Created on 07-26-2016 08:03 AM - edited 09-16-2022 03:31 AM
Our reports manager is currently using 19G of 24G on /var, all in /var/lib.
More specifically:
/var/lib/cloudera-scm-headlamp/cloudera-scm-headlamp
We have 9 <cluster>-hdfs directories here which are taking up most of the space.
The contents are mostly fsimage.tmp.
Some of these clusters no longer exist, and we have other live clusters that aren’t here. The timestamps are from Mar 22.
Why is the reports manager keeping fsimage.tmp files, what are some of the configs we can use to manage these, and what’s the main goal (meaning if it’s meant as a backup strategy, why a one time copy instead of continual)?
Created 07-26-2016 09:07 AM
Hello,
In short, you can stop Reports Manager and then safely remove any fsimage.tmp file.
Reports Manager will download the fsimage to a temp file, index it, then rename the fsimage.tmp to fsimage.
If you have fsimage files lying around or directories named after HDFS that no longer exists, they can be removed. The fact that there are fsimage.tmp files left around indicates that indexing did not complete for those files. If an HDFS service is removed, Reports Manager does not clean up the previous files automatically.
By default, every hour the fsimage will be downloaded from your NameNode and indexed. If indexing is taking longer than an hour, then you can increase the interval in the Cloudera Management Service Reports Manager "Reports Manager Update Frequency" configuration.
If you have specific questions about what to delete, let us know.
In generaly, you should have one fsimage (or fsimage.tmp while indexing is occurring) per HDFS service that Cloudera Manager manages. If you have 2 clusters managed by Cloudera Manager, you will have two fsimage files.
Created 07-26-2016 09:07 AM
Hello,
In short, you can stop Reports Manager and then safely remove any fsimage.tmp file.
Reports Manager will download the fsimage to a temp file, index it, then rename the fsimage.tmp to fsimage.
If you have fsimage files lying around or directories named after HDFS that no longer exists, they can be removed. The fact that there are fsimage.tmp files left around indicates that indexing did not complete for those files. If an HDFS service is removed, Reports Manager does not clean up the previous files automatically.
By default, every hour the fsimage will be downloaded from your NameNode and indexed. If indexing is taking longer than an hour, then you can increase the interval in the Cloudera Management Service Reports Manager "Reports Manager Update Frequency" configuration.
If you have specific questions about what to delete, let us know.
In generaly, you should have one fsimage (or fsimage.tmp while indexing is occurring) per HDFS service that Cloudera Manager manages. If you have 2 clusters managed by Cloudera Manager, you will have two fsimage files.