Hi everyone! I'm trying to get fluent-bit's memory usage under control on a very constrained embedded Linux system. I'm using the node_exporter_metrics as well as a few tail and exec inputs. Outputs are

prometheus_remote_write

and

loki

. I have limited the memory usage of all inputs via

mem_buf_limit: 56KB

. When flushing of the output fails, I can see that inputs are paused:

[2025/11/06 06:32:19] [ warn] [input] tail.1 paused (mem buf overlimit)
[2025/11/06 06:32:19] [ info] [input] pausing tail.1
[2025/11/06 06:33:01] [ info] [input] resume tail.1
[2025/11/06 06:33:01] [ info] [input] tail.1 resume (mem buf overlimit)
[2025/11/06 06:33:01] [ warn] [input] tail.1 paused (mem buf overlimit)
[2025/11/06 06:33:01] [ info] [input] pausing tail.1
[2025/11/06 06:33:09] [ info] [input] resume tail.1

However, the total memory usage of the fluent-bit process continues to rise. Checking storage metrics I can see that

storage_layer.chunks.mem_chunks

is continuously increasing. Is there a way to limit that? In the Buffering and storage docs I find that

storage.total_limit_size

can be set for

storage.type == filesystem

but I can't find information on how to limit chunks in memory. Thanks!

maybe https://docs.fluentbit.io/manual/data-pipeline/inputs/tail#file-buffers-versus-fluent-bit-chunks

thanks @Celalettin. It seems both

buffer_chunk_size

and

buffer_max_size

default to

32k

which is pretty small. I see fluent-bit's RSS RAM usage growing from 500K to 15MB within 70min.