pegasus-s3 — Upload, download, delete objects in Amazon S3
pegasus-s3 ls [options] URL
pegasus-s3 mkdir [options] URL…
pegasus-s3 rmdir [options] URL…
pegasus-s3 rm [options] [URL…]
pegasus-s3 put [options] FILE URL
pegasus-s3 get [options] URL [FILE]
pegasus-s3 lsup [options] URL
pegasus-s3 rmup [options] URL [UPLOAD]
pegasus-s3 cp [options] SRC… DEST
pegasus-s3 is a client for the Amazon S3 object storage service and any other storage services that conform to the Amazon S3 API, such as Eucalyptus Walrus.
- -h , --help
- Show help message for subcommand and exit
- -d , --debug
- Turn on debugging
- -v , --verbose
- Show progress messages
- -C FILE , --conf=FILE
- Path to configuration file
- -f , --force
- If the URL does not exist, then ignore the error.
- -F FILE , --file=FILE
- File containing a list of URLs to delete
- -r , --recursive
- Upload all files in the directory named FILE to keys with prefix URL.
- -c X , --chunksize=X
- Set the chunk size for multipart uploads to X MB. A value of 0 disables multipart uploads. The default is 10MB, the min is 5MB and the max is 1024MB. This parameter only applies for sites that support multipart uploads (see multipart_uploads configuration parameter in the CONFIGURATION section). The maximum number of chunks is 10,000, so if you are uploading a large file, then the chunk size is automatically increased to enable the upload. Choose smaller values to reduce the impact of transient failures.
- -p N , --parallel=N
- Use N threads to upload FILE in parallel. The default value is 4, which enables parallel uploads with 4 threads. This parameter is only valid if the site supports mulipart uploads and the --chunksize parameter is not 0. Otherwise parallel uploads are disabled.
- -b , --create-bucket
- Create the destination bucket if it does not already exist
- -r , --recursive
- Download all keys that match URL exactly or begin with URL+"/". For example, pegasus-s3 get -r s3://u@h/bucket/key will match both key and key/foo but not keyfoo. Since S3 allows names to exist as both keys (the bare key) and folders (the key in key/foo), but file systems do not, you will get an error when using -r/--recursive on a bucket that contains such duplicate names. An entire bucket can be downloaded at once by specifying only the bucket name in URL.
- -c X , --chunksize=X
- Set the chunk size for parallel downloads to X megabytes. A value of 0 will avoid chunked reads. This option only applies for sites that support ranged downloads (see ranged_downloads configuration parameter). The default chunk size is 10MB, the min is 1MB and the max is 1024MB. Choose smaller values to reduce the impact of transient failures.
- -p N , --parallel=N
- Use N threads to upload FILE in parallel. The default value is 4, which enables parallel downloads with 4 threads. This parameter is only valid if the site supports ranged downloads and the --chunksize parameter is not 0. Otherwise parallel downloads are disabled.
pegasus-s3 has several subcommands for different storage service operations.
- The help subcommand lists all available subcommands.
- The ls subcommand lists the contents of a URL. If the URL does not contain a bucket, then all the buckets owned by the user are listed. If the URL contains a bucket, but no key, then all the keys in the bucket are listed. If the URL contains a bucket and a key, then all keys in the bucket that begin with the specified key are listed.
- The mkdir subcommand creates one or more buckets.
- The rmdir subcommand deletes one or more buckets from the storage service. In order to delete a bucket, the bucket must be empty.
- The rm subcommand deletes one or more keys from the storage service.
The put subcommand stores the file specified by FILE in the storage service under the bucket and key specified by URL. If the URL contains a bucket, but not a key, then the file name is used as the key. If URL ends with a "/", then the file name is appended to the URL to create the key name (e.g. pegasus-s3 put foo s3://u@h/bucket/key will create a key called "key", while pegasus-s3 put foo s3://u@h/bucket/key/ will create a key called "key/foo". The same is true of directories when used with the -r/--recursive option.
The put subcommand can do both chunked and parallel uploads if the service supports multipart uploads (see multipart_uploads in the CONFIGURATION section). Currently only Amazon S3 supports multipart uploads.
This subcommand will check the size of the file to make sure it can be stored before attempting to store it.
Chunked uploads are useful to reduce the probability of an upload failing. If an upload is chunked, then pegasus-s3 issues separate PUT requests for each chunk of the file. Specifying smaller chunks (using --chunksize) will reduce the chances of an upload failing due to a transient error. Chunksizes can range from 5 MB to 1GB (chunk sizes smaller than 5 MB produced incomplete uploads on Amazon S3). The maximum number of chunks for any single file is 10,000, so if a large file is being uploaded with a small chunksize, then the chunksize will be increased to fit within the 10,000 chunk limit. By default, the file will be split into 10 MB chunks if the storage service supports multipart uploads. Chunked uploads can be disabled by specifying a chunksize of 0. If the upload is chunked, then each chunk is retried independently under transient failures. If any chunk fails permanently, then the upload is aborted.
