fluentbit vs vector configs

[ledbit]

It seems like you're configuring fluentbit to reassemble CRI logs in Lua, which is pretty expensive, but not in Vector or Fluentd. This would definitely impact performance, is there any reason why that's there or just left over configs?

[eranra]

@ledbit

This should be apples to apples -- all collectors are configured to merge partial messages::

--> in vector this is default: look here: https://vector.dev/docs/reference/configuration/sources/kubernetes_logs/#auto_partial_merge
--> in fluentd we use https://github.com/ViaQ/cluster-logging-collector-benchmarks/blob/main/conf/collector/fluentd/fluentd-clf.conf#L98
--> in fluentbit --> we use Lua filter

[agup006]

@eranra Fluent Bit partial message merging is done automatically with Docker mode (https://docs.fluentbit.io/manual/pipeline/inputs/tail), no lua required