On Feb 12, 2013, at 18:58 , aaron morton <aa...@thelastpickle.com> wrote:
> Just checking if this sorted it's self out?
Well, partially. :) Situation at the moment:
>> 1) Is the new node not showing up in the "nodetool status" expected behavior?
Not showing still, don't know if should. :)
>> 2) Do I have to wait for all the data streaming to end until the new node
>> starts to accept writes? If yes, has this changed from 1.1? (I remember
>> nodes becoming writeable immediately after booting, but readable only after
>> streaming has ended)
This does not seem to be the case. Writes are indeed relayed to the
bootstrapping node during the bootstrapping even if it is not shown as a part
of the cluster.
>> 3) "nodetool info" on the new node raises an exception "Exception in thread
>> "main" java.lang.AssertionError at
>> org.apache.cassandra.locator.TokenMetadata.getTokens(TokenMetadata.java:465)"
>> This is probably not normal?
Still an issue. Should probably file a bug on this one?
>> 4) Is it so that the virtual tokens for the new node are not assigned before
>> the data streaming ends? ("select * from peers;" on .11 returns null for the
>> tokens of .14) If so, how the existing nodes decide what data to stream to a
>> new node?
By reading the code it seems that the tokens are indeed generated early in the
bootstrapping sequence and propagated (via gossip?) to the other nodes.
However, I haven't still figured out why they are not shown in the list of
peers when querying using CQL.
FYI.. While playing with this I bumbed into a bug that prevents compactions
from running while a node is in the bootstrapping state:
https://issues.apache.org/jira/browse/CASSANDRA-5244 Might be an issue at least
for those running clusters with lots of data per node (= long bootstrap times)
and heavy write load (= lots of data to compacted).
Cheers,
-Jouni
