Connection pool exhaustion during NetworkManager interface flapping on Fedora
0 reputation · 24 Feb 2021, 18:18 UTC
On Fedora systems using NetworkManager, rapid transitions between Wi-Fi and Ethernet profiles may coincide with application-level connection pool exhaustion. NetworkManager manages connection profiles and state transitions over D-Bus, but the interaction between interface flapping and kernel socket cleanup is not always straightforward to isolate.
The uncertainty centers on whether stale sockets in TIME_WAIT state are an expected consequence of profile switching or a race condition in the daemon's state machine. Application-layer pool settings and kernel-level file descriptor limits can produce similar symptoms, making it difficult to determine whether the operating system's network daemon is contributing to exhaustion or merely reflecting application demand.
Does NetworkManager's handling of interface flapping on current Fedora releases create a race condition that prolongs TIME_WAIT socket accumulation? When pool exhaustion events align with NetworkManager re-initializations, should diagnosis focus on the daemon's D-Bus state transitions or on kernel socket limits first? Is the interaction between systemd-resolved and NetworkManager during profile changes a known source of masked timeouts in connection pools?
fedora