[
https://issues.apache.org/jira/browse/PHOENIX-1457?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14380512#comment-14380512
]
ASF GitHub Bot commented on PHOENIX-1457:
-----------------------------------------
Github user jyates commented on a diff in the pull request:
https://github.com/apache/phoenix/pull/55#discussion_r27154098
--- Diff:
phoenix-core/src/main/java/org/apache/hadoop/hbase/ipc/PhoenixRpcScheduler.java
---
@@ -40,28 +40,34 @@
private static final int DEFAULT_MAX_CALLQUEUE_LENGTH_PER_HANDLER = 10;
private RpcScheduler delegate;
- private int minPriority;
- private int maxPriority;
- private RpcExecutor callExecutor;
+ private int indexPriority;
+ private int metadataPriority;
+ private RpcExecutor indexCallExecutor;
+ private RpcExecutor metadataCallExecutor;
private int port;
- public PhoenixIndexRpcScheduler(int indexHandlerCount, Configuration
conf,
- RpcScheduler delegate, int minPriority, int maxPriority) {
+ public PhoenixRpcScheduler(int indexHandlerCount, int
metadataHandlerCount, Configuration conf,
--- End diff --
Can we do layered handlers instead of clumping them all into one? HBase did
this with its meta, regular, and replication queues and it is terribly ugly.
Instead, maybe have this class instantiate the necessary handler and then
chain through them until one of the handlers range (min/max priority) applies
to the incoming message and it then passes it onto the queue. You could also do
some checking to ensure the ranges don't overlap, etc.
At the very least, this will get rid of the layers of if/else below.
> Use high priority queue for metadata endpoint calls
> ---------------------------------------------------
>
> Key: PHOENIX-1457
> URL: https://issues.apache.org/jira/browse/PHOENIX-1457
> Project: Phoenix
> Issue Type: Bug
> Reporter: James Taylor
> Assignee: Thomas D'Silva
> Labels: 4.3.1
>
> If the RS hosting the system table gets swamped, then we'd be bottlenecked
> waiting for the response back before running a query when we check if the
> metadata is in sync. We should run endpoint coprocessor calls for
> MetaDataService at a high priority to avoid that.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
