[Bug 1719287] [NEW] libnss-resolve causes gnome-shell to hang when there's a high rate of getaddrinfo() calls

[Chow Loong Jin]

Public bug reported:

I've been running a locust[1]-based web load test on my Ubuntu machine
running GNOME Shell (wayland). When starting a new test with a client
spawn rate of ~500/s, it issues a huge amount of getaddrinfo() calls.
When using libnss-resolve, this causes a huge number of NameOwnerChanged
events on the system bus:


signal time=1506335482.865950 sender=org.freedesktop.DBus -> destination=(null destination) serial=227290 path=/org/freedesktop/DBus; interface=org.freedesktop.DBus; member=NameOwnerChanged
   string ":1.190697"
   string ""
   string ":1.190697"
signal time=1506335482.868601 sender=org.freedesktop.DBus -> destination=(null destination) serial=227291 path=/org/freedesktop/DBus; interface=org.freedesktop.DBus; member=NameOwnerChanged
   string ":1.190697"
   string ":1.190697"
   string ""
signal time=1506335483.305614 sender=org.freedesktop.DBus -> destination=(null destination) serial=227292 path=/org/freedesktop/DBus; interface=org.freedesktop.DBus; member=NameOwnerChanged
   string ":1.190698"
   string ""
   string ":1.190698"
signal time=1506335483.310873 sender=org.freedesktop.DBus -> destination=(null destination) serial=227293 path=/org/freedesktop/DBus; interface=org.freedesktop.DBus; member=NameOwnerChanged
   string ":1.190698"
   string ":1.190698"
   string ""

This in turn causes gnome-shell to spend a huge amount of CPU churning
through these events, and if sufficiently starved of CPU (those load
testing scripts also consume quite a bit) when interacting with Google
Chome, it goes into key repeat runaway (similar to [2]).

I'm not completely certain, but there were quite a number of other
processes affected by this, and all of them start fighting over CPU time
to handle these NameOnwerChanged events.

As a workaround, disabling libnss-resolve in /etc/nsswitch.conf and just
using the stub resolver fixes things.


[1] https://locust.io
[2] https://bugzilla.redhat.com/show_bug.cgi?id=1459692

** Affects: systemd (Ubuntu)
     Importance: Undecided
         Status: New

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to systemd in Ubuntu.
https://bugs.launchpad.net/bugs/1719287

Title:
  libnss-resolve causes gnome-shell to hang when there's a high rate of
  getaddrinfo() calls

Status in systemd package in Ubuntu:
  New

Bug description:
  I've been running a locust[1]-based web load test on my Ubuntu machine
  running GNOME Shell (wayland). When starting a new test with a client
  spawn rate of ~500/s, it issues a huge amount of getaddrinfo() calls.
  When using libnss-resolve, this causes a huge number of
  NameOwnerChanged events on the system bus:

  
  signal time=1506335482.865950 sender=org.freedesktop.DBus -> destination=(null destination) serial=227290 path=/org/freedesktop/DBus; interface=org.freedesktop.DBus; member=NameOwnerChanged
     string ":1.190697"
     string ""
     string ":1.190697"
  signal time=1506335482.868601 sender=org.freedesktop.DBus -> destination=(null destination) serial=227291 path=/org/freedesktop/DBus; interface=org.freedesktop.DBus; member=NameOwnerChanged
     string ":1.190697"
     string ":1.190697"
     string ""
  signal time=1506335483.305614 sender=org.freedesktop.DBus -> destination=(null destination) serial=227292 path=/org/freedesktop/DBus; interface=org.freedesktop.DBus; member=NameOwnerChanged
     string ":1.190698"
     string ""
     string ":1.190698"
  signal time=1506335483.310873 sender=org.freedesktop.DBus -> destination=(null destination) serial=227293 path=/org/freedesktop/DBus; interface=org.freedesktop.DBus; member=NameOwnerChanged
     string ":1.190698"
     string ":1.190698"
     string ""

  This in turn causes gnome-shell to spend a huge amount of CPU churning
  through these events, and if sufficiently starved of CPU (those load
  testing scripts also consume quite a bit) when interacting with Google
  Chome, it goes into key repeat runaway (similar to [2]).

  I'm not completely certain, but there were quite a number of other
  processes affected by this, and all of them start fighting over CPU
  time to handle these NameOnwerChanged events.

  As a workaround, disabling libnss-resolve in /etc/nsswitch.conf and
  just using the stub resolver fixes things.

  
  [1] https://locust.io
  [2] https://bugzilla.redhat.com/show_bug.cgi?id=1459692

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/1719287/+subscriptions