Cluster retry of pipeline commands #326
Description
Currently when some pipeline commands fail because of a TRYAGAIN, MOVE, ASK error. Only the commands that receive the errors are run again. This can cause issues eg a successful readonly command that comes after a failed write command using the same key. There are also situations where a write command that follows another write command might need to be run again if they affect the same keys, but then in theory they should both receive the same error.
To get this correct we should probably store the keys affected by commands that failed and then retry any commands that follow which also use those keys.
@nilanshu-sharma what are your thoughts?