Incorrect Labels in Hadoop Log Data

[mmantyla]

Recently, I began working on a demo for our log analysis tool, LogDelta, using your Hadoop. However, during the demo's creation, I grew increasingly suspicious of certain labels in the Hadoop data. As a result, what started as a simple demo evolved into a label investigation, ultimately requiring far more effort than initially anticipated.

I focused solely on the PageRank application, meaning that the WordCount application might still contain additional incorrect labels. Below are the identified incorrect labels along with their corresponding fixes:

ID	Orig Label	Fixed Label
1445144423722_0024	Normal	Disk Full
1445182159119_0017	Machine Down	Normal
1445062781478_0020	Machine Down	Normal
1445182151478_0015	Machine Down	Disk Full
1445182159119_0013	Disk Full	Machine Down
1445182159119_0011	Disk Full	Machine Down

If you're curious about how I reached these conclusions, the process is documented in a YouTube playlist.

• The key part of the label correction is covered in the final video.
• The earlier videos provide details on how the suspicions began to arise.
• I have also shared the text script of the video, which includes some visuals.