Failed to start td-agent with aarch64 about fluent-package-builder HOT 8 CLOSED

Easy to add its option to only Arm64 CentOS package?

from fluent-package-builder.

@rluisr it may be that jemalloc is not compiled appropriately on aarch64 (not sure yet).

As a workaround,
just comment out #Environment=LD_PRELOAD=/opt/td-agent/lib/libjemalloc.so in td-agent.servce then execute systemctl daemon-reload && systemctl start td-agent works as expected. (even though jemalloc can't be enabled 😭 )

$ cat /lib/systemd/system/td-agent.service 
[Unit]
Description=td-agent: Fluentd based data collector for Treasure Data
Documentation=https://docs.treasuredata.com/articles/td-agent
After=network-online.target
Wants=network-online.target

[Service]
User=td-agent
Group=td-agent
LimitNOFILE=65536
#Environment=LD_PRELOAD=/opt/td-agent/lib/libjemalloc.so
Environment=GEM_HOME=/opt/td-agent/lib/ruby/gems/2.7.0/
Environment=GEM_PATH=/opt/td-agent/lib/ruby/gems/2.7.0/
Environment=FLUENT_CONF=/etc/td-agent/td-agent.conf
Environment=FLUENT_PLUGIN=/etc/td-agent/plugin
Environment=FLUENT_SOCKET=/var/run/td-agent/td-agent.sock
Environment=TD_AGENT_LOG_FILE=/var/log/td-agent/td-agent.log
Environment=TD_AGENT_OPTIONS=
EnvironmentFile=-/etc/sysconfig/td-agent
PIDFile=/var/run/td-agent/td-agent.pid
RuntimeDirectory=td-agent
Type=forking
# XXX: Fix fluentd executables path
ExecStart=/opt/td-agent/bin/fluentd --log $TD_AGENT_LOG_FILE --daemon /var/run/td-agent/td-agent.pid $TD_AGENT_OPTIONS
ExecStop=/bin/kill -TERM ${MAINPID}
ExecReload=/bin/kill -HUP ${MAINPID}
Restart=always
TimeoutStopSec=120

[Install]
WantedBy=multi-user.target

from fluent-package-builder.

I confirmed no problem on m6g with amazon linux 2 and ubuntu bionic .
I'm now trying to check centos 8 with non-AWS environment.

from fluent-package-builder.

@kenhys working perfectly! thanks!

@repeatedly perhaps, AL2 is important?

from fluent-package-builder.

AL2 is important?

Yes.

The problem is different PAGE_SIZE on Arm64 with CentOS.
Arm64 with Ubuntu/AL2 and x86_64 with all OSes use 4096 page size but Arm64 with CentOS uses 65536 page size.

from fluent-package-builder.

FYI: --with-lg-page configure option may be able to control page size when building jemalloc

from fluent-package-builder.

@repeatedly I'll check it.

from fluent-package-builder.

The cause is that the difference of the kernel between build time vs run time.
When the package is built using qemu:

 rake yum:build YUM_TARGETS=centos-8-aarch64

non-default kernel is used:

$ uname -a
Linux 5b7a1d11123f 5.4.0-42-generic #46-Ubuntu SMP Fri Jul 10 00:24:02 UTC 2020 aarch64 aarch64 aarch64 GNU/Linux
$ getconf PAGESIZE
4096

FYI: --with-lg-page configure option may be able to control page size when building jemalloc

It should fix this issue.

refs:

• jemalloc/jemalloc#467
• jemalloc/jemalloc#769

from fluent-package-builder.