[ 
https://issues.apache.org/jira/browse/CASSANDRA-7510?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14056228#comment-14056228
 ] 

Brandon Williams commented on CASSANDRA-7510:
---------------------------------------------

bq. Are you sure that it means "I'm ready to receive client connections"? My 
reading of the code is that STATUS_NORMAL is sent as part of SS.joinTokenRing 
which is itself part of SS.initServer. The latter being called in 
CassandraDaemon.setup before we start the thrift and native protocol servers.

If the node needs to bootstrap, it will send JOINING because joinTokenRing will 
call bootstrap which sets it.  It won't send NORMAL until bootstrap finishes 
and then it calls setTokens.  Now, this is all before it returns control from 
initServer back to CassandraDaemon.setup, so there is a *very* small chance it 
could gossip it to another node before the rpc servers are started, but the 
only way around that is giving control of rpc startup to SS which feels ugly.

> Notify clients that bootstrap is finished over binary protocol
> --------------------------------------------------------------
>
>                 Key: CASSANDRA-7510
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-7510
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Joost Reuzel
>            Priority: Minor
>             Fix For: 2.0.10
>
>         Attachments: 7510.txt
>
>
> Currently, Cassandra will notify clients when a new node is added to a 
> cluster. However, that node is typically not usable yet. It first needs to 
> gossip its key range and finish loading all its assigned data before it 
> allows clients to connect. Depending on the amount of data this may take 
> quite a while. The clients in the mean time have no clue about the bootstrap 
> status of that node. The only thing they can do is periodically check if it 
> will accept a connection. 
> My proposal would be to send an additional UP event when the bootstrap is 
> done, this allows clients to mark the node initially as down/unavailable and 
> simply wait for the UP event to arrive.
> Kind regards,
> Joost



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Reply via email to