For HEC (httpInputQ) set maxSize in server.conf

[hrawat]

There is a practice of setting queueSize in inputs.conf [http://<token>] stanza. queueSize over writes server.conf stanza

[queue=httpInputQ]
maxSize

Now if you have multiple tokens with different queueSize.

inputs.conf
[http://1]
queueSize=1

[http://2]
queueSize=2

[http://3]
queueSize=3

[http://4]
queueSize=4

Globally only one inputs.conf stanza wins for final httpInputQ size.

This setting should only be set if setting 'persistentQueueSize' as well. If there are multiple http inputs configured and each input has set 'queueSize' but persistentQueueSize is not is set, splunkd will create one in-memory queue and pick the 'queueSize' value from first stanza after sorting http stanzas with matching token of first received http event in ascending order. With multiple pipelines configured, each pipeline will create one in-memory queue depending on the first http event received by the pipeline thus each pipeline might have different sized httpInputQ created. If there are multiple http stanzas configured and 'persistentQueueSize' is not set, prefer to set 'maxSize' under 'queue=httpInputQ' stanza in server.conf.

So best practice would be to never set per token queueSize in inputs.conf. Instead set one time in server.conf, if not setting persistentQueueSize.

[queue=httpInputQ]
maxSize

[hrawat]

We are updating docs to reflect layering of multiple http stanzas with different queueSize values.
Eventually all tokens share one input queue httpInputQ. Once all tokens are read in-memory the first token(shorted in ascending order) wins and creates final httpInputQ. Other queueSize values are no-op since the queue is already created.

Above is also applicable for multiple splunktcpin or tcpin ports having different queueSize but sharing splunktcp queue or tcpin queue.

[isoutamo]

Thanks @hrawat 

I just check this from docs and nether inputs.conf, server.conf or Set up and use HTTP Event Collector with configuration files says anything that there is only one value for queueSize. At least me, as non native English speaker, cannot get that conclusion based on those documents. It's much easier to understand this just opposite way. 

Have you already asked fixes/additional information for those documents?

r. Ismo

[PickleRick]

Yes. That is how I'd interpret the inputs.conf spec as well.

I can understand though why would just one value be effective (it's after all just one input bound to one port and the data is just internally split between various tokens) but the docs are ambiguous on this one to say the least.