[
https://issues.apache.org/jira/browse/STORM-885?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15010807#comment-15010807
]
Longda Feng commented on STORM-885:
-----------------------------------
@d2r
Can we defer this patch . Why not do this like Heron. Every Topology has one
component "TopologyMaster", TopologyMaster will collect all heartbeat then send
this heartbeats to Nimbus.
1. Add Pacemaker will add difficulty of maintain one cluster, especially take
care of HA.
2. It is hard to do extension, the Pacemaker will become the cluster's
bottleneck, if it is become the cluster's bottleneck, it is very hard to do
extension. If something is wrong with Pacemaker, it will impact all topology,
not only one topology.
3. The logic is a little complicated, the JStorm's TopologyMaster logic is more
simple and clear, the code of handle heartbeat is no more than 300 lines.
worker --> topologymaster --> nimbus, due to topologymaster is one component of
Topology, it is easy to do HA. Even if the topologyMaster is down, it won't
impact any other topology.
> Heartbeat Server (Pacemaker)
> ----------------------------
>
> Key: STORM-885
> URL: https://issues.apache.org/jira/browse/STORM-885
> Project: Apache Storm
> Issue Type: Improvement
> Components: storm-core
> Reporter: Robert Joseph Evans
> Assignee: Kyle Nusbaum
>
> Large highly connected topologies and large clusters write a lot of data into
> ZooKeeper. The heartbeats, that make up the majority of this data, do not
> need to be persisted to disk. Pacemaker is intended to be a secure
> replacement for storing the heartbeats without changing anything within the
> heartbeats. In the future as more metrics are added in, we may want to look
> into switching it over to look more like Heron, where a metrics server is
> running for each node/topology. And can be used to aggregate/per-aggregate
> them in a more scalable manor.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
