search

folbricht / desync

Alternative casync implementation
BSD 3-Clause "New" or "Revised" License
343 stars 44 forks source link
archive casync chunking golang synchronization

readme

desync

GoDoc

This project re-implements many features of upstream casync in Go. It seeks to maintain compatibility with casync's data structures, protocols and types, such as chunk stores (castr), index files (caibx/caidx) and archives (catar) in order to function as a drop-in replacement in many use cases. It also tries to maintain support for platforms other than Linux and simplify build/installation. It consists of a library that implements the features, available for integration into any 3rd-party product as well as a command-line tool.

For support and discussion, see Gitter chat. Feature requests should be discussed there before filing, unless you're interested in doing the work to implement them yourself.

Goals And Non-Goals

Among the distinguishing factors:

Terminology

The documentation below uses terms that may not be clear to readers not already familiar with casync.

Parallel chunking

One of the significant differences to casync is that desync attempts to make chunking faster by utilizing more CPU resources, chunking data in parallel. Depending on the chosen degree of concurrency, the file is split into N equal parts and each part is chunked independently. While the chunking of each part is ongoing, part1 is trying to align with part2, and part3 is trying to align with part4 and so on. Alignment is achieved once a common split point is found in the overlapping area. If a common split point is found, the process chunking the previous part stops, eg. part1 chunker stops, part2 chunker keeps going until it aligns with part3 and so on until all split points have been found. Once all split points have been determined, the file is opened again (N times) to read, compress and store the chunks. While in most cases this process achieves significantly reduced chunking times at the cost of CPU, there are edge cases where chunking is only about as fast as upstream casync (with more CPU usage). This is the case if no split points can be found in the data between min and max chunk size as is the case if most or all of the file consists of 0-bytes. In this situation, the concurrent chunking processes for each part will not align with each other and a lot of effort is wasted. The table below shows how the type of data that is being chunked can influence runtime of each operation. make refers to the process of chunking, while extract refers to re-assembly of blobs from chunks.

Command Mostly/All 0-bytes Typical data
make Slow (worst-case) - Likely comparable to casync Fast - Parallel chunking
extract Extremely fast - Effectively the speed of a truncate() syscall Fast - Done in parallel, usually limited by I/O

Seeds and reflinks

Copy-on-write filesystems such as Btrfs and XFS support cloning of blocks between files in order to save disk space as well as improve extraction performance. To utilize this feature, desync uses several seeds to clone sections of files rather than reading the data from chunk-stores and copying it in place:

chunks-from-seeds

Even if cloning is not available, seeds are still useful. desync automatically determines if reflinks are available (and the block size used in the filesystem). If cloning is not supported, sections are copied instead of cloned. Copying still improves performance and reduces the load created by retrieving chunks over the network and decompressing them.

Reading and writing tar streams

In addition to packing local filesystem trees into catar archives, it is possible to read a tar archive stream. Various tar formats such as GNU and BSD tar are supported. See https://golang.org/pkg/archive/tar/ for details on supported formats. When reading from tar archives, the content is no re-ordered and written to the catar in the same order. This may create output files that are different when comparing to using the local filesystem as input since the order depends entirely on how the tar file is created. Since the catar format does not support hardlinks, the input tar stream needs to follow hardlinks for desync to process them correctly. See the --hard-dereference option in the tar utility.

catar archives can also be extracted to GNU tar archive streams. All files in the output stream are ordered the same as in the catar.

Tool

The tool is provided for convenience. It uses the desync library and makes most features of it available in a consistent fashion. It does not match upsteam casync's syntax exactly, but tries to be similar at least.

