Hi, we have a cluster with hot and warm nodes. Our warm nodes have around 400 shards.
In 7.4.2, we see an increase of 10% in CPU usage in comparison to 6.3.2.
According to the hot threads call stacks, it is related to management tasks (see below).
99.6% (497.7ms out of 500ms) cpu usage by thread 'elasticsearch[NODE-NAME][management][T#5]' 2/10 snapshots sharing following 27 elements firstname.lastname@example.org/sun.nio.fs.UnixFileSystemProvider.readAttributes(UnixFileSystemProvider.java:149) email@example.com/sun.nio.fs.LinuxFileSystemProvider.readAttributes(LinuxFileSystemProvider.java:99) firstname.lastname@example.org/java.nio.file.Files.readAttributes(Files.java:1842) email@example.com/java.nio.file.Files.getLastModifiedTime(Files.java:2393)
We were able to trace the reason for the management tasks to the _node/stats query that our exporter is executing.
Is there any change that could affect the performance of _node/stats in nodes with many shards that you know of?