[
https://issues.apache.org/jira/browse/TAJO-1412?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14497418#comment-14497418
]
ASF GitHub Bot commented on TAJO-1412:
--------------------------------------
Github user jihoonson commented on a diff in the pull request:
https://github.com/apache/tajo/pull/432#discussion_r28477732
--- Diff: tajo-core/src/main/java/org/apache/tajo/master/QueryInfo.java ---
@@ -119,14 +119,37 @@ public int getQueryMasterClientPort() {
return queryMasterClientPort;
}
- public TajoProtos.QueryState getQueryState() {
+ public synchronized TajoProtos.QueryState getQueryState() {
return queryState;
}
- public void setQueryState(TajoProtos.QueryState queryState) {
+ public synchronized boolean isTerminalState() {
+ return isTerminalState(queryState);
+ }
+
+ public static boolean isTerminalState(TajoProtos.QueryState state) {
+ return state == TajoProtos.QueryState.QUERY_FAILED ||
+ state == TajoProtos.QueryState.QUERY_ERROR ||
+ state == TajoProtos.QueryState.QUERY_KILLED ||
+ state == TajoProtos.QueryState.QUERY_SUCCEEDED;
+ }
+
+ public synchronized void setQueryState(TajoProtos.QueryState queryState)
{
this.queryState = queryState;
+ notifyAll();
}
+ public synchronized boolean waitState(TajoProtos.QueryState expect, long
timeout)
+ throws InterruptedException {
+ long prev = System.currentTimeMillis();
+ while (timeout > 0 && expect != queryState && !isTerminalState()) {
+ wait(timeout);
--- End diff --
I think that this approach has a potential problem of process hang. Once
wait() is called, it can wake up only when setQueryState() is called. During
query processing, setQueryState() is called only when TajoMaster receives
heartbeats from workers. That means, if some heartbeats are lost, this process
cannot wake up even though waiting time exceeds the given timeout.
This will be not a big problem because we usually expect to receive at
least one heartbeat from a lot of workers. However, I think that it is still a
problem because the process's work can be different from our intention.
What do you think?
> Add polling API in TajoMasterClientProtocol to wait result effectively
> ----------------------------------------------------------------------
>
> Key: TAJO-1412
> URL: https://issues.apache.org/jira/browse/TAJO-1412
> Project: Tajo
> Issue Type: Improvement
> Components: client
> Reporter: Navis
> Assignee: Navis
> Priority: Trivial
>
> Currently, client polls master node to acquire result frequently (every 500
> msec), which seemed not small enough for light queries and not big enough for
> heavy queries.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