Installation

The following builds the binary and installs it into $HOME/go/bin by default.

go install -v github.com/folbricht/desync/cmd/desync@latest

Alternative method using a clone, building from the tip of the master branch.

git clone https://github.com/folbricht/desync.git
cd desync/cmd/desync && go install

Subcommands

Options (not all apply to all commands)

Environment variables

Caching

The -c <store> option can be used to either specify an existing store to act as cache or to populate a new store. Whenever a chunk is requested, it is first looked up in the cache before routing the request to the next (possibly remote) store. Any chunks downloaded from the main stores are added to the cache. In addition, when a chunk is read from the cache and it is a local store, mtime of the chunk is updated to allow for basic garbage collection based on file age. The cache store is expected to be writable. If the cache contains an invalid chunk (checksum does not match the chunk ID), the operation will fail. Invalid chunks are not skipped or removed from the cache automatically. verfiy -r can be used to evict bad chunks from a local store or cache.

Multiple chunk stores

One of the main features of desync is the ability to combine/chain multiple chunk stores of different types and also combine it with a cache store. For example, for a command that reads chunks when assembling a blob, stores can be chained in the command line like so: -s <store1> -s <store2> -s <store3>. A chunk will first be requested from store1, and if not found there, the request will be routed to <store2> and so on. Typically, the fastest chunk store should be listed first to improve performance. It is also possible to combine multiple chunk stores with a cache. In most cases the cache would be a local store, but that is not a requirement. When combining stores and a cache like so: -s <store1> -s <store2> -c <cache>, a chunk request will first be routed to the cache store, then to store1 followed by store2. Any chunks that is not yet in the cache will be stored there upon first request.

Not all types of stores support all operations. The table below lists the supported operations on all store types.

Operation Local store S3 store HTTP store SFTP SSH (casync protocol)
Read chunks yes yes yes yes yes
Write chunks yes yes yes yes no
Use as cache yes yes yes yes no
Prune yes yes no yes no
Verify yes yes no no no

Store failover

Given stores with identical content (same chunks in each), it is possible to group them in a way that provides resilience to failures. Store groups are specified in the command line using | as separator in the same -s option. For example using -s "http://server1/|http://server2/", requests will normally be sent to server1, but if a failure is encountered, all subsequent requests will be routed to server2. There is no automatic fail-back. A failure in server2 will cause it to switch back to server1. Any number of stores can be grouped this way. Note that a missing chunk is treated as a failure immediately, no other servers will be tried, hence the need for all grouped stores to hold the same content.

Dynamic store configuration

Some long-running processes, namely chunk-server and mount-index may require a reconfiguration without having to restart them. This can be achieved by starting them with the --store-file options which provides the arguments that are normally passed via command line flags --store and --cache from a JSON file instead. Once the server is running, a SIGHUP to the process will trigger a reload of the configuration and replace the stores internally without restart. This can be done under load. If the configuration in the file is found to be invalid, and error is printed to STDERR and the reload ignored. The structure of the store-file is as follows:

{
  "stores": [
    "/path/to/store1",
    "/path/to/store2"
  ],
  "cache": "/path/to/cache"
}

This can be combined with store failover by providing the same syntax as is used in the command-line, for example {"stores":["/path/to/main|/path/to/backup"]}, See Examples for details on how to use the --store-file option.

Remote indexes

Indexes can be stored and retrieved from remote locations via SFTP, S3, and HTTP. Storing indexes remotely is optional and deliberately separate from chunk storage. While it's possible to store indexes in the same location as chunks in the case of SFTP and S3, this should only be done in secured environments. The built-in HTTP chunk store (chunk-server command) can not be used as index server. Use the index-server command instead to start an index server that serves indexes and can optionally store them as well (with -w).

Using remote indexes, it is possible to use desync completely file-less. For example when wanting to share a large file with mount-index, one could read the index from an index store like this:

desync mount-index -s http://chunk.store/store http://index.store/myindex.caibx /mnt/image

No file would need to be stored on disk in this case.

