rclone/docs/content/cache.md

title	description	date
Cache	Rclone docs for cache remote	2017-09-03

Cache (BETA)

The cache remote wraps another existing remote and stores file structure
and its data for long running tasks like rclone mount.

To get started you just need to have an existing remote which can be configured
with cache.

Here is an example of how to make a remote called test-cache. First run:

 rclone config

This will guide you through an interactive setup process:

No remotes found - make a new one
n) New remote
r) Rename remote
c) Copy remote
s) Set configuration password
q) Quit config
n/r/c/s/q> n
name> test-cache
Type of storage to configure.
Choose a number from below, or type in your own value
...
 5 / Cache a remote
   \ "cache"
...
Storage> 5
Remote to cache.
Normally should contain a ':' and a path, eg "myremote:path/to/dir",
"myremote:bucket" or maybe "myremote:" (not recommended).
remote> local:/test
Optional: The URL of the Plex server
plex_url> http://127.0.0.1:32400
Optional: The username of the Plex user
plex_username> dummyusername
Optional: The password of the Plex user
y) Yes type in my own password
g) Generate random password
n) No leave this optional password blank
y/g/n> y
Enter the password:
password:
Confirm the password:
password:
The size of a chunk. Lower value good for slow connections but can affect seamless reading.
Default: 5M
Choose a number from below, or type in your own value
 1 / 1MB
   \ "1m"
 2 / 5 MB
   \ "5M"
 3 / 10 MB
   \ "10M"
chunk_size> 2
How much time should object info (file size, file hashes etc) be stored in cache. Use a very high value if you don't plan on changing the source FS from outside the cache.
Accepted units are: "s", "m", "h".
Default: 5m
Choose a number from below, or type in your own value
 1 / 1 hour
   \ "1h"
 2 / 24 hours
   \ "24h"
 3 / 24 hours
   \ "48h"
info_age> 2
The maximum size of stored chunks. When the storage grows beyond this size, the oldest chunks will be deleted.
Default: 10G
Choose a number from below, or type in your own value
 1 / 500 MB
   \ "500M"
 2 / 1 GB
   \ "1G"
 3 / 10 GB
   \ "10G"
chunk_total_size> 3
Remote config
--------------------
[test-cache]
remote = local:/test
plex_url = http://127.0.0.1:32400
plex_username = dummyusername
plex_password = *** ENCRYPTED ***
chunk_size = 5M
info_age = 48h
chunk_total_size = 10G

You can then use it like this,

List directories in top level of your drive

rclone lsd test-cache:

List all the files in your drive

rclone ls test-cache:

To start a cached mount

rclone mount --allow-other test-cache: /var/tmp/test-cache

Write Features

Offline uploading

In an effort to make writing through cache more reliable, the backend
now supports this feature which can be activated by specifying a
cache-tmp-upload-path.

A files goes through these states when using this feature:

1. An upload is started (usually by copying a file on the cache remote)
2. When the copy to the temporary location is complete the file is part
   of the cached remote and looks and behaves like any other file (reading included)
3. After cache-tmp-wait-time passes and the file is next in line, rclone move
   is used to move the file to the cloud provider
4. Reading the file still works during the upload but most modifications on it will be prohibited
5. Once the move is complete the file is unlocked for modifications as it
   becomes as any other regular file
6. If the file is being read through cache when it's actually
   deleted from the temporary path then cache will simply swap the source
   to the cloud provider without interrupting the reading (small blip can happen though)

Files are uploaded in sequence and only one file is uploaded at a time.
Uploads will be stored in a queue and be processed based on the order they were added.
The queue and the temporary storage is persistent across restarts and even purges of the cache.

Write Support

Writes are supported through cache.
One caveat is that a mounted cache remote does not add any retry or fallback
mechanism to the upload operation. This will depend on the implementation
of the wrapped remote. Consider using Offline uploading for reliable writes.

One special case is covered with cache-writes which will cache the file
data at the same time as the upload when it is enabled making it available
from the cache store immediately once the upload is finished.

Read Features

Multiple connections

To counter the high latency between a local PC where rclone is running
and cloud providers, the cache remote can split multiple requests to the
cloud provider for smaller file chunks and combines them together locally
where they can be available almost immediately before the reader usually
needs them.

This is similar to buffering when media files are played online. Rclone
will stay around the current marker but always try its best to stay ahead
and prepare the data before.

Plex Integration

There is a direct integration with Plex which allows cache to detect during reading
if the file is in playback or not. This helps cache to adapt how it queries
the cloud provider depending on what is needed for.

Scans will have a minimum amount of workers (1) while in a confirmed playback cache
will deploy the configured number of workers.

This integration opens the doorway to additional performance improvements
which will be explored in the near future.

Note: If Plex options are not configured, cache will function with its
configured options without adapting any of its settings.

How to enable? Run rclone config and add all the Plex options (endpoint, username
and password) in your remote and it will be automatically enabled.

Affected settings:

• cache-workers: Configured value during confirmed playback or 1 all the other times

Known issues

Mount and --dir-cache-time

--dir-cache-time controls the first layer of directory caching which works at the mount layer.
Being an independent caching mechanism from the cache backend, it will manage its own entries
based on the configured time.

To avoid getting in a scenario where dir cache has obsolete data and cache would have the correct
one, try to set --dir-cache-time to a lower time than --cache-info-age. Default values are
already configured in this way.

Windows support - Experimental

There are a couple of issues with Windows mount functionality that still require some investigations.
It should be considered as experimental thus far as fixes come in for this OS.

Most of the issues seem to be related to the difference between filesystems
on Linux flavors and Windows as cache is heavily dependant on them.

