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
Using tail plugin, and polling (i.e. not inotify yet). There is no log file rotation. I realize that multiple workers can be used to break up this path further (e.g. even/odd, every third file, etc) to limit scope of the search for new files and updates.
What we're seeing though in 12-50% of cases (depending on load) that fluentD misses the beginning part of the file and only after seeing a log message like the below are the lines seen and sent by fluentD.
e.g. if we see this:
td-agent.log:2024-08-19 13:39:10 -0400 [info]: #0 following tail of /my/long/file/path/with/ahundred_thousand_subdirs/job-NNNNN-task-1.out
... then all lines from the start of the log (e.g. 13:34:00-0100 and onwards until the above time get missed.
I understand the latency can be large in polling such a large hierarchy, but would expect the logs to be complete once noticed by FluentD if read_from_head true is set.
I've looked through the Coolio and FileWatcher code in FluentD but do not see an obvious issue.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
What is a problem?
Using tail plugin, and polling (i.e. not inotify yet). There is no log file rotation. I realize that multiple workers can be used to break up this path further (e.g. even/odd, every third file, etc) to limit scope of the search for new files and updates.
What we're seeing though in 12-50% of cases (depending on load) that fluentD misses the beginning part of the file and only after seeing a log message like the below are the lines seen and sent by fluentD.
e.g. if we see this:
... then all lines from the start of the log (e.g. 13:34:00-0100 and onwards until the above time get missed.
I understand the latency can be large in polling such a large hierarchy, but would expect the logs to be complete once noticed by FluentD if read_from_head true is set.
I've looked through the Coolio and FileWatcher code in FluentD but do not see an obvious issue.
Describe the configuration of Fluentd
FLuentD Ruby - 1.16.1
Describe the logs of Fluentd
td-agent.log:2024-08-19 13:39:10 -0400 [info]: #0 following tail of /my/long/file/path/with/ahundred_thousand_subdirs/job-NNNNN-task-1.out
Environment
Beta Was this translation helpful? Give feedback.
All reactions