I’m revisiting Appgate SDP after a hiatus of about eight months, prompted by a new zero-trust initiative at my organization. My previous deployment, version 5.2, was ultimately shelved due to persistent stability issues that made it untenable for production. The problems were multifaceted, but primarily manifested in the controller cluster and client connectivity.
The core instability I observed fell into a few specific categories:
* **Controller Cluster "Split-Brain" Scenarios:** Under moderate load, we would occasionally see the controller nodes lose consensus, leading to a state where policies would be inconsistently applied. This required manual intervention to restore quorum.
* **Ephemeral Port Exhaustion on Gateways:** The gateways, particularly when handling a large volume of concurrent, short-lived connections from IoT devices, would exhibit symptoms of port exhaustion. This was traced to socket `TIME_WAIT` accumulation, despite our tuning efforts within the OS parameters.
* **Unpredictable Client Disconnects:** The Windows client, while generally reliable for standard 8-hour days, would frequently and silently drop connections during long-running data transfers (e.g., database backups), without triggering a reconnection or adequate user notification. Our workaround was a cron-job-style ping, which defeated the purpose of a seamless SDP.
My configuration at the time was fairly standard, but the issues arose under real-world load. A snippet of the gateway sysctl tuning we attempted illustrates the depth we had to go to:
```bash
# Attempted mitigations for connection churn
net.ipv4.tcp_tw_reuse = 1
net.ipv4.tcp_fin_timeout = 30
net.ipv4.ip_local_port_range = 10000 65000
```
My question to the community is whether the stability curve has meaningfully improved in the last two major releases. I am specifically interested in:
* Experiences with controller clustering in versions 5.4 and beyond. Has the consensus mechanism been hardened?
* Gateway performance under sustained, high-connection-churn workloads.
* The reliability of the client, especially for persistent, long-lived tunnels required for legacy application access.
I am preparing a new proof-of-concept and need to determine if the platform's operational integrity now matches its robust security model, or if we should consider the architectural trade-offs of a more API-centric, event-driven access proxy built from composable services.
API whisperer