S3 chunk stores

desync supports reading from and writing to chunk stores that offer an S3 API, for example hosted in AWS or running on a local server. When using such a store, credentials are passed into the tool either via environment variables S3_ACCESS_KEY, S3_SECRET_KEY and S3_SESSION_TOKEN (if needed) or, if multiples are required, in the config file. Care is required when building those URLs. Below a few examples:

AWS

This store is hosted in eu-west-3 in AWS. s3 signals that the S3 protocol is to be used, https should be specified for SSL connections. The first path element of the URL contains the bucket, desync.bucket in this example. Note, when using AWS, no port should be given in the URL!

s3+https://s3-eu-west-3.amazonaws.com/desync.bucket

It's possible to use prefixes (or "directories") to object names like so:

s3+https://s3-eu-west-3.amazonaws.com/desync.bucket/prefix

Other service with S3 API

This is a store running on the local machine on port 9000 without SSL.

s3+http://127.0.0.1:9000/store

Setting S3 bucket addressing style for other services

desync uses minio as an S3 client library. It has an auto-detection mechanism for determining the addressing style of the buckets which should work for Amazon and Google S3 services but could potentially fail for your custom implementation. You can manually specify the addressing style by appending the "lookup" query parameter to the URL.

By default, the value of "?lookup=auto" is implied.

s3+http://127.0.0.1:9000/bucket/prefix?lookup=path
s3+https://s3.internal.company/bucket/prefix?lookup=dns
s3+https://example.com/bucket/prefix?lookup=auto

Compressed vs Uncompressed chunk stores

By default, desync reads and writes chunks in compressed form to all supported stores. This is in line with upstream casync's goal of storing in the most efficient way. It is however possible to change this behavior by providing desync with a config file (see Configuration section below). Disabling compression and store chunks uncompressed may reduce latency in some use-cases and improve performance. desync supports reading and writing uncompressed chunks to SFTP, S3, HTTP and local stores and caches. If more than one store is used, each of those can be configured independently, for example it's possible to read compressed chunks from S3 while using a local uncompressed cache for best performance. However, care needs to be taken when using the chunk-server command and building chains of chunk store proxies to avoid shifting the decompression load onto the server (it's possible this is actually desirable).

In the setup below, a client reads chunks from an HTTP chunk server which itself gets chunks from S3.

<Client> ---> <HTTP chunk server> ---> <S3 store>

If the client configures the HTTP chunk server to be uncompressed (chunk-server needs to be started with the -u option), and the chunk server reads compressed chunks from S3, then the chunk server will have to decompress every chunk that's requested before responding to the client. If the chunk server was reading uncompressed chunks from S3, there would be no overhead.

Compressed and uncompressed chunks can live in the same store and don't interfere with each other. A store that's configured for compressed chunks by configuring it client-side will not see the uncompressed chunks that may be present. prune and verify too will ignore any chunks written in the other format. Both kinds of chunks can be accessed by multiple clients concurrently and independently.

Configuration

For most use cases, it is sufficient to use the tool's default configuration not requiring a config file. Having a config file $HOME/.config/desync/config.json allows for further customization of timeouts, error retry behaviour or credentials that can't be set via command-line options or environment variables. All values have sensible defaults if unconfigured. Only add configuration for values that differ from the defaults. To view the current configuration, use desync config. If no config file is present, this will show the defaults. To create a config file allowing custom values, use desync config -w which will write the current configuration to the file, then edit the file.

Available configuration values:

Example config

{
  "s3-credentials": {
       "http://localhost": {
           "access-key": "MYACCESSKEY",
           "secret-key": "MYSECRETKEY"
       },
       "https://127.0.0.1:9000": {
           "aws-credentials-file": "/Users/user/.aws/credentials",
       },
       "https://127.0.0.1:8000": {
           "aws-credentials-file": "/Users/user/.aws/credentials",
           "aws-profile": "profile_static"
       },
       "https://s3.us-west-2.amazonaws.com": {
           "aws-credentials-file": "/Users/user/.aws/credentials",
           "aws-region": "us-west-2",
           "aws-profile": "profile_refreshable"
       }
  },
  "store-options": {
    "https://192.168.1.1/store": {
      "client-cert": "/path/to/crt",
      "client-key": "/path/to/key",
      "error-retry": 1
    },
    "https://10.0.0.1/": {
      "http-auth": "Bearer abcabcabc"
    },
    "https://example.com/*/*/": {
      "http-auth": "Bearer dXNlcjpwYXNzd29yZA=="
    },
    "https://cdn.example.com/": {
      "http-cookie": "PHPSESSID=298zf09hf012fh2; csrftoken=u32t4o3tb3gg43"
    },
    "/path/to/local/cache": {
      "uncompressed": true
    }
  }
}

