Use this procedure to split a large dataset into two processing tasks when cloud processing does not finish successfully. Splitting the dataset reduces processing time and keeps both parts easier to align after processing.
When to Use This Procedure
Use two separate processing tasks when the original dataset is too large for reliable cloud processing. A common sign is that processing stops without a clear error in the main log output, but cloud monitoring shows a runtime timeout.
The log ends after point cloud filtering or similar steps without a clear failure message.
The cloud monitoring entry shows a timeout-related failure, such as Job attempt duration exceeded timeout.
How to Confirm That the Dataset Is Too Large
Open the cloud processing logs for the failed task.
Scroll to the section just before Verifying license and data integrity.
Find the last reindexed_bagfiles entry.
Check the number in the file name, for example bag_laser_vert_97.bag.
Use that number to estimate the recording length in minutes. Since indexing starts at zero, 97 means roughly 98 minutes of recorded data.
Datasets that are much longer than 60 minutes are more likely to reach cloud processing limits. They can also require more memory than cloud processing can provide.
Procedure
Start the first cloud processing task with the original dataset.
Set the range slider start value to 0%.
Set the range slider end value to about 55%.
Start the first processing job.
Create a second cloud processing task with the same dataset.
Set the second range slider start value to about 45%.
Set the second range slider end value to 100%.
Start the second processing job.
Use about 10% overlap between the two processing ranges.
Why the Overlap Matters
The overlap gives both processed outputs shared geometry and scan content. Use this shared area to align the two datasets after processing, especially if you did not capture control points during acquisition.
After Processing
If you used control points, verify that both processed parts align correctly in the shared reference system.
If you did not use control points, align the two processed datasets manually with the overlapping area.
Check the seam area for visible shift, rotation error, or height mismatch.
Best Practices
Keep each split below the practical processing limit instead of dividing the dataset exactly in half.
Choose an overlap with recognizable geometry, such as corridors, rooms, or structural features.
Split very large datasets into more than two parts if needed.
Document the percentage range for each task.
Expected Outcome
After splitting and processing the two parts separately, both jobs should complete more reliably. The final result can then be combined by aligning the outputs using the shared overlap section.