Heap usage in versions 6 and 7

After upgrading from 6.7.0 to 7.10.2, the amount of data in each data node has decreased, but the heap utilization rate has increased, leading to frequent circuit_breaking. The 7 version FST has moved to the off heap memory, so the memory utilization rate should be lower?I don't understand. Is there a bug?

3.3TB data per data node of the 6.7.0 version
only 1.5TB data per data node of the 7.10.2 version


image
The data format and cluster settings are the same, but the heap usage of cluster 7 is higher

You can dump the heap to check which class lead to high usage of heap.

What is the full output of the cluster stats API?

I would agree with DJ_Zhu and recommend having a look at a heap dump to see what is causing the increase.

Hi, we got the same problem. Upgrading from 6.8 to 7.17.8 I can see x3 increase in fixed_bit_set_memory_in_bytes usage. I would assume it's connected to heavy nested field usage in our mapping, but such an increase is a bit strange. We can see a decrease in terms memory usage but it's not enough to offset the 3x increase in fixed bitset.

How did you solve it?

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.