lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andrzej Bialecki (JIRA)" <>
Subject [jira] [Commented] (SOLR-12652) SolrMetricManager.overridableRegistryName should be removed; it doesn't work
Date Mon, 10 Sep 2018 13:10:00 GMT


Andrzej Bialecki  commented on SOLR-12652:

I don't see any way how to fix it so that it works properly with Gauge metrics... As it is
now the feature is broken and leads to incorrect and misleading results, so I think it's ok
to remove it.

Although there's no 1:1 replacement for this feature, the aggregation of metrics across registries
can be now achieved in a more consistent way using {{SolrShardReporter}}, {{SolrClusterReporter}}
and {{/admin/metrics/collector}}.

> SolrMetricManager.overridableRegistryName should be removed; it doesn't work
> ----------------------------------------------------------------------------
>                 Key: SOLR-12652
>                 URL:
>             Project: Solr
>          Issue Type: Improvement
>      Security Level: Public(Default Security Level. Issues are Public) 
>          Components: metrics
>    Affects Versions: 7.1
>            Reporter: David Smiley
>            Priority: Minor
>          Time Spent: 10m
>  Remaining Estimate: 0h
> The {{SolrMetricManager.overridableRegistryName()}} method is a great idea but unfortunately
in practice I've found it doesn't really work; it seems fundamentally flawed.  +I wish it
could work+.  The main issue I think is that the callers of SMM.registerGauge/registerMetric
assumes it can place a gauge/metric and have it be the only once there (force==true).  But
it won't be if it's shared.  
> Another problem is in at least one of the reporters -- {{JmxMetricsReporter.JmxListener#registerMBean}}
will get in a race condition to remove an already-registered MBean but in the process of removing
it, it'll already get removed concurrently by some other core working on the same name.  This
results in {{}} logged as a warning; nothing serious.
 But I suspect conceptually there is a problem since which MBean should "win"?  Shrug.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message