A large MogileFS storage rebalance was moving data between two sites.

After 12.17 days, the job was 9.91 percent complete. The working estimate showed another 110.64 days remaining. The system was moving the equivalent of 1.81 storage devices, or about 3,517 gigabytes, per day.

Those numbers came from a small internal calculation shared with the thread so other people could check the math. The underlying script and internal locations are not public, but the progress figures are enough to show the length of the operation.

MogileFS is a distributed filesystem. Shutterstock used distributed storage because the image and video collection was too large and important to depend on one machine or one storage array. Files were spread across devices, and the system kept track of where copies lived.

A rebalance moved data so that capacity and replicas were distributed in the intended way. At this scale, the work could not be treated like copying a folder. It ran while customers searched and downloaded content and contributors continued uploading new files.

The long estimate did not automatically mean the job was failing. Moving several terabytes per day for months could be the safe rate if faster movement would compete with customer traffic or normal processing. The email does not state the rate limit or performance impact, so those details remain open.

The progress report also needed to be readable. The thread discussed how often the data should be produced and how to present it. A percentage alone was not enough. The team wanted elapsed time, estimated time remaining, device rate, and data rate so it could see whether the job was holding steady.

By the end of 2012, Shutterstock had more than 23 million images in its collection. In April 2013, the company announced that the collection had passed 25 million images and 300 million cumulative paid downloads. Storage work happened underneath that growth.

This skeleton does not claim the rebalance finished on the estimate or that every calculation proved accurate. It records the measured state on April 8: roughly ten percent complete, twelve days elapsed, and a large amount of data still to move.

Archive