Seeing various messages in the logs when under load about vernemq

[ioolkos]

@wernermorgenstern This is likely coming from the ports of your outgoing sessions (the 2 consumers).
You're probably not yet getting disconnects, though.
I cannot give you an elaborated recipe here, but a couple of keywords to look into:

• configuring adapted TCP buffers (buffer_sizes, tcp_listen_options, maybe increasing high_watermark setting using tcp_listen_options)
• increasing the non-acked inflight window (but only if your messages use QoS > 0)
• using shared subscriptions to have consumer side load balancing between multiple consumer instances

---

👉 Thank you for supporting VerneMQ: https://github.com/sponsors/vernemq
👉 Using the binary VerneMQ packages commercially (.deb/.rpm/Docker) requires a paid subscription.

[wernermorgenstern]

Thank you.
We do use shared subscriptions. So that is good.

Here is our config:
DOCKER_VERNEMQ_MAX_MESSAGE_RATE = 0
DOCKER_VERNEMQ_MAX_OFFLINE_MESSAGES = 0
DOCKER_VERNEMQ_MAX_ONLINE_MESSAGES = 1000
DOCKER_VERNEMQ_MAX_INFLIGHT_MESSAGES = 1000
DOCKER_VERNEMQ_SHARED_SUBSCRIPTION_POLICY = prefer_local
DOCKER_VERNEMQ_UPGRADE_OUTGOING_QOS = off
DOCKER_VERNEMQ_COORDINATE_REGISTRATIONS = on
DOCKER_VERNEMQ_QUEUE_TYPE = lifo
DOCKER_VERNEMQ_LISTENER__NR_OF_ACCEPTORS = 400
DOCKER_VERNEMQ_TCP_LISTEN_OPTIONS = [{sndbuf, 32768}, {recbuf, 32768}, {buffer, 65536}, {nodelay, true}, {linger, {true, 10}}, {send_timeout, 30000}, {send_timeout_close, true}]

systree_enabled = on
systree_interval = 20000
graphite_interval = 100000
plugins.vmq_bridge = off
plugins.vmq_webhooks = off
vmq_bcrypt.pool_size = 1
erlang.async_threads = 64
erlang.max_ports = 262144
leveldb.maximum_memory.percent = 20

Our QoS is 1.

How would I configure these two options:

- configuring adapted TCP buffers (buffer_sizes, tcp_listen_options, maybe increasing high_watermark setting using tcp_listen_options)
- increasing the non-acked inflight window (but only if your messages use QoS > 0)

[ioolkos]

How would I configure these two options:

• configuring adapted TCP buffers (buffer_sizes, tcp_listen_options, maybe increasing high_watermark setting using tcp_listen_options)
• increasing the non-acked inflight window (but only if your messages use QoS > 0)

For the first, by adding values (for high_watermark and low_watermark) to tcp_listen_options, the second is max_inflight_messages

---

👉 Thank you for supporting VerneMQ: https://github.com/sponsors/vernemq
👉 Using the binary VerneMQ packages commercially (.deb/.rpm/Docker) requires a paid subscription.