You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@activemq.apache.org by "Larkin, Cian via users" <us...@activemq.apache.org> on 2026/09/29 10:57:18 UTC
Re: [External] Re: Consumer connection imbalance behind ingress-nginx SSL-passthrough: any native Artemis mechanism we're missing?
Hi Matt,
Thanks for the heads up, appreciate it! Will subscribe and repost there.
Cheers
Cian.
________________________________
From: Matt Pavlovich <ma...@apache.org>
Sent: Monday, September 28, 2026 4:50 PM
To: users@activemq.apache.org <us...@activemq.apache.org>
Cc: Larkin, Cian <Ci...@fmr.com>
Subject: [External] Re: Consumer connection imbalance behind ingress-nginx SSL-passthrough: any native Artemis mechanism we're missing?
⚠️ External Email
This email originated from outside the organization. Do not click links or open attachments unless you recognize the sender and trust the content.
Hi Cian-
Heads up— Artemis has been spun out into its own project (November 2025) and has its own mailing list, issue tracker, etc.
For Artemis questions, sending your question to those lists is your best bet.
Thanks,
Matt Pavlovich
Apache ActiveMQ PMC
> On Sep 28, 2026, at 10:39 AM, Larkin, Cian via users <us...@activemq.apache.org> wrote:
>
> We're running a 3-node Artemis cluster (replicated, Kubernetes/EKS) behind ingress-nginx, with a single shared hostname for client connections (SSL-passthrough on the AMQP/Core port). We're seeing significant consumer-connection imbalance across brokers — e.g. 300 consumer connections landing as ~150/75/75 instead of ~100/100/100 — which is causing consumption slowness on the overloaded broker.
> Root cause we've identified: SSL-passthrough bypasses nginx's own load-balancer entirely , so the actual "which broker" decision is made once, at connect time, by kube-proxy — and once a connection lands, it's sticky for its lifetime with no rebalancing. Any broker restart/rollout produces a fresh, uncorrected skew that compounds over time.
> We evaluated:
>
> * <connection-router> — redirects new connections round-robin, but our client library only follows a redirect on retry, not on the very first connection attempt (confirmed still true on current mainline ClientSessionFactoryImpl/ActiveMQClientProtocolManager).
> * useTopologyForLoadBalancing — redirects to internal pod hostnames unreachable from outside the cluster in our ingress setup.
> * Static multi-connector + RoundRobinConnectionLoadBalancingPolicy — works, but is client-side only and doesn't rebalance existing connections after a scale event.
>
> Question for the community: has anyone solved this at the broker/client level rather than the infra level? Specifically — is there any roadmap/plan for the client library to follow a connection-router redirect on the first connection attempt (not just retries)? That would make <connection-router> viable for our use case. Or is there a recommended pattern for exposing an Artemis cluster behind a single shared external hostname that we're missing?
>