Example aws credentials

[default]
aws_access_key_id = DEFAULT_PROFILE_KEY
aws_secret_access_key = DEFAULT_PROFILE_SECRET

[profile_static]
aws_access_key_id = OTHERACCESSKEY
aws_secret_access_key = OTHERSECRETKEY

[profile_refreshable]
aws_access_key_id = PROFILE_REFRESHABLE_KEY
aws_secret_access_key = PROFILE_REFRESHABLE_SECRET
aws_session_token = PROFILE_REFRESHABLE_TOKEN

Examples

Re-assemble somefile.tar using a remote chunk store and a blob index file.

desync extract -s ssh://192.168.1.1/path/to/casync.store/ -c /tmp/store somefile.tar.caibx somefile.tar

Use multiple stores, specify the local one first to improve performance.

desync extract -s /some/local/store -s ssh://192.168.1.1/path/to/casync.store/ somefile.tar.caibx somefile.tar

Extract version 3 of a disk image using the previous 2 versions as seed for cloning (if supported), or copying. Note, when providing a seed like --seed <file>.ext.caibx, it is assumed that <file>.ext is available next to the index file, and matches the index.

desync extract -s /local/store \
  --seed image-v1.qcow2.caibx \
  --seed image-v2.qcow2.caibx \
  image-v3.qcow2.caibx image-v3.qcow2

Extract an image using several seeds present in a directory. Each of the .caibx files in the directory needs to have a matching blob of the same name. It is possible for the source index file to be in the same directory also (it'll be skipped automatically).

desync extract -s /local/store --seed-dir /path/to/images image-v3.qcow2.caibx image-v3.qcow2

Mix and match remote stores and use a local cache store to improve performance. Also group two identical HTTP stores with | to provide failover in case of errors on one.

desync extract \
       -s "http://192.168.1.101/casync.store/|http://192.168.1.102/casync.store/" \
       -s ssh://192.168.1.1/path/to/casync.store/ \
       -s https://192.168.1.3/ssl.store/ \
       -c /path/to/cache \
       somefile.tar.caibx somefile.tar

Extract a file in-place (-k option). If this operation fails, the file will remain partially complete and can be restarted without the need to re-download chunks from the remote SFTP store. Use -k when a local cache is not available and the extract may be interrupted.

desync extract -k -s sftp://192.168.1.1/path/to/store file.caibx file.tar

Extract an image directly onto a block device. The -k or --in-place option is needed.

desync extract -k -s /mnt/store image.caibx /dev/sdc

Extract a file using a remote index stored in an HTTP index store

desync extract -k -s sftp://192.168.1.1/path/to/store http://192.168.1.2/file.caibx file.tar

Verify a local cache. Errors will be reported to STDOUT, since -r is not given, nothing invalid will be removed.

desync verify -s /some/local/store

Cache the chunks used in a couple of index files in a local store without actually writing the blob.

desync cache -s ssh://192.168.1.1/path/to/casync.store/ -c /local/cache somefile.tar.caibx other.file.caibx

Copy all chunks referenced in an index file from a remote HTTP store to a remote SFTP store.

desync cache -s ssh://192.168.1.2/store -c sftp://192.168.1.3/path/to/store /path/to/index.caibx

Cache chunks from remote locally with non-standard port. Ignore existing files that are available locally from seed(s). This will only download chunks from the remote if they do not exist in the seed. Works with multiple seeds.

desync cache -s http://cdn:9876 -c /tmp/chunkstore --ignore /tmp/indices/existing-image.raw.caibx /tmp/images/existing-image.raw

List the chunks referenced in a caibx.

desync list-chunks somefile.tar.caibx

Chop an existing file according to an existing caibx and store the chunks in a local store. This can be used to populate a local cache from a possibly large blob that already exists on the target system.

