It recovered right away, bounced again at 12:05 UTC, and appears to be OK
for the moment. This appears to be the same issue we found in March: Network operations is looking into this issue.

Be sure that your streaming client looks for keep-alives. If neither data
nor keep-alive is detected for about 90 seconds -- 3 keep-alive cycles,
assume that the connection has been left high-and-dry and reconnect. During
normal operations, if your connection is dropped, you'll receive a TCP CLOSE
or TCP RESET. When this unusual load balancer situation develops, existing
connections are just ignored. Load balancers are wonderful, until the
instant they are awful.

-John Kalucki
Infrastructure, Twitter Inc.

Reply via email to