents=POLLIN}]) [pid 5283] 04:07:47 read(3, "\0\0\0\0\0\0\0\3", 16) = 8 [pid 5283] 04:07:47 poll([{fd=3, events=POLLIN}, {fd=4, events=POLLIN}, {fd=5, events=POLLIN}, {fd=7, events=POLLOUT}, {fd=8, events=POLLIN}], 5, 60000 <unfinished ...> ... [pid 5283] 04:08:47 <... poll resumed> ) = 0 (Timeout) Version-Release number of selected component (if applicable): restraint-0.1.30-1.el7_2 How reproducible: high Actual results: Delays combined doesn't allow /distribution/install task to complete and system hits EWD. Expected results: If IPv6 can't be used, restraint gives up on re-trying it in every request. Additional info:
Jan, Thank you :) Best, -pbunyan
I think it's reasonable for restraint to try IPv6 on every request, I mean there is an AAAA record and a route, it just doesn't work. Ideally we would treat a problem like this as a critical network failure which brings down the whole lab, and just get the network configuration fixed up. If we had the same problem on IPv4 it would be treated that way...
I guess another possible workaround is to disable IPv6 on the system in kickstart %post, until that particular lab network problem is fixed.
Using beah or kernel command line option to disable IPv6 as workaround is also possible. Artem also pointed out, that this would be difficult to achieve in restraint, because some log/result upload is running in separate processes. I'm closing this and we'll follow up with lab admins.