desync chop -s /some/local/store somefile.tar.caibx somefile.tar

Chop a blob according to an existing index, while ignoring any chunks that are referenced in another index. This can be used to improve performance when it is known that all chunks referenced in image-v1.caibx are already present in the target store and can be ignored when chopping image-v2.iso.

desync chop -s /some/local/store --ignore image-v1.iso.caibx image-v2.iso.caibx image-v2.iso

Pack a directory tree into a catar file.

desync tar archive.catar /some/dir

Pack a directory tree into an archive and chunk the archive, producing an index file.

desync tar -i -s /some/local/store archive.caidx /some/dir

Unpack a catar file.

desync untar archive.catar /some/dir

Unpack a directory tree using an index file referencing a chunked archive.

desync untar -i -s /some/local/store archive.caidx /some/dir

Pack a directory tree currently available as tar archive into a catar. The tar input stream can also be read from STDIN by providing '-' instead of the file name.

desync tar --input-format=tar archive.catar /path/to/archive.tar

Process a tar stream into a catar. Since catar don't support hardlinks, we need to make sure those are dereferenced in the input stream.

tar --hard-dereference -C /path/to/dir -c . | desync tar --input-format tar archive.catar -

Unpack a directory tree from an index file and store the output filesystem in a GNU tar file rather than the local filesystem. Instead of an archive file, the output can be given as '-' which will write to STDOUT.

desync untar -i -s /some/local/store --output-format=gnu-tar archive.caidx /path/to/archive.tar

Prune a store to only contain chunks that are referenced in the provided index files. Possible data loss.

desync prune -s /some/local/store index1.caibx index2.caibx

Start a chunk server serving up a local store via port 80.

desync chunk-server -s /some/local/store

Start a chunk server on port 8080 acting as proxy for other remote HTTP and SSH stores and populate a local cache.

desync chunk-server -s http://192.168.1.1/ -s ssh://192.168.1.2/store -c cache -l :8080

Start a chunk server with a store-file, this allows the configuration to be re-read on SIGHUP without restart.

# Create store file
echo '{"stores": ["http://192.168.1.1/"], "cache": "/tmp/cache"}` > stores.json

# Start the server
desync chunk-server --store-file stores.json -l :8080

# Modify
echo '{"stores": ["http://192.168.1.2/"], "cache": "/tmp/cache"}` > stores.json

# Reload
killall -1 desync

Start a writable index server, chunk a file and store the index.

server# desync index-server -s /mnt/indexes --writable -l :8080

client# desync make -s /some/store http://192.168.1.1:8080/file.vmdk.caibx file.vmdk

Start a TLS chunk server on port 443 acting as proxy for a remote chunk store in AWS with local cache. The credentials for AWS are expected to be in the config file under key https://s3-eu-west-3.amazonaws.com.

desync chunk-server -s s3+https://s3-eu-west-3.amazonaws.com/desync.bucket/prefix -c cache -l 127.0.0.1:https --cert cert.pem --key key.pem

Split a blob, store the chunks and create an index file.

desync make -s /some/local/store index.caibx /some/blob

Split a blob, create an index file and store the chunks in an S3 bucket named store.

S3_ACCESS_KEY=mykey S3_SECRET_KEY=mysecret desync make -s s3+http://127.0.0.1:9000/store index.caibx /some/blob

Index an existing local file. Does not create chunks

desync make image.raw.caibx /tmp/image.raw

Verify the index you just created

desync verify-index image.raw.caibx /tmp/image.raw

FUSE mount an index file. This will make the indexed blob available as file underneath the mount point. The filename in the mount matches the name of the index with the extension removed. In this example /some/mnt/ will contain one file index.

desync mount-index -s /some/local/store index.caibx /some/mnt

FUSE mount a chunked and remote index file. First a (small) index file is read from the index-server which is used to re-assemble a larger index file and pipe it into the 2nd command that then mounts it.

desync cat -s http://192.168.1.1/store http://192.168.1.2/small.caibx | desync mount-index -s http://192.168.1.1/store - /mnt/point

Long-running FUSE mount that may need to have its store setup changed without unmounting. This can be done by using the --store-file option rather than speicifying store+cache in the command line. The process will then reload the file when a SIGHUP is sent.

