consoleproxy.session.timeout is honoured for noVNC sessions but not legacy VNC/RDP/AJAX viewers

[DaanHoogland]

problem

consoleproxy.session.timeout is now honoured for noVNC console sessions (#12810, PR #13058), but VNC/RDP/AJAX-based console viewers still rely on a separate, hardcoded idle threshold: ConsoleProxy.VIEWER_LINGER_SECONDS (180 seconds), used by isFrontEndAlive() in ConsoleProxyVncClient, ConsoleProxyRdpClient, and ConsoleProxyNoVncClient.

This means the effective idle timeout for a console session now depends on which viewer/hypervisor path served it:

• noVNC sessions: idle timeout follows the configured consoleproxy.session.timeout (both via ConsoleProxyGCThread and, since PR Backport: Honour consoleproxy.session.timeout for noVNC console sessions #13058, the WebSocket session's own idle timeout).
• Legacy VNC/RDP/AJAX sessions: idle timeout is still fixed at 180 seconds regardless of the consoleproxy.session.timeout setting.

An admin who sets consoleproxy.session.timeout to, say, 30 minutes to keep long-idle sessions open will still see legacy VNC/RDP/AJAX sessions get dropped after 3 minutes — an inconsistency that's confusing and undocumented.

versions

ACS 4.20+ (any branch carrying the PR #13058 / equivalent noVNC timeout fix)

The steps to reproduce the bug

1. Set the global setting consoleproxy.session.timeout to a value well above 180000 ms (e.g. 1800000 ms / 30 minutes).
2. Open a console session that uses the legacy VNC/RDP/AJAX path (e.g. a hypervisor or console mode that doesn't route through noVNC).
3. Leave the session idle.
4. Observe the session is torn down after ~180 seconds, not after the configured consoleproxy.session.timeout.
5. Compare against a noVNC console session under the same setting, which correctly honours the configured value.

What to do about it?

Derive VIEWER_LINGER_SECONDS from the same configured consoleproxy.session.timeout value (ConsoleProxy.sessionTimeoutMillis) instead of keeping it as an independent hardcoded constant, so all console viewer types (noVNC, VNC, RDP, AJAX) honour one single, consistently-configured idle timeout.

This was flagged during review of PR #13058 (#13058 (comment)) but deliberately left out of that PR's scope, since it's a noVNC-focused backport and touching isFrontEndAlive() for the legacy viewer types is a broader behavioural change that deserves its own review/testing pass.

[DaanHoogland]

@nvazquez can you have a look (AI aided research!), I am not sure if it is worth continuing as websockets seem to have become the standard, but want more opinions. cc @weizhouapache @winterhazel @wido (no more W’s to add now) @sureshanaparti

[wido]

@nvazquez can you have a look (AI aided research!), I am not sure if it is worth continuing as websockets seem to have become the standard, but want more opinions. cc @weizhouapache @winterhazel @wido (no more W’s to add now) @sureshanaparti

I thought WebSockets are the default now. No idea if we need to fix the legacy protocols.