docs: Cleaned up and corrected bugs docs verbiage

This commit is contained in:
Evan Harris 2020-08-03 15:51:52 -05:00 committed by Nick Craig-Wood
parent 841edc729c
commit 78abd21eec

View File

@ -12,25 +12,27 @@ description: "Rclone Bugs and Limitations"
Rclone doesn't currently preserve the timestamps of directories. This
is because rclone only really considers objects when syncing.
### Rclone struggles with millions of files in a directory
### Rclone struggles with millions of files in a directory/bucket
Currently rclone loads each directory entirely into memory before
using it. Since each Rclone object takes 0.5k-1k of memory this can
take a very long time and use an extremely large amount of memory.
Currently rclone loads each directory/bucket entirely into memory before
using it. Since each rclone object takes 0.5k-1k of memory this can take
a very long time and use a large amount of memory.
Millions of files in a directory tend caused by software writing cloud
storage (eg S3 buckets).
Millions of files in a directory tends to occur on bucket-based remotes
(e.g. S3 buckets) since those remotes do not segregate subdirectories within
the bucket.
### Bucket based remotes and folders
Bucket based remotes (eg S3/GCS/Swift/B2) do not have a concept of
Bucket based remotes (e.g. S3/GCS/Swift/B2) do not have a concept of
directories. Rclone therefore cannot create directories in them which
means that empty directories on a bucket based remote will tend to
disappear.
Some software creates empty keys ending in `/` as directory markers.
Rclone doesn't do this as it potentially creates more objects and
costs more. It may do in future (probably with a flag).
costs more. This ability may be added in the future (probably via a
flag/option).
## Bugs