When a container is restarted, first ~50 seconds of logs are lost

[mfornasa]

We noticed that first ~50 seconds of logs after a container startup are not captured.

For example:

• I started a container at 12:50:00
• After 50 seconds, Sumologic pods logs the following:

2017-02-15 12:50:42 +0000 [info]: following tail of /mnt/log/containers/my-pod-name-87d9f1b535e11c02690567e7b4bb65016346c263b683c96f720b90e52ca6982e.log

• Logs between 12:50:00 and 12:50:42 are lost.

This is a showstopper for us. Not only we are loosing logs, but we are loosing the most important log entries (application startup). Crashing application are not showing any log!

[stevezau]

Hi @mfornasa this repo used fluentd tail plugin. I suspect maybe the files are disappearing before the tail plugin can pick it up.

I wonder if there are some docker/kube settings we can tune to keep it around for longer or perhaps the better solution is to look at using the native fluentd driver as suggested in #9

[frankreno]

@mfornasa bc96bcc should fix this issue. I need to work on getting this pushed to Dockerhub though. Will update once available.

[frankreno]

@mfornasa: new image with the fix is now available in docker hub

[mfornasa]

@frankreno seems good. Thanks!