Skip to content
Notifications
Clear all

Troubleshooting: Sensor showing 'disconnected' but it's running.

4 Posts
4 Users
0 Reactions
5 Views
(@latency_king)
Trusted Member
Joined: 4 months ago
Posts: 44
Topic starter   [#3700]

I have been conducting a detailed latency analysis of our endpoint telemetry pipeline and have encountered a persistent, performance-impacting anomaly that appears as a state discrepancy within the Trend Micro Vision One console. Specifically, multiple sensors across a heterogeneous fleet (Windows Server 2019, Windows 10 22H2) are reported as 'Disconnected' in the Vision One 'Endpoints' view, while the local `tmawf` and `tmas` services are unequivocally running and consuming expected CPU/memory resources. This state desynchronization introduces significant noise into our monitoring and incident response metrics.

The core issue is not a complete loss of data—some events still trickle into the Workbench—but the connection heartbeat appears to be failing or being discarded. This suggests a failure in the persistent, low-latency control channel, distinct from the bulk event upload channel. From a networking perspective, I have ruled out the obvious culprits:

* **Local Firewall/IDS:** We have verified the necessary egress rules for `tmawf.exe` to `agents.trendmicro.com` (TCP/443) and the regional collector FQDNs are in place. No asymmetric routing is present.
* **Proxy Configuration:** The sensors are configured for direct internet egress; no proxy is involved in these specific cases.
* **DNS Latency & Resolution:** Resolving the collector hostnames from the affected endpoints returns consistent, low-latency results (<30ms) with no packet loss. The IPs are correct and match the Vision One documentation for our region.
* **TCP Connection Health:** A continuous raw socket test to the collector IP:443 from affected hosts shows a stable connection with a consistent 0.1% packet loss rate, which is within acceptable thresholds and should not cause a 'Disconnected' state.

This leads me to hypothesize the fault lies in the application-layer handshake or credential renewal process. I have gathered the following diagnostic artifacts from a representative affected host:

**1. Service Status (Elevated PowerShell):**
```powershell
Get-Service -Name "tmawf", "tmas" | Select-Object Name, Status, StartType
```
Output confirms both services are `Running`.

**2. Recent Agent Logs (C:Program FilesTrend MicroVision One Agentlogs):**
Scouring `agent.log` and `trace.log` reveals a repeating pattern of seemingly successful collector communication interspersed with ominous but vague warnings. The key snippet:
```
[WARN] [2024-05-15 14:23:17 UTC] [HeartbeatThread] Failed to receive valid ACK from collector within timeout. Retrying...
[INFO] [2024-05-15 14:23:45 UTC] [UploadThread] Successfully uploaded 12 events to .
```
This indicates a bifurcated channel state: the event upload channel is functional (hence sporadic Workbench events), but the dedicated heartbeat/control channel is impaired.

**My primary questions for the community are:**

* Has anyone performed a packet-level analysis (e.g., Wireshark with `ssl.key_log_file`) to isolate the exact TLS handshake or heartbeat request/response failure between a 'running' but 'disconnected' sensor and the Trend Micro backend?
* Are there known, specific registry keys or configuration files (beyond the generic `agent.ini`) that govern the heartbeat interval, timeout thresholds, or collector failover logic that may be corrupted?
* Could this be a certificate (client-auth) renewal failure on the agent side? If so, what is the precise cleanup and forced renewal procedure that does not require a complete sensor reinstall?
* Is there a deterministic method to force a full state resynchronization from the agent to the Vision One console, beyond the standard service restarts (`net stop`/`net start`) which we have attempted multiple times?

The operational impact is a degradation in our real-time security posture visibility. Any detailed, technical insights into the agent-collector protocol, especially concerning state management and fault detection, would be invaluable.


Every microsecond counts.


   
Quote
(@brianw5)
Estimable Member
Joined: 1 week ago
Posts: 75
 

Yeah, I've seen this exact disconnect state before when the heartbeat channel gets blocked, even with services running. It can be maddening because the event channel uses a different, more tolerant retry logic.

Since you've ruled out the network basics, check the local agent logs. The key is usually in `tmawf.log` around the `SrvReportStatus` entries. If those heartbeat attempts are timing out or getting a specific HTTP error (like a 403 from a stale token), it'll show there. I've had it happen after a regional collector had a TLS certificate change that the older agent versions didn't handle gracefully. The event upload kept retrying on a different path, but the control channel just gave up.

You could also try a forced re-registration from the agent CLI to refresh its endpoint token, but that's a bit nuclear. The logs are your best bet first. Good luck.


Automate all the things.


   
ReplyQuote
(@ethanf)
Eminent Member
Joined: 1 week ago
Posts: 22
 

Interesting point about the different retry logic between channels. The forced re-registration you mentioned, is that done through the tmashell command? I'm hesitant to go that route on production systems without clearer triggers from the logs.



   
ReplyQuote
(@monitor_king)
Eminent Member
Joined: 3 months ago
Posts: 18
 

You're overcomplicating it. The key isn't network, it's the agent's internal state. If the heartbeat channel is dead but events flow, your agent holds a bad session token or can't parse the control response.

Check the `tmawf.log` for `SrvReportStatus` like user609 said, but look for the specific error code. I've seen this with a malformed API response after a backend update that older agent builds choke on.

Don't run a full re-registration yet. Force a heartbeat cycle from the CLI with the diagnostic command. If that fails, then you've isolated the control channel.



   
ReplyQuote