Fix broken navoidverify on Linux

[lunkwill42]

While working on SNMPv3 implementation for the async/ipdevpoll parts of NAV, I incidentally found the navoidverify command to be broken before I started.

For the same reasons as documented in nav.ipdevpoll.epollreactor2, navoidverify seems to have stopped working (at least on Linux), due to incompatibilites between the default epollreactor implementation and pynetsnmp.

This change installs NAV's epollreactor2 in Twisted before any of the twisted imports run and install the default reactor. ipdevpoll does the same, but since this is an all-in-one script, this mucks about with code in between imports to ensure it runs before twisted, while a multi-module system like ipdevpoll can simply do it before calling the main() function.

Reference

nav/python/nav/ipdevpoll/epollreactor2.py

Lines 34 to 53 in ea4f3c5

	def _remove(self, xer, primary, other, selectables, event, antievent):
	"""
	Private method for removing a descriptor from the event loop.
	It does the inverse job of _add, and also add a check in case of the fd
	has gone away.
	It overrides the inherited epollreactor functionality to ensure that file
	descriptors closed behind our back are ignored and properly removed from the
	reactor's internal data structures. This is needed mostly because pynetsnmp
	adds reactor readers for file descriptors that are managed by the NET-SNMP C
	library. There is no way for Python code close these file descriptors in a
	controlled way, wherein they are removed from the reactor first - the
	NET-SNMP library will close them "behind our backs", so to speak.
	Attempting to unregister a closed file descriptor from the epoll object will
	cause an OSError that the original implementation left the client to handle -
	but this also caused the internal data structures of the reactor to become
	inconsistent.
	"""

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
0 Code Smells

No Coverage information
0.0% Duplication

[github-actions]

Test results

     12 files       12 suites   12m 27s ⏱️
3 228 tests 3 228 ✔️ 0 💤 0 ❌
9 159 runs  9 159 ✔️ 0 💤 0 ❌

Results for commit 5356a3e.

♻️ This comment has been updated with latest results.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (0388823) 55.32% compared to head (5356a3e) 55.32%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2737   +/-   ##
=======================================
  Coverage   55.32%   55.32%           
=======================================
  Files         567      567           
  Lines       41052    41052           
=======================================
  Hits        22713    22713           
  Misses      18339    18339           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[hmpf]

What happens on FreeBSD? Is this only a problem on Linux?

[lunkwill42]

What happens on FreeBSD? Is this only a problem on Linux?

I have not heard any complaints from UiT, which run NAV on FreeBSD (i.e. the only complaint was that ipdevpoll installed the epollreactor2 regardless of the OS, and it doesn't work on FreeBSD).

Apparently, this problem is exclusive to Twisted's built-in epollreactor, which it selects by default on Linux systems.

It might also be that a newer version of Twisted would work better (after all, we are locked to an older version atm).