logrotate and copytruncate

[dodysw2]

Question: is it expected for logrotate to always use copytruncate option for exporter to be able to read new log file properly? I have a quite busy nginx that needs hourly rotation, on which copytruncate would be too disk i/o intensive as one hour worth of log file size is already a bit large. However the exporter almost always fail to realize the file has been rotated -- it's still holding the old file descriptor -- some times to the point of the file already deleted (exporter still kept holding on the fd). Using copytruncate fix this problem, but why is the exporter behave this way?

[martin-helmich]

Sorry for the late response.

This package uses the github.com/hpcloud/tail package for following log files, which also has some reported issues regarding log rotation (e.g. hpcloud/tail#154, hpcloud/tail#98 or hpcloud/tail#94).

github.com/hpcloud/tail looks mostly abandoned, though. There seems to be a better-maintained fork (github.com/nxadm/tail), which apparently has addressed some of the open issues. I'll try bumping the package to its fork; possibly some of these issues will have already been fixed.