On 03/03/2020 04:06 PM, Mike Christie wrote: > On 03/03/2020 03:13 PM, Josef Bacik wrote: >> On 2/28/20 1:40 AM, Hou Pu wrote: >>> In commit 2da22da5734 (nbd: fix zero cmd timeout handling v2), >>> it is allowed to reset timer when it fires if tag_set.timeout >>> is set to zero. If the server is shutdown and a new socket >>> is reconfigured, the request should be requeued to be processed by >>> new server instead of waiting for response from the old one. >>> >>> Signed-off-by: Hou Pu <houpu@xxxxxxxxxxxxx> >> >> I'm confused by this, if we get here we've already timed out and >> requeued once right? Why do we need to requeue again? Thanks, >> > > We may not have timed out already. If the tag_set.timeout=0, then the > block timer will fire every 30 seconds. This could be the first time the > timer has fired. If it has fired multiple times already then it still > would not have been requeued because the num_connections=1 code just > does a BLK_EH_RESET_TIMER when timeout=0 and does not have support for > detecting reconnects. > > In this second patch if timeout=0 and num_connections=1 we restart the > command when the command timer fires and we detect a new connection > (nsock->cookie has incremented). > > I was saying in the last patch, maybe waiting for reconnect is wrong. > Does a cmd timeout=0 mean to wait for a reconnect or in this patch > should we do: > > 1. if timeout=0, num_connections=1, and the cmd timer fires and the > conneciton is marked dead then requeue the command. > 2. we then rely on the dead_conn_timeout code to decide how long to wait > for a reconnect. > Oh yeah, I had thought Hou implemented timeout=0 to wait for a reconnect to handle existing apps. However, I am not sure if they exist. When we had timeout=0 support the first time then we did not have multi conn and reconnect support yet. The current timeout=0 and reconnect support does not work since that is what Hou is implementing, so we can decide the behavior now.