banyan-julia
banyan-julia copied to clipboard
S3FS takes time to sync up
Sometimes S3FS takes time to sync up files or file metadata. On one occassion with the Black Scholes tests, we would have HDF5.ishdf5
cause an error about the file that we were trying to write to being invalid because the file still existed according to S3FS. The file had recently been deleted by some code that ran on the client side.
On another occasion, we got a ERROR: LoadError: ArgumentError: "/home/ec2-user/s3fs/banyan-cluster-data-pumpkincluster02-f47c1c35/iris_large.csv" is not a valid file
when trying to load a file that definitely got written (by some code on the client side).
This issue could affect usage of S3FS on the client side or on the cluster. There are several action items to take:
- [ ] Ensure that we are not configuring S3FS to allow caching.
- [ ] Use
fsync
to sync up files that are written to.