How does ElasticSearch calculate memory pressure?

Question

We have a couple of Java services that we would like to add monitoring/alarms to.

We were collecting memory usage information from the JVM. Then, we noticed that ElasticSearch has a metric called memory pressure which sounds like a useful metric for our Java services as well.

The problem is that I can't figure out how it works. I've tried searching for "memory pressure" in the code on their GitHub , to no avail.

When I try to search for the formula or an explanation of how it calculates this, I somehow always end up finding the same two articles: Understanding the Memory Pressure Indicator and JVM memory pressure indicator .

Those nicely explain how to interpret the values ElasticSearch calculates, but not how it calculates it.

Does anyone know how the ElasticSearch memory pressure calculation works under the hood?

Answer 1

As mentioned on the third link you referenced:

the JVM memory pressure indicator is actually the fill rate of the old generation pool

So it's simply the percentage of the old pool usage

When retrieving the GET _nodes/stats you'll get the following information regarding the JVM metrics:

  ...
  "jvm" : {
    "timestamp" : 1614957874986,
    "uptime_in_millis" : 274397434,
    "mem" : {
      "heap_used_in_bytes" : 20094897664,
      "heap_used_percent" : 65,
      "heap_committed_in_bytes" : 30836523008,
      "heap_max_in_bytes" : 30836523008,
      "non_heap_used_in_bytes" : 310525488,
      "non_heap_committed_in_bytes" : 321077248,
      "pools" : {
        "young" : {
          "used_in_bytes" : 6710886400,
          "max_in_bytes" : 0,
          "peak_used_in_bytes" : 12264144896,
          "peak_max_in_bytes" : 0
        },
        "old" : {
          "used_in_bytes" : 13098798592,
          "max_in_bytes" : 30836523008,
          "peak_used_in_bytes" : 14881037312,
          "peak_max_in_bytes" : 30836523008
        },
        "survivor" : {
          "used_in_bytes" : 285212672,
          "max_in_bytes" : 0,
          "peak_used_in_bytes" : 1300234240,
          "peak_max_in_bytes" : 0
        }
      }
    },

So the memory pressure is simply computed with the following formula:

100 * jvm.mem.pools.old.used_in_bytes / jvm.mem.pools.old.max_in_bytes 

Which in my case yields 42% and can be seen on the following screenshot