Hello, To stop the cluster gracefully use one of this methods: - call Ignite.active(false) from an application script. - use {ignite}/bin/control.sh script for the deactivation. - use Ignite Web Console “Active/Deactivate” switch that is on monitoring screen.
After that you can bring down your server nodes. The documentation is already in progress and will be ready the next week: https://issues.apache.org/jira/browse/IGNITE-6682 Anyway, if the cluster is halted abruptly then ongoing operations might be stopped in the middle of their execution causing exceptions on applications side. However, it will NOT cause data corruption or inconsistency because of the way the persistence operates (WAL, etc.). — Denis > On Oct 19, 2017, at 11:08 AM, blackfield <charles.kat...@maxpoint.com> wrote: > > Hello, > > I searched this topic, I can't seem to find it anywhere. Apology in advance > if this is covered somewhere.. > > In multi nodes cluster and persistence is enabled, what is the best practice > to bring down a node for maintenance (e.g applying OS patches, hardware > upgrade, etc)? > > If one just stops ignite node, what would happen if the node is in the > middle of processing a query/update etc? Will the client get exception? Is > there a graceful way to perform this to avoid no service interruption? > > > Somewhat related to the above, is there a best practice in bringing down a > cluster for maintenance (e.g. due to issue, ignite upgrade, etc)? > > For example, if we configure our cluster with TcpDiscoveryVmIpFinder and > provide lists of IP addresses of just few nodes, should these nodes be shut > down last? > > > Note: I am aware of GridGain's rolling upgrade capability, this question is > for Ignite specific. > > > > > -- > Sent from: http://apache-ignite-users.70518.x6.nabble.com/