Migration to read-only cache
Changes and implications of migrating the `/cache` filesystem to read-only access on November 19, 2024:
What is Changing?
- Access: The /cache filesystem will become read-only side-wide (farm, ifarm, Globus, XRootD), with the exception of the following transfer methods.
- File Transfer Methods:
- jcache: To read files from tape and write them to /cache.
- jput: To write files to tape. The `-cache` option will inform the system to put a copy in /cache as soon as it is written to tape.
- jmirror: Designed to support data ingestion from experimental halls but also available for general use. Also has a `-cache` option to place a copy of the file in /cache as soon as it is on tape.
Why is This Happening?
- Improved File Management: This change ensures timely movement of files to tape and maintains /cache as an accurate reflection of tape storage.
- Conflict Resolution: Reduces conflicts where files in /cache did not match those stored on tape, leading to slowdowns and data management confusion.
- Small File Handling: Addresses historical issues with small files that were not backed up or stored on tape.
Effects on Farm Job Workflows
- SWIF Workflows: Jobs using SWIF that specify output to `mss:/mss/...` and `/cache` will continue to function as before.
- Direct Writes: Jobs trying to write directly to /cache using standard commands (like `open()`, `cp`, `mv`) will fail.
- Slurm Jobs: Outputs from Slurm jobs not part of a SWIF workflow should go to `/volatile` and be manually moved to tape using `jput` on ifarm. It is recommended that Slurm workflows involving tape use SWIF.
- jput Availability: The `jput` command will not be available on non-interactive farm nodes to prevent job stalls.
Examples
What is Not Changing?
- Cache Deletion Policy: Remains unchanged.
- Cache File Pinning: Continues to work as before.
- jcache client: Continues to work as before.
- SWIF outputs to /cache: Continues to work as before.
Summary
This transition aims to enhance data management efficiency and reliability by establishing a clear process for how files are stored and accessed, particularly with regard to tape storage. Users need to adapt their workflows accordingly, especially those using Slurm jobs.