[Bug 1577844] Re: don't run ifup at .service before networking.service

[Martin Pitt]

Turns out this was all just an illusion. ifup at .service already has
"After=network-pre.target", i. e. is actual execution will already be
deferred until after network-pre.target. cloud-init-local.service is
Before=network-pre.target, so the effect of any hotplug event will sort
correctly.

I added a "ExecStart=/bin/sleep 5" to cloud-init-local.service, and this
shows that network-pre.target and ifup@ correctly blocks:

May 04 14:57:05 autopkgtest cloud-init[336]: [CLOUDINIT] stages.py[INFO]: network config is disabled by /var/lib/cloud/data/upgraded-network
May 04 14:57:05 autopkgtest cloud-init[336]: Cloud-init v. 0.7.7 running 'init-local' at Wed, 04 May 2016 12:57:04 +0000. Up 3.64 seconds.
May 04 14:57:10 autopkgtest systemd[1]: Started Initial cloud-init job (pre-networking).
May 04 14:57:10 autopkgtest systemd[1]: Reached target Network (Pre).
May 04 14:57:10 autopkgtest systemd[1]: Starting Raise network interfaces...
May 04 14:57:11 autopkgtest systemd[1]: Started ifup for ens3.

So there is nothing to fix in ifupdown, and it should be safe to drop
all the hackery in cloud-init: /run/cloud-init/network-config-ready from
cloud-init-local.service, the whole cloud-init-wait, and
/lib/udev/rules.d/79-cloud-init-net-wait.rules.

** Changed in: ifupdown (Ubuntu)
       Status: In Progress => Invalid

** Changed in: cloud-init (Ubuntu)
       Status: New => Triaged

** Summary changed:

- don't run ifup at .service before networking.service
+ Drop unnecessary blocking of all net udev rules

** Description changed:

+ cloud-inits networking setup currently jumps through a lot of bad hoops
+ to make sure that ifup@ does not run until after cloud-init-
+ local.service. This includes blocking udev rules for an indefinite time,
+ which is racy, a potential deadlock, and highly non-elegant.
+ 
+ This is also not necessary: while ifupdown's net udev rule certainly can
+ fire before cloud-init-local, it only asynchronously starts
+ ifup at .service which will be deferred until after network-pre.target and
+ thus after cloud-init-local.service.
+ 
+ -------
+ Original description, which turned out to be completely false and just us being misled:
+ 
  ifup at .service can (and often does) run for a particular interface before
  networking.service runs. This is brittle as during early boot ifup is
  prone to fail: / might still be read-only, /var might not yet exist or
  be writable, dhclient-enter-hooks.d/ or if-up.d/ hooks might silently
  fail, etc. It is also unnecessary as networking.service will bring up
  all "auto" and all present "allow-hotplug" interfaces anyway, and it
  runs at the right time.
  
  We should make either 80-ifupdown.rules or ifup at .service ignore events
  until networking.service is active, or wait until after it has run
  (slower, but avoids race conditions when hotplug events happen while
  networking.service is running). Thus we need to add
  After=networking.service to ifup at .service, so that this only does stuff
  after doing the "coldplug" configuration.
  
  This also affects cloud-init's setup of networking: this currently jumps
  through a lot of bad hoops to make sure that ifup@ does not run until
  after cloud-init-local.service. This includes blocking udev rules for an
  indefinite time, which is racy, a potential deadlock, and highly non-
  elegant.
  
  https://bugs.debian.org/752919 is related to this issue.

** Changed in: cloud-init (Ubuntu)
     Assignee: (unassigned) => Scott Moser (smoser)

** Changed in: ifupdown (Ubuntu)
     Assignee: Martin Pitt (pitti) => (unassigned)

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to ifupdown in Ubuntu.
https://bugs.launchpad.net/bugs/1577844

Title:
  Drop unnecessary blocking of all net udev rules

Status in cloud-init package in Ubuntu:
  Triaged
Status in ifupdown package in Ubuntu:
  Invalid

Bug description:
  cloud-inits networking setup currently jumps through a lot of bad
  hoops to make sure that ifup@ does not run until after cloud-init-
  local.service. This includes blocking udev rules for an indefinite
  time, which is racy, a potential deadlock, and highly non-elegant.

  This is also not necessary: while ifupdown's net udev rule certainly
  can fire before cloud-init-local, it only asynchronously starts
  ifup at .service which will be deferred until after network-pre.target
  and thus after cloud-init-local.service.

  -------
  Original description, which turned out to be completely false and just us being misled:

  ifup at .service can (and often does) run for a particular interface
  before networking.service runs. This is brittle as during early boot
  ifup is prone to fail: / might still be read-only, /var might not yet
  exist or be writable, dhclient-enter-hooks.d/ or if-up.d/ hooks might
  silently fail, etc. It is also unnecessary as networking.service will
  bring up all "auto" and all present "allow-hotplug" interfaces anyway,
  and it runs at the right time.

  We should make either 80-ifupdown.rules or ifup at .service ignore events
  until networking.service is active, or wait until after it has run
  (slower, but avoids race conditions when hotplug events happen while
  networking.service is running). Thus we need to add
  After=networking.service to ifup at .service, so that this only does
  stuff after doing the "coldplug" configuration.

  This also affects cloud-init's setup of networking: this currently
  jumps through a lot of bad hoops to make sure that ifup@ does not run
  until after cloud-init-local.service. This includes blocking udev
  rules for an indefinite time, which is racy, a potential deadlock, and
  highly non-elegant.

  https://bugs.debian.org/752919 is related to this issue.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/cloud-init/+bug/1577844/+subscriptions