[
https://issues.apache.org/jira/browse/IGNITE-4633?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15893977#comment-15893977
]
Vladimir Ozerov commented on IGNITE-4633:
-----------------------------------------
[~al.psc], I did some refactoring and simplification. Please see my changes.
Currently my only comment is how we determine set of participating nodes. It
should happen on client, not on coordinator, because in future client will
track alive nodes and complete operation with error if all participants have
left the grid. So we should make set of participating nodes final field and
initialize it on the client. Current {{nodeState}} field must be renamed to
{{nodeErrs}}.
By the way, do we really need to collect errors from all nodes? Will it be
enough to collect only the very first error and simply ignore init message on
further nodes?
> Initiate DDL operation through custom discovery message
> -------------------------------------------------------
>
> Key: IGNITE-4633
> URL: https://issues.apache.org/jira/browse/IGNITE-4633
> Project: Ignite
> Issue Type: Sub-task
> Components: SQL
> Reporter: Vladimir Ozerov
> Assignee: Alexander Paschenko
> Fix For: 2.0
>
>
> Design considerations:
> 1) Create generic DDL custom discovery message pair - {{INIT}} and {{ACK}}
> messages - which will hold concrete DDL commands.
> 2) Originator must generate unique message ID to be able to track it later.
> 3) Coordinator calculates all participants through affinity API and adds them
> to message.
> 4) {{INIT}} message goes through the ring and:performs some fast preliminary
> checks if necessary and collects error; in particular it may verify whether
> schema change is valid to allow for fail-fast client notification.;
> 5) If at least one error occurs during {{INIT}} stage, originator may throw
> exception to the user.
> 6) {{ACK}} message schedules asynchronous execution of the command.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
