Maybe because when you kill your cardano node process, this doesn’t kill the established socket connection that the database uses, and so to “reset it” and let your system know that a new socket connection needs to be created because the db is using the old one, so in order to run a new instance of cardano node, maybe you have to ‘kill the socket connection’, and export the “path socket” again because the system thinks it doesn’t exist because it’s still in use even though you shut down the node,or it could be that if you wait too long to restart your node, maybe there is a socket timeout, and if you don’t restart it soon enough, the cardano-cli or system will treat the socket as being open when it actually isn’t…I don’t know anything about it…I’m just taking a wild stab…what else would explain the bizarre intermittently working behavior of the “export path” command for the socket when you have already written it to .bashrc and ‘sourced’ it? I’m just taking a wild guess…maybe this weird behavior can be pinpointed by using a combination of commands like netstat -a | grep -i socket, echo $CARDANO_NODE_SOCKET_PATH, etc.
I didn’t bother trying to pinpoint to figure out why it fails sometimes, I just make a short alias to circumvent this ‘headache’ whenever it appears. Yes, it appears to be a persistent problem for many users.