# Create the store file
echo '{"stores": ["http://192.168.1.1/"], "cache": "/tmp/cache"}` > stores.json

# Start the mount
desync mount-index --store-file stores.json index.caibx /some/mnt

# Modify the store setup
echo '{"stores": ["http://192.168.1.2/"], "cache": "/tmp/cache"}` > stores.json

# Reload
killall -1 desync

Show information about an index file to see how many of its chunks are present in a local store or an S3 store. The local store is queried first, S3 is only queried if the chunk is not present in the local store. The output will be in JSON format (--format=json) for easier processing in scripts.

desync info --format=json -s /tmp/store -s s3+http://127.0.0.1:9000/store /path/to/index

Start an HTTP chunk server that will store uncompressed chunks locally, configured via JSON config file, and serve uncompressed chunks over the network (-u option). This chunk server could be used as a cache, minimizing latency by storing and serving uncompressed chunks. Clients will need to be configured to request uncompressed chunks from this server.

# Chunk server
echo '{"store-options": {"/path/to/store/":{"uncompressed": true}}}' > /path/to/server.json

desync --config /path/to/server.json chunk-server -w -u -s /path/to/store/ -l :8080

# Client
echo '{"store-options": {"http://store.host:8080/":{"uncompressed": true}}}' > /path/to/client.json

desync --config /path/to/client.json cache -s sftp://remote.host/store -c http://store.host:8080/ /path/to/blob.caibx

HTTP chunk server using basic authorization. The server is configured to expect an Authorization header with the correct value in every request. The client configuration defines what the value should be on a per-server basis. The client config could be added to the default $HOME/.config/desync/config.json instead.

# Server
DESYNC_HTTP_AUTH="Bearer abcabcabc" desync chunk-server -s /path/to/store -l :8080

# Client
echo '{"store-options": {"http://127.0.0.1:8080/":{"http-auth": "Bearer abcabcabc"}}}' > /path/to/client.json

desync --config /path/to/client.json extract -s http://127.0.0.1:8080/ /path/to/blob.caibx /path/to/blob

HTTPS chunk server using key and certificate signed by custom CA.

# Building the CA and server certficate
openssl genrsa -out ca.key 4096
openssl req -x509 -new -nodes -key ca.key -sha256 -days 3650 -out ca.crt
openssl genrsa -out server.key 2048
openssl req -new -key server.key -out server.csr (Common Name should be the server name)
openssl x509 -req -in server.csr -CA ca.crt -CAkey ca.key -CAcreateserial -out server.crt -days 3650 -sha256

# Chunk server
desync chunk-server -s /path/to/store --key server.key --cert server.crt -l :8443

# Client
desync extract --ca-cert ca.crt -s https://hostname:8443/ image.iso.caibx image.iso

HTTPS chunk server with client authentication (mutual-TLS).

# Building the CA, server and client certficates
openssl genrsa -out ca.key 4096
openssl req -x509 -new -nodes -key ca.key -sha256 -days 3650 -out ca.crt
openssl genrsa -out server.key 2048
openssl req -new -key server.key -out server.csr (Common Name should be the server name)
openssl x509 -req -in server.csr -CA ca.crt -CAkey ca.key -CAcreateserial -out server.crt -days 3650 -sha256
openssl genrsa -out client.key 2048
openssl req -new -key client.key -out client.csr
openssl x509 -req -in client.csr -CA ca.crt -CAkey ca.key -CAcreateserial -out client.crt -days 3650 -sha256

# Chunk server
desync chunk-server -s /path/to/store --key server.key --cert server.crt --mutual-tls --client-ca ca.crt -l :8443

# Client
desync extract --client-key client.key --client-cert client.crt --ca-cert ca.crt -s https://hostname:8443/ image.iso.caibx image.iso

Links