23:43 - 24:29
87.9% of the time in this time range was spent in Garbage Collection pauses. This means that reducing garbage collections would have a high impact on performance. Usually, the most effective way to do is to reduce allocations.
Head over to the allocations view to learn what allocations were done in this period.
Use the tabs to select different profiling visualisations. You can Zoom in on a time range in the chart.
This demo runs on a pre-defined dataset. Blunders can show you live (within 30s) profiling data from your production servers. Interested? Get in touch!
This is a demo of Blunders at which we look at an Elasticsearch® instance under heavy, but synthetic load. There are different spikes in resource usage, each triggered by different kinds of load. The queries were ran as fast as the node could handle.
The version of ElasticSearch used was 7.4.2. The different spikes in the chart comes from different query patterns, selected to provoke certain behaviours. For instance, last large spike comes from particularly bad use of the Span multi-term query.
Other examples include leading and trailing wildcard searches, which have widely different performance characteristics. The experiment using, as well as not using, the "map" execution hint.
Play around. And let me know what you think!