- Make long directory listings faster (remove O(N^2))
- Fix long directory listings where some files could be skipped due to inode expiration
- Remove extra empty COPY requests when metadata is not changed
- Add --fsync-on-close option for synchronous s3fs-like mode
- Fix --disable-xattr error code
- Do not try to switch between STANDARD and COLD based on the object size
- Update GeeseFS (now it also limits metadata memory usage)
- Return s3fs-fuse & rclone packages back because they are now fixed in Alpine
- Do not use hostNetwork: true for mounter... but it won't help much because GeeseFS is now started on host :)
- Publish helm chart to github pages
- Update GeeseFS
- Add CSIDriver and remove external-attacher
- Add missing mounts to Helm chart
- Allow to configure mounter in Helm chart
- Allow to configure kubelet path in Helm chart
- Add nodeSelector for provisioner in Helm chart
- Do not use os.Setenv to pass keys
- Fix minio client constructor argument
Previously, multiple containers with the same mounted volume resulted in multiple
FUSE processes. This behaviour was breaking parallel modifications from different
containers, consumed extra resources, and after mounting via systemd was introduced,
led to the total inability to mount the same volume into multiple containers on
the same host.
Now only one FUSE process is started per volume, per host.
* rclone use 1.54.1
* add EnsureFSMetaExist func
* fix removeObjects have err check
* add removeObjectsOneByOne, when removeObjects failed, will try to use it
* fix repeat in yaml
With this, each volume will get its own prefix within the bucket if it
is configured in the storage class. This also ensures backwards
compatibility with older volumes that have been created in earlier
versions of csi-s3.
