You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We've consistently experienced google-fluentd opening around 40,000 files under Debian Jessie when started with all the catch-all-config *.conf files installed and few of the associated services installed.
Removing all but the few *.conf we need (syslog and a tcp listener) reduces this number to around 3,000 on a quiet machine.
We haven't isolated whether particular *.conf's are the culprit, but suspect that attempting--and re-attempting--to tail non-existent log files might be part of the problem.
The best fix would be for fluentd to detect this situation and stop leaking file handles.
Until then, it seems a good idea to provide the catch-all configs as examples, not automatically installed. They could be put in an examples directory or named *.conf.example.
The text was updated successfully, but these errors were encountered:
We're noticing this issue too on Ubuntu 16.04 LTS. google-fluentd seems to be openingthread.rb up to 150,000 times with the catch-all config. By keeping only the following config files, we reduced that number to 172:
cassandra.conf
redis.conf
elasticsearch.conf
jenkins.conf
syslog.conf
tomcat.conf
This is all with latest google-fluentd at the time of writing (1.5.27-1)
@igorpeshansky can you take a look, as you had some thoughts about the catch-all config? I'm surprised it causes the agent to open thread.rb so many times, since it's clearly not a log file.
We've consistently experienced google-fluentd opening around 40,000 files under Debian Jessie when started with all the catch-all-config *.conf files installed and few of the associated services installed.
Removing all but the few *.conf we need (syslog and a tcp listener) reduces this number to around 3,000 on a quiet machine.
We haven't isolated whether particular *.conf's are the culprit, but suspect that attempting--and re-attempting--to tail non-existent log files might be part of the problem.
The best fix would be for fluentd to detect this situation and stop leaking file handles.
Until then, it seems a good idea to provide the catch-all configs as examples, not automatically installed. They could be put in an examples directory or named *.conf.example.
The text was updated successfully, but these errors were encountered: