It stops accepting writes immediately. New requests won't be sent to the node, but existing in-flight queries should be completed. There could be some client exceptions for coordinated queries, but if you have speculative execution set up on your clients these cases should be covered.
Consistency ALL queries will fail for any query that hits a replica on the down node. On 23 November 2017 at 06:12, Daniel Woo <daniel.y....@gmail.com> wrote: > Hi guys, > > I have two questions about the drain status. > > The doc says when a node is drained it stops accepting new requests and > flush memtable to disk. Does it stops accepting new requests immediately or > wait for the majority of the nodes to gossip about the DRAINED status? If > it stops accepting new requests immediately there could be some noisy > exceptions to the client, right? > > After a node is drained, it might take 10 minutes to upgrade and restart > it, during that 10 minutes, what happens to the read and write requests if > the consistency level is ALL? > > -- > Thanks & Regards, > Daniel >
