[
https://issues.apache.org/jira/browse/IOTDB-373?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17080349#comment-17080349
]
WangChao commented on IOTDB-373:
--------------------------------
To check whether the node is busy, I think we have three methods:
# only check cpu, memory, network, io usage
# check cost of the working query plan and insert plan
## could be number of query plan, insert plan
## or the number of affected timeseries of the working query plan or insert
plan
# combine the above two
I think we could check node's total number of affected timeseries byt working
plan to check whether it is busy.
> [Distributed] Query coordinating
> --------------------------------
>
> Key: IOTDB-373
> URL: https://issues.apache.org/jira/browse/IOTDB-373
> Project: Apache IoTDB
> Issue Type: Sub-task
> Reporter: Tian Jiang
> Priority: Major
> Labels: distributed
>
> When using more than one replicas, query options are enriched and
> complicated. To ensure load balancing, we may have to choose the node with
> lowest load to perform the query, which requests knowing the load of each
> node and a formula to rank the nodes based on their status.
> We can also issue the same query to multiple replicas and pick up the fastest
> one to respond to the user as MapReduce has done. But this may result in
> resource wasting unless we feasibly support quick query cancellation.
> In a word, we should decide which replica(s) to serve a query and what
> information we need to collect to make the decision.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