Any reports or feedback on how cache behaves on this OS is greatly appreciated.

• https://github.com/ncw/rclone/issues/1935
• https://github.com/ncw/rclone/issues/1907
• https://github.com/ncw/rclone/issues/1834

Risk of throttling

Future iterations of the cache backend will make use of the pooling functionality
of the cloud provider to synchronize and at the same time make writing through it
more tolerant to failures.

There are a couple of enhancements in track to add these but in the meantime
there is a valid concern that the expiring cache listings can lead to cloud provider
throttles or bans due to repeated queries on it for very large mounts.

Some recommendations:

• don't use a very small interval for entry informations (--cache-info-age)
• while writes aren't yet optimised, you can still write through cache which gives you the advantage
  of adding the file in the cache at the same time if configured to do so.

Future enhancements:

• https://github.com/ncw/rclone/issues/1937
• https://github.com/ncw/rclone/issues/1936

cache and crypt

One common scenario is to keep your data encrypted in the cloud provider
using the crypt remote. crypt uses a similar technique to wrap around
an existing remote and handles this translation in a seamless way.

There is an issue with wrapping the remotes in this order:
cloud remote -> crypt -> cache

During testing, I experienced a lot of bans with the remotes in this order.
I suspect it might be related to how crypt opens files on the cloud provider
which makes it think we're downloading the full file instead of small chunks.
Organizing the remotes in this order yelds better results:
cloud remote -> cache -> crypt

Specific options

Here are the command line options specific to this cloud storage
system.

--cache-chunk-path=PATH

Path to where partial file data (chunks) is stored locally. The remote
name is appended to the final path.

This config follows the --cache-db-path. If you specify a custom
location for --cache-db-path and don't specify one for --cache-chunk-path
then --cache-chunk-path will use the same path as --cache-db-path.

Default: /cache-backend/
Example: /.cache/cache-backend/test-cache

--cache-db-path=PATH

Path to where the file structure metadata (DB) is stored locally. The remote
name is used as the DB file name.

Default: /cache-backend/
Example: /.cache/cache-backend/test-cache

--cache-db-purge

Flag to clear all the cached data for this remote before.

Default: not set

--cache-chunk-size=SIZE

The size of a chunk (partial file data). Use lower numbers for slower
connections.

Default: 5M

--cache-total-chunk-size=SIZE

The total size that the chunks can take up on the local disk. If cache
exceeds this value then it will start to the delete the oldest chunks until
it goes under this value.

Default: 10G

--cache-chunk-clean-interval=DURATION

How often should cache perform cleanups of the chunk storage. The default value
should be ok for most people. If you find that cache goes over cache-total-chunk-size
too often then try to lower this value to force it to perform cleanups more often.

Default: 1m

--cache-info-age=DURATION

How long to keep file structure information (directory listings, file size,
mod times etc) locally.

If all write operations are done through cache then you can safely make
this value very large as the cache store will also be updated in real time.

Default: 6h

--cache-read-retries=RETRIES

How many times to retry a read from a cache storage.

Since reading from a cache stream is independent from downloading file data,
readers can get to a point where there's no more data in the cache.
Most of the times this can indicate a connectivity issue if cache isn't
able to provide file data anymore.

For really slow connections, increase this to a point where the stream is
able to provide data but your experience will be very stuttering.

Default: 10

--cache-workers=WORKERS

How many workers should run in parallel to download chunks.

Higher values will mean more parallel processing (better CPU needed) and
more concurrent requests on the cloud provider.
This impacts several aspects like the cloud provider API limits, more stress
on the hardware that rclone runs on but it also means that streams will
be more fluid and data will be available much more faster to readers.

Note: If the optional Plex integration is enabled then this setting
will adapt to the type of reading performed and the value specified here will be used
as a maximum number of workers to use.
Default: 4

--cache-chunk-no-memory

By default, cache will keep file data during streaming in RAM as well
to provide it to readers as fast as possible.

This transient data is evicted as soon as it is read and the number of
chunks stored doesn't exceed the number of workers. However, depending
on other settings like cache-chunk-size and cache-workers this footprint
can increase if there are parallel streams too (multiple files being read
at the same time).

If the hardware permits it, use this feature to provide an overall better
performance during streaming but it can also be disabled if RAM is not
available on the local machine.

Default: not set

--cache-rps=NUMBER

This setting places a hard limit on the number of requests per second that cache
will be doing to the cloud provider remote and try to respect that value
by setting waits between reads.

If you find that you're getting banned or limited on the cloud provider
through cache and know that a smaller number of requests per second will
allow you to work with it then you can use this setting for that.

A good balance of all the other settings should make this
setting useless but it is available to set for more special cases.

NOTE: This will limit the number of requests during streams but other
API calls to the cloud provider like directory listings will still pass.

Default: disabled

--cache-writes

If you need to read files immediately after you upload them through cache
you can enable this flag to have their data stored in the cache store at the
same time during upload.

Default: not set

--cache-tmp-upload-path=PATH

This is the path where cache will use as a temporary storage for new files
that need to be uploaded to the cloud provider.

Specifying a value will enable this feature. Without it, it is completely disabled
and files will be uploaded directly to the cloud provider

Default: empty

--cache-tmp-wait-time=DURATION

This is the duration that a file must wait in the temporary location
cache-tmp-upload-path before it is selected for upload.

Note that only one file is uploaded at a time and it can take longer to
start the upload if a queue formed for this purpose.

Default: 15m

--cache-db-wait-time=DURATION

Only one process can have the DB open at any one time, so rclone waits
for this duration for the DB to become available before it gives an
error.

If you set it to 0 then it will wait forever.

Default: 1s