+++ This bug was initially created as a clone of Bug #1268382 +++

netcf calls aug_load() at the start of every public API that uses augeas, and libvirt's iface-list calls netcf-->augeas at least 3 times for each interface. In one test system with ~600 interfaces (300 bridges each attached to a vlan) this results in a runtime for "virsh iface-list --all" of ~22 seconds (it was 28 seconds prior to libvirt code being optimized to eliminate one of the three calls).

This is in spite of the fact that augeas itself isn't actually re-reading the data unless a file has changed - it's merely checking the mtime of every ifcfg-* file, but even that takes a very long time when there are a lot of interfaces.

There is an upstream patch that allows netcf to used cached config data from augeas during a public API call, refreshing the cache at most once per second (assuming a single persistent instance of netcf; if netcf is instead hutdown and re-inited, the entire config tree must be re-read). This lowers the time for virsh iface-list --all on the same test setup down to < 1 second.

Here is the upstream netcf patch that solves the problem:

commit 9b5f4eb57af28a604cd7ac8b2c1be9e49f0b517d
Author: Laine Stump <laine>
Date:   Mon Sep 28 17:11:11 2015 -0400

    call aug_load() at most once per second