[datrixlab]

Description

The connections plugin stops using a probe that fails — psutil.net_connections() raising, or the nf_conntrack files not being present — and records that by writing net_connections_enabled / nf_conntrack_enabled into stats.

But update() starts every refresh with stats = self.get_init_value(), which is copy.copy(self.stats_init_value), and stats_init_value has both flags True (glances/plugins/connections/__init__.py:98). The flag is written into a dict that is thrown away at the end of the refresh, and the next refresh calls the failing probe again.

Measured by pointing conntrack at a file that does not exist and refreshing three times:

cycle 0: stats['nf_conntrack_enabled'] = False
cycle 1: stats['nf_conntrack_enabled'] = False
cycle 2: stats['nf_conntrack_enabled'] = False
conntrack read attempts: 3          (1 after this PR)
"Can not get network connections track" warnings: 3   (1 after this PR)

So on any host without /proc/sys/net/netfilter/nf_conntrack_* — macOS, Windows, a Linux kernel without the module — the warning plus its Disable connections track info line are logged on every single refresh.

The same applies to the psutil.net_connections() guard, and that one matters beyond the log: the plugin is disabled by default "because it consumes lots of CPU", so a user who turns it on (config or the K key) on a host where the call raises, such as macOS without root, pays for the failing call every refresh.

Every other plugin that gives up keeps the flag outside the per-refresh stats: mem with self.zfs_enabled (mem/__init__.py:134, read at :180), sensors with its grabber's init flag, wifi with a module-level check.

Resume

• Keep net_connections_enabled / nf_conntrack_enabled on the plugin instance, set to False the first time a probe fails.
• Copy them into stats at the start of each refresh, so update_views(), msg_curse() and the REST/WebUI payload still carry the same two fields as before.

Tests in tests/test_connections_states.py:

• a failed conntrack read happens once across three refreshes, and the flag stays False
• a raising psutil.net_connections is called once across three refreshes
• guard: when both probes work, they keep running every refresh and nf_conntrack_percent is still computed

The tests drive the real update() with the refresh timer pushed past its deadline between calls, the way the main loop reaches it; without that, the refresh timer alone would make any version of the code look correct.

With glances/plugins/connections/__init__.py reverted, the first two fail (3 reads / 3 calls instead of 1) and the guard passes. pytest tests/test_connections_states.py tests/test_core.py tests/test_plugin_ports.py → 75 passed, 6 skipped. ruff check and ruff format --check clean.

Fixed tickets

None open; I did not find an issue for this.

[nicolargo]

Thanks @datrixlab

Merged into develop.