Parallel uploads can increase performance for services that support multipart uploads. In a parallel upload the file is split into N chunks and each chunk is uploaded concurrently by one of M threads in first-come, first-served fashion. If the chunksize is set to 0, then parallel uploads are disabled. If M > N, then the actual number of threads used will be reduced to N. The number of threads can be specified using the --parallel argument. If --parallel is 1, then only a single thread is used. The default value is 4. There is no maximum number of threads, but it is likely that the link will be saturated by 4 to 8 threads.
Under certain circumstances, when a multipart upload fails it could leave behind data on the server. When a failure occurs the put subcommand will attempt to abort the upload. If the upload cannot be aborted, then a partial upload may remain on the server. To check for partial uploads run the lsup subcommand. If you see an upload that failed in the output of lsup, then run the rmup subcommand to remove it.
The get subcommand retrieves an object from the storage service identified by URL and stores it in the file specified by FILE. If FILE is not specified, then the part of the key after the last "/" is used as the file/directory name, and the results are placed in the current working directory. If FILE ends with a "/", then the last component of the key name is appended to FILE to create the output path (e.g. pegasus-s3 get s3://u@h/bucket/key /tmp/ will create a file called /tmp/key while pegasus-s3 get s3://u@h/bucket/key /tmp/foo will put the contents of key in a file called /tmp/foo). The same is true of folders/directories with the -r/--recursive option.
The get subcommand can do both chunked and parallel downloads if the service supports ranged downloads (see ranged_downloads in the CONFIGURATION section). Currently only Amazon S3 has good support for ranged downloads. Eucalyptus Walrus supports ranged downloads, but version 1.6 is inconsistent with the Amazon interface and has a bug that causes ranged downloads to hang in some cases. It is recommended that ranged downloads not be used with Walrus 1.6.
Chunked downloads can be used to reduce the probability of a download failing. When a download is chunked, pegasus-s3 issues separate GET requests for each chunk of the file. Specifying smaller chunks (using --chunksize) will reduce the chances that a download will fail to do a transient error. Chunk sizes can range from 1 MB to 1 GB. By default, a download will be split into 10 MB chunks if the site supports ranged downloads. Chunked downloads can be disabled by specifying a --chunksize of 0. If a download is chunked, then each chunk is retried independently under transient failures. If any chunk fails permanently, then the download is aborted.
Parallel downloads can increase performance for services that support ranged downloads. In a parallel download, the file to be retrieved is split into N chunks and each chunk is downloaded concurrently by one of M threads in a first-come, first-served fashion. If the chunksize is 0, then parallel downloads are disabled. If M > N, then the actual number of threads used will be reduced to N. The number of threads can be specified using the --parallel argument. If --parallel is 1, then only a single thread is used. The default value is 4. There is no maximum number of threads, but it is likely that the link will be saturated by 4 to 8 threads.
The lsup subcommand lists active multipart uploads. The URL specified should point to a bucket. This command is only valid if the site supports multipart uploads. The output of this command is a list of keys and upload IDs.
This subcommand is used with rmup to help recover from failures of multipart uploads.
The rmup subcommand cancels and active upload. The URL specified should point to a bucket, and UPLOAD is the long, complicated upload ID shown by the lsup subcommand.
This subcommand is used with lsup to recover from failures of multipart uploads.
- The cp subcommand copies keys on the server. Keys cannot be copied between accounts.
All URLs for objects stored in S3 should be specified in the following format:
The protocol part can be s3:// or s3s://. If s3s:// is used, then pegasus-s3 will force the connection to use SSL and override the setting in the configuration file. If s3:// is used, then whether the connection uses SSL or not is determined by the value of the endpoint variable in the configuration for the site.
The USER@SITE part is required, but the BUCKET and KEY parts may be optional depending on the context.
The USER@SITE portion is referred to as the “identity”, and the SITE portion is referred to as the “site”. Both the identity and the site are looked up in the configuration file (see CONFIGURATION) to determine the parameters to use when establishing a connection to the service. The site portion is used to find the host and port, whether to use SSL, and other things. The identity portion is used to determine which authentication tokens to use. This format is designed to enable users to easily use multiple services with multiple authentication tokens. Note that neither the USER nor the SITE portion of the URL have any meaning outside of pegasus-s3. They do not refer to real usernames or hostnames, but are rather handles used to look up configuration values in the configuration file.
The BUCKET portion of the URL is the part between the 3rd and 4th slashes. Buckets are part of a global namespace that is shared with other users of the storage service. As such, they should be unique.
The KEY portion of the URL is anything after the 4th slash. Keys can include slashes, but S3-like storage services do not have the concept of a directory like regular file systems. Instead, keys are treated like opaque identifiers for individual objects. So, for example, the keys a/b and a/c have a common prefix, but cannot be said to be in the same directory.
