Subject: zookeeper: Hit by ZOOKEEPER-706
Package: zookeeper
Version: 3.4.5+dfsg-2
Severity: normal
Tag: upstream fixed-upstream
Dear Maintainer,
In situations where a large number of watches are in place (such as when
heavy YARN + heavy HBase traffic takes place on a moderately busy
cluster), zookeeper session re-establishment can fail.
As a consequence, YARN ResourceManager fails to start after a fail-over
situation, causing all sort of trouble down the line.
This has been identified and fixed upstream in
https://issues.apache.org/jira/browse/ZOOKEEPER-706, which made it into
release 3.4.7
As a workaround, I have been able to add this to /etc/zookeeper/zoo.cf:
#
# 2015-12-08 -- workaround for ZOOKEEPER-706
# default is 0xfffff
https://zookeeper.apache.org/doc/r3.3.2/zookeeperAdmin.html
# got issues with lengths as big as 1820946
jute.maxbuffer=4194303
# 0x3fffff
Although the messages from ZOOKEEPER-1162 suggested deploying the
jute.maxbuffer increase in both client and server side settings, adding
that just to zoo.cfg seemed enough to clear proceed on my workload.
-- Cyrille
-- System Information:
Debian Release: 8.1
APT prefers stable
APT policy: (950, 'stable'), (850, 'testing'), (800, 'unstable'),
(500, 'stable-updates')
Architecture: amd64 (x86_64)
Kernel: Linux 3.16.0-4-amd64 (SMP w/8 CPU cores)
Locale: LANG=fr_FR.UTF-8, LC_CTYPE=fr_FR.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash
Init: systemd (via /run/systemd/system)
Versions of packages zookeeper depends on:
ii adduser 3.113+nmu3
ii libzookeeper-java 3.4.6-8
ii openjdk-8-jre-headless [java6-runtime-headless] 8u45-b14-2
zookeeper recommends no packages.
zookeeper suggests no packages.
-- Configuration Files:
/etc/zookeeper/conf_example/myid changed [not included]
/etc/zookeeper/conf_example/zoo.cfg changed [not included]
-- no debconf information
