As address the I/O delays in SQUID file system (Lustre) that occurred in September and October of 2023,
we have changed the mount option for SQUID file system (Lustre) from "flock" to "localflock".
 

The "flock" option enables file locking when mounting the file system, supporting cluster-wide control through flock. In contrast, the "localflock" option provides support for flock on a per-node basis.
 

Due to a software-related issue in the file system (Lustre), it was discovered that using the "flock" option can cause significant job delays on computing nodes and command operation delays on frontend nodes, especially when running high-load jobs that meet certain conditions. Therefore, we have switched to the "localflock" option.
 

Since a permanent solution has not yet been implemented for the file system (Lustre), we do not plan to revert back to the "flock" option at this time. The developers of file system are currently considering possible solutions.
 

We apologize for the delayed communication.
 

We believe that this change should not affect the execution of applications in general. However, there may be issues with applications that intentionally require consistency through flock across nodes. If you encounter any problems due to this change, please do not hesitate to contact us.