Skip to main content

Troubleshooting & FAQ

Common Issues​

Issue: Unable to Mount PFS​

Symptoms: PFS fails to mount on compute instance

Possible Causes & Solutions:

  1. PFS not unmounted from previous resize: Complete any pending operations
  2. Incorrect mount path: Ensure mount path is valid and doesn't conflict with existing directories
  3. PFS in error state: Check status in dashboard and contact support

Issue: Resize Operation Failed​

Symptoms: Resize doesn't complete or PFS enters error state

Possible Causes & Solutions:

  1. PFS still mounted: Ensure PFS is completely unmounted from all Instances before resizing
  2. Active I/O operations: Wait for all operations to complete
  3. Backend issue: Contact support with error details

Critical Step

Always unmount PFS from all attached Instances before attempting to resize. Failure to do so will cause the resize operation to fail.

Issue: Poor Performance​

Symptoms: I/O operations slower than expected

Possible Causes & Solutions:

  1. Network bottleneck: Verify network bandwidth and latency
  2. Suboptimal access patterns: Review and optimize I/O patterns
  3. Resource contention: Check if compute nodes have sufficient resources
  4. Wrong performance tier: You may need a higher performance tier
  5. Too many concurrent operations: Optimize application parallelism

Issue: Cannot Delete PFS​

Symptoms: Deletion operation fails or is unavailable

Possible Causes & Solutions:

  1. PFS still mounted: Unmount from all attached Instances first
  2. Active services using PFS: Stop all services accessing the PFS
  3. Pending operations: Wait for any resize or maintenance operations to complete

Frequently Asked Questions​

General Questions​

Q: What makes PFS different from SFS?
A: PFS is optimized for high-performance parallel I/O workloads, offering superior throughput and IOPS compared to SFS. It's designed for HPC applications and intensive multi-node training scenarios.

Q: Can I convert an existing SFS to PFS?
A: No, you cannot directly convert SFS to PFS. You would need to create a new PFS and migrate your data.

Q: How many Instances can access a PFS simultaneously?
A: PFS is designed for high-concurrency parallel access. The exact number depends on your configuration and performance tier.

Mounting & Access​

Q: Why must I unmount PFS before resizing?
A: Unmounting ensures data consistency and prevents corruption during the resize operation. Active I/O during resize can lead to data loss.

Q: Can I specify custom mount paths?
A: Yes, you can specify your preferred mount path when attaching PFS to compute instances, allowing flexible integration with your applications.

Q: What happens to data when I unmount PFS?
A: Data remains intact on the PFS. Unmounting only disconnects the file system from that specific node. You can remount it later without data loss.

Q: Can I mount the same PFS to Instances in different regions?
A: No, PFS is region-specific. Instances must be in the same region as the PFS for optimal performance.

Performance​

Q: How do I know if I need PFS instead of SFS?
A: Consider PFS if you:

  • Require maximum I/O throughput
  • Run HPC or parallel computing workloads
  • Need consistent high performance under heavy concurrent access
  • Have budget for premium storage

Q: Why is my PFS not performing as expected?
A: Check:

  • Network bandwidth between nodes and storage
  • I/O access patterns (sequential vs random)
  • Number of concurrent operations
  • Node configuration and resources

Q: What throughput can I expect from PFS?
A: Performance varies based on your specific configuration and workload.

Data Management​

Q: How do I backup PFS data?
A: Implement regular backups by:

  • Copying data to Datasets (EOS) for long-term archival
  • Using rsync or parallel copy tools for large datasets

Q: What happens if I delete PFS by accident?
A: Data deletion is permanent and cannot be recovered. Always maintain backups and follow the deletion confirmation process carefully.

Data Safety Recommendations​

Backup Strategy​

Critical Reminder

PFS deletion is permanent and irreversible. Always maintain backups of mission-critical data.

Recommended Approach:

  1. Automated Backups: Schedule regular backups to external storage
  2. Datasets Integration: Archive important data to Datasets (EOS)
  3. Version Control: Use appropriate tools for code and configuration
  4. Disaster Recovery: Maintain copies in different regions for critical data

Before Resizing​

Pre-Resize Checklist
  1. Unmount from all Instances - Verify no Instances have PFS mounted
  2. Backup critical data - Create backups as additional safety measure
  3. Stop dependent services - Ensure no applications are accessing PFS
  4. Verify requirements - Confirm new size meets your needs

Before Deletion​

Pre-Deletion Checklist
  1. Verify backups - Confirm all important data is backed up elsewhere
  2. Check dependencies - Ensure no active workloads depend on this PFS
  3. Unmount completely - Remove PFS from all attached Instances
  4. Double-check PFS ID - Verify you're deleting the correct file system

Security Best Practices​

  • Access Control: Limit PFS access to only necessary compute instances
  • Data Sensitivity: Understand compliance requirements for stored data
  • Regular Audits: Review which Instances have access to each PFS
  • Unmount When Idle: Unmount PFS from Instances that aren't actively using it