Skip to content

Loki Ingester: Working Set Memory vs Resident Set Memory #19916

@a7i

Description

@a7i

We are noticing a big different in Memory WSS vs RSS. This presents a challenge for HPA scaling on Memory, given that metrics-server only scrapes container_memory_working_set_bytes. Is there some configuration or flag that we need to set to reclaim memory more aggressively?

Ingester config:

    ingester:
      autoforget_unhealthy: true
      chunk_block_size: 10490000
      chunk_encoding: zstd
      chunk_target_size: 0
      max_chunk_age: 1h
      wal:
        enabled: true
        replay_memory_ceiling: 24GB
Image

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions