[
https://issues.apache.org/jira/browse/STORM-200?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13920964#comment-13920964
]
ASF GitHub Bot commented on STORM-200:
--------------------------------------
Github user revans2 commented on a diff in the pull request:
https://github.com/apache/incubator-storm/pull/38#discussion_r10303756
--- Diff: storm-core/src/jvm/backtype/storm/spout/ShellSpout.java ---
@@ -120,12 +170,23 @@ private void querySubprocess(Object query) {
} else {
_collector.emitDirect((int)task.longValue(),
stream, tuple, messageId);
}
+ } else if (command.equals("metrics")) {
+ handleMetrics(action);
}
}
} catch (IOException e) {
throw new RuntimeException(e);
}
}
+
+ public <T extends IShellMetric> T registerMetric(String name, T
metric) {
--- End diff --
This still requires two calls to register a metric. I really would like it
to be a single call, and preferably one that looks like all the other register
metric calls that will happen elsewhere.
Perhaps we can do something with TopologyContext. Perhaps we can add a way
to pull out the map of metrics that match a specific type so that the spout and
bolt can get cache this after all of the metrics have been added and use it for
the updates.
```java
public <T extends IMetric> Map<String,T> getRegisteredMetrics(Class<T>
type) {
Map<String, T> ret = new HashMap<String, T>();
for (Map<Integer, Map<String, IMetric>> taskIdToNameToMetric:
_registeredMetric.values()) {
Map<String, IMetric> nameToMetric =
taskIdToNameToMetric.get(_taskId);
if (nameToMetric != null) {
for (Map.Entry<String, IMetric> entry :
nameToMetric.entrySet()) {
if (type.isAssignableFrom(entry.getValue().getClass())) {
ret.put(entry.getKey(), (T)entry.getValue());
}
}
}
}
return ret;
}
```
> Proposal for Multilang's Metrics feature
> ----------------------------------------
>
> Key: STORM-200
> URL: https://issues.apache.org/jira/browse/STORM-200
> Project: Apache Storm (Incubating)
> Issue Type: New Feature
> Reporter: DashengJu
> Priority: Minor
>
> Storm 0.9.0.1 exposes a metrics interface to report summary statistics across
> the full topology. We can build our own metric, and build metrics consumer to
> use those statistics.
> But when we use Multilang(ie. Python), we can not use this feature. So we
> want to summit a proposal for multilang's metrics.
> The specifics of the proposal:
> 1. The main idea is: when user want to add a metric statistics in
> multilang(python) bolt，
> a) he need first create a metric object and register in ShellBolt's
> sub-class,
> b) then update the metric in Python bolt process through RPC call.
> 2. In Metrics API:
> a) extend IMetric interface add a method for RPC call：public void
> updateMetricFromRPC(List<Object> params);
> b) modify IMetric implements, to support updateMetricFromRPC;
> 3. In ShellBolt，
> a) we have a Map<String, IMetric> to hold user's registered metrics
> object；
> b) we have a method registerMetric(String name, T metric) for user to
> register their metris object;
> c) we have a method handleMetrics(Map action) to handle RPC call from
> Python process；
> 4) In Multilang protocol: add a command "metrics" for shell process to make
> RPC call. The protocol is: {"command":"metrics",
> "name":"metric-registerd-name", "params":["param-1", param-2]}
> 5) In storm.py：add rpcMetrics(name, params), user can update remote metric
> through this RPC call.
> any suggestions?
--
This message was sent by Atlassian JIRA
(v6.2#6252)
