TL;DR: update your bin/qgtunnel
.
I've recently noticed an increase in my web dyno's memory usage. After digging a bit, I could see that the LD_PRELOAD
variable that should be set with heroku-buildpack-jemalloc was not set correctly. I used a tiny script (bin/show_preload
) that helped me debug that and trace which program was overriding LD_PRELOAD
.
#!/usr/bin/env bash
echo "buildpack=foo preload='$LD_PRELOAD' at=start-app cmd='$@'"
$@
I introduced that in our Procfile
:
web: bin/show_preload bin/qgtunnel bin/show_preload bin/start-nginx bin/show_preload bin/start-pgbouncer bin/show_preload bundle exec puma -C config/puma.rb
And when lauching on heroku I can see that bin/qgtunnel
overrides our LD_PRELOAD
configuration.
I created a tiny helper for the time being which makes sure I keep original value as well as what is added by bin/qgtunnel
:
#!/usr/bin/env bash
after_qgtunnel_script=$(mktemp)
echo <<-BASH > $after_qgtunnel_script
# Retrieve previous LD_PRELOAD value
export LD_PRELOAD="\$LD_PRELOAD $LD_PRELOAD"
# Clean after usage
rm $after_qgtunnel_script
# Start following commands
$@
BASH
chmod +x $after_qgtunnel_script
bin/qgtunnel $after_qgtunnel_script $@
If you ever need this script use it in place of bin/qgtunnel
After reaching out to Quotaguard, they patched the qgtunnel
binary and there is no error anymore:
curl https://quotaguard.s3.amazonaws.com/qgtunnel-2.4.1.tar.gz | tar xz
git add bin/qgtunnel vendor/nss_wrapper/libnss_wrapper.so
git commit -m "Update qgtunnel to fix LD_PRELOAD"
NOTE: new versions may occur since that one, see the related documentation