All our 3.2 kernels (and also sodium's 2.6.32) have the livelock fix (6b43ae8a619d17c4935c3320d2ef9e92bdeed05d). The system with the oldest kernel (es1006) was really close, though (it's uptime is 1039 days and Ubuntu pushed their fix two weeks before that).

The followup fix (6b1859dba01c7) only seems to apply to 3.4 and wasn't backported to older kernels.

The Java fix is present in all our openjdk-7 packages and the few systems with a openjdk-8 backport. This covers the complex services like Hadoop, Cassandra and Elastic.

However, Oracle didn't backport the fix to openjdk-6. The following systems use openjdk-6. In the worst case we need to restart some services, but possibly most of these should be moved to Java 7 anyway (e.g. lanthanum, since other part of Jenkins use Java 7 already).
labcontrol2001.wikimedia.org
ytterbium.wikimedia.org
labsdb1004.eqiad.wmnet
zirconium.wikimedia.org
labsdb1006.eqiad.wmnet
lanthanum.eqiad.wmnet

In addition three further systems have openjdk-6 installed, but the standard Java is based on 7 (based on java -version). I'll clean these up:
nembus.wikimedia.org
neptunium.wikimedia.org
gallium.wikimedia.org

@MoritzMuehlenhoff I copy pasted your last comment ( T103479#1391521 ) to the task detail and added some checkbox in front of each machine.

I guess you want to fill sub tasks for each of them.

ytterbium.wikimedia.org is Gerrit production host. Gerrit itself uses Java 7, so we can probably just purge Java 6: T103668: Remove Java 6 from ytterbium.wikimedia.org (Gerrit production host)

@MoritzMuehlenhoff: is the plan to smear the leap, or will we let it happen normally?

In principle it does make a difference to Cassandra as it's using timestamps for conflict resolution. However, the way we use it (data model and concurrency) makes it very unlikely that it would matter in practice.

The patches to smear the leap second have only merged into the Linux kernel last week.

We plan to disable NTP on the 29th, so the leap second won't be communicated to the systems. The individual hosts will run on hardware clock until the 1st, where we'll enable NTP again. A proposal is currently worked out and will be sent around.

@MoritzMuehlenhoff Danke Schoen! Thank you a ton to have taken extra care :-}