Some example URLs are:
s3://ewa@amazon s3://juve@skynet/gideon.isi.edu s3://juve@magellan/pegasus-images/centos-5.5-x86_64-20101101.part.1 s3s://ewa@amazon/pegasus-images/data.tar.gz
Each user should specify a configuration file that pegasus-s3 will use to look up connection parameters and authentication tokens.
This client will look in the following locations, in order, to locate the user’s configuration file:
- The -C/--conf argument
- The S3CFG environment variable
If it does not find the configuration file in one of these locations it will fail with an error. The $HOME/.s3cfg location is only supported for backward-compatibility. $HOME/.pegasus/s3cfg should be used instead.
Configuration File Format
The configuration file is in INI format and contains two types of entries.
The first type of entry is a site entry, which specifies the configuration for a storage service. This entry specifies the service endpoint that pegasus-s3 should connect to for the site, and some optional features that the site may support. Here is an example of a site entry for Amazon S3:
[amazon] endpoint = http://s3.amazonaws.com/
The other type of entry is an identity entry, which specifies the authentication information for a user at a particular site. Here is an example of an identity entry:
[pegasus@amazon] access_key = 90c4143642cb097c88fe2ec66ce4ad4e secret_key = a0e3840e5baee6abb08be68e81674dca
It is important to note that user names and site names used are only logical—they do not correspond to actual hostnames or usernames, but are simply used as a convenient way to refer to the services and identities used by the client.
The configuration file should be saved with limited permissions. Only the owner of the file should be able to read from it and write to it (i.e. it should have permissions of 0600 or 0400). If the file has more liberal permissions, then pegasus-s3 will fail with an error message. The purpose of this is to prevent the authentication tokens stored in the configuration file from being accessed by other users.
- endpoint (site)
- The URL of the web service endpoint. If the URL begins with https, then SSL will be used.
- max_object_size (site)
- The maximum size of an object in GB (default: 5GB)
- multipart_uploads (site)
- Does the service support multipart uploads (True/False, default: False)
- ranged_downloads (site)
- Does the service support ranged downloads? (True/False, default: False)
- access_key (identity)
- The access key for the identity
- secret_key (identity)
- The secret key for the identity
This is an example configuration that specifies a two sites (amazon and
magellan) and three identities (
voeckler@magellan). For the amazon site the maximum object size
is 5TB, and the site supports both multipart uploads and ranged downloads,
so both uploads and downloads can be done in parallel.
[amazon] endpoint = https://s3.amazonaws.com/ max_object_size = 5120 multipart_uploads = True ranged_downloads = True [pegasus@amazon] access_key = 90c4143642cb097c88fe2ec66ce4ad4e secret_key = a0e3840e5baee6abb08be68e81674dca [magellan] # NERSC Magellan is a Eucalyptus site. It doesn't support multipart uploads, # or ranged downloads (the defaults), and the maximum object size is 5GB # (also the default) endpoint = https://18.104.22.168:8773/services/Walrus [juve@magellan] access_key = quwefahsdpfwlkewqjsdoijldsdf secret_key = asdfa9wejalsdjfljasldjfasdfa [voeckler@magellan] # Each site can have multiple associated identities access_key = asdkfaweasdfbaeiwhkjfbaqwhei secret_key = asdhfuinakwjelfuhalsdflahsdl
List all buckets owned by identity user@amazon:
$ pegasus-s3 ls s3://user@amazon
List the contents of bucket bar for identity user@amazon:
$ pegasus-s3 ls s3://user@amazon/bar
List all objects in bucket bar that start with hello:
$ pegasus-s3 ls s3://user@amazon/bar/hello
Create a bucket called mybucket for identity user@amazon:
$ pegasus-s3 mkdir s3://user@amazon/mybucket
Delete a bucket called mybucket:
$ pegasus-s3 rmdir s3://user@amazon/mybucket
Upload a file foo to bucket bar:
$ pegasus-s3 put foo s3://user@amazon/bar/foo
Download an object foo in bucket bar:
$ pegasus-s3 get s3://user@amazon/bar/foo foo
Upload a file in parallel with 4 threads and 100MB chunks:
$ pegasus-s3 put --parallel 4 --chunksize 100 foo s3://user@amazon/bar/foo
Download an object in parallel with 4 threads and 100MB chunks:
$ pegasus-s3 get --parallel 4 --chunksize 100 s3://user@amazon/bar/foo foo
List all partial uploads for bucket bar:
$ pegasus-s3 lsup s3://user@amazon/bar
Remove all partial uploads for bucket bar:
$ pegasus-s3 rmup --all s3://user@amazon/bar
pegasus-s3 returns a zero exist status if the operation is successful. A non-zero exit status is returned in case of failure.
Pegasus Team http://pegasus.isi.edu