I reported here: https://jira.mesosphere.com/projects/DCOS_OSS/issues/DCOS_OSS-2112
Out of curiosity, does ignoring stdout.logrotate.state help with the lost documents?
Is this the 'state' file as used by logrotate? Depending on distribution you might also find this file at /var/lib/logrotate.status or /var/lib/logrotate/status.
Logrotates updates this file by first writing its state into a new temporary file and then using mv to replace the old state file with a new file. Filebeat detects this by inode change and would start reporting the status file from beginning. If the old status file has not been fully processed, contents is lost.
When DC/OS log rotation happens DC/OS touches this file and makes 2 line entry in it.
Are these lines appended, or is the file truncated before?
Hi @steffens, yes, as indicated using exclude_files ["stdout.logrotate.state"] helps.
According to the DC/OS guys, the standard Linux logrotate is used. Each "containerized" application gets its own state file.
As far as I can tell, the file is truncated. It only ever contains 2 lines. Exactly this...
logrotate state -- version 2
"/var/lib/mesos/slave/slaves/be77bad4-de55-4c2a-b832-babfb5e43646-S1/frameworks/be77bad4-de55-4c2a-b832-babfb5e43646-0001/executors/auto-scale-prod-raange.8e08cb64-01fb-11e8-a0a1-bed7ac162486/runs/a33d447f-5e32-4b09-8e3a-7440a0d2ddd1/stdout" 2018-1-25-21:54:29This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.