glacier-cli

Fork of glacier-cli modified for uploading timemachine/clonezilla backups to glacier for offsite backup.

Background

Timemachine backups are saved in sparsebundle images, which are saved on disk as (encrypted) 8MB chunks - clonezilla can be configured to produce similar output.

Timemachine will overwrite (possibly delete?) existing chunks when data changes. Relying on local last modified time to identify such changes would be iffy at best; relying on checksums is much safer. Luckily, glacier uses hashes in order to guarantee data integrity on upload, and these hashes are persisted and accessible throughout the lifecycle of each archive.

This is basically the sole purpose of this fork - be aware of glacier hashes and rely on them as checksums for all local vs remote reconciliation operations.

Changes in this fork

• Persist archive checksums sent to glacier (tree hash) in the local archive db
• Use these checksums when comparing local vs remote (cached) files
• Reconcile filesystem changes vs cache, and automatically update/upload modified/new files
• Support uploading/deleting multiple files
• Incorporate checksums in existing sync operation (cache rebuild)

Example

 $ ./glacier.py archive multi-upload example-vault /local/dir

(Uploads all files in dir and its subdirectories to example-vault)

 $ ./glacier.py vault reconcile example-vault /local/dir

(Reconciles local files under dir against cached files for example-vault. Automatically update cache/glacier to reflect local changes if --commit is specified.)

Notes

• I am not using or testing with git-annex - this fork might break integration.
• For my use case, I am hoping I will never have to pull this data from glacier so retrieval is not a priority. Any glacier client (including this one) should be able to pull the archives. The only requirement is that the archives must be renamed with their description when saved locally (this client does so).
• TM default settings update backups every hour. TM must be turned off while uploading/syncing to guarantee that the remote archive is in a consistent state.
• Uploads can take a loooong time.. use tmux to reattach to the session and/or tee the output of multi-upload to a file, e.g: ./glacier.py archive multi-upload example-vault /local/dir 2>&1 | tee -a ./upload.log
• If a multi-upload fails (due to connection errors for example), --resume can be used to only upload files not previously uploaded.

Original readme

This tool provides a sysadmin-friendly command line interface to Amazon Glacier, turning Glacier into an easy-to-use storage backend. It automates tasks which would otherwise require a number of separate steps (job submission, polling for job completion and retrieving the results of jobs). It provides integration with git-annex, making Glacier even more useful.

glacier-cli uses Amazon Glacier's archive description field to keep friendly archive names, although you can also address archives directly by using their IDs. It keeps a local cache of archive IDs and their corresponding names, as well as housekeeping data to keep the cache up-to-date. This will save you time because you won't have to wait spend hours retrieving inventories all the time, and will save you mental effort because you won't have to keep track of the obtuse archive IDs yourself.

glacier-cli is fully interoperable with other applications using the same Glacier vaults. It can deal gracefully with vaults changing from other machines and/or other applications, and introduces no new special formats from the point of view of a vault.

Example with git-annex

$ echo 42 > example-content
$ git annex add example-content
add example-content (checksum...) ok
(Recording state in git...)
$ git commit -m'Add example-content'
[master cc632d1] Add example-content
 1 file changed, 1 insertion(+)
 create mode 120000 example/example-content

(the local annex now stores example-content)

$ git annex copy --to glacier example-content
copy example-content (gpg) (checking glacier...) (to glacier...) 
ok

(copying content to Amazon Glacier is straightforward)

$ git annex drop example-content
drop example-content (gpg) (checking glacier...) ok

(now the only copy of the data is in Amazon Glacier)

$ git annex get --from glacier example-content
get example-content (from glacier...) (gpg) 
glacier: queued retrieval job for archive 'GPGHMACSHA1--2945f64be96ccbb9feb4d8ff44ac9692fdbe654e'

  retrieve hook exited nonzero!
failed
git-annex: get: 1 failed

(this fails on the first attempt since the data isn't immediately available; but it does submit a job to Amazon Glacier requesting the data, so a later retry will work)

(...four hours later...)

$ git annex get --from glacier example-content
get example-content (from glacier...) (gpg) 
ok
$ cat example-content
42

(content successfully retrieved from Glacier)

Example without git-annex

$ glacier vault list

(empty result with zero exit status)

$ glacier vault create example-vault

(silently successful: like other Unix commands, only errors are noisy)

$ glacier vault list
example-vault

(this list is retrieved from Glacier; a relatively quick operation)

$ glacier archive list example-vault

(empty result with zero exit status; nothing is in our vault yet)

$ echo 42 > example-content
$ glacier archive upload example-vault example-content

(Glacier has now stored example-content in an archive with description example-content and in a vault called example-vault)

$ glacier archive list example-vault
example-content

(this happens instantly, since glacier-cli maintains a cached inventory)

$ rm example-content

(now the only place the content is stored is in Glacier)

$ glacier archive retrieve example-vault example-content
glacier: queued retrieval job for archive 'example-content'
$ glacier archive retrieve example-vault example-content
glacier: job still pending for archive 'example-content'
$ glacier job list
a/p 2012-09-19T21:41:35.238Z example-vault example-content
$ glacier archive retrieve --wait example-vault example-content

(...hours pass while Amazon retrieves the content...)

$ cat example-content
42

(content successfully retrieved from Glacier)

Costs

Before you use Amazon Glacier, you should make yourself familiar with how much it costs. Note that archive retrieval costs are complicated and may be a lot more than you expect. Files are uploaded in chunks, so uploading an archive can cause many requests. The default size of the parts is determined by the boto library; check DefaultPartSize in the documentation. Changes have been proposed to this tool in order to allow the user to specify the chunk size, but they have not been merged yet.

Installation

First ensure that boto is installed on your system.

Then clone this repository:

git clone git://github.com/basak/glacier-cli.git

and either, for all users:

sudo ln -s $PWD/glacier-cli/glacier.py /usr/local/bin/glacier

or for just yourself, if you have ~/bin in your path:

ln -s $PWD/glacier-cli/glacier.py ~/bin/glacier

Integration with git-annex

Using glacier-cli via git-annex is the easiest way to use Amazon Glacier from the CLI.

git-annex now has native glacier-cli integration. See the git-annex Glacier documentation and tutorial for details.

You probably want to set git-annex to only use glacier as a last resort in order to control your costs:

git config remote.glacier.annex-cost 1000

Copying to the remote works as normal. Retrieving from the remote initially fails after a job is queued. If you try again after the job is complete (usually around four hours), then retrieval should work successfully. You can monitor the status of the jobs using glacier job list; when the job status changes from p (pending) to d (done), a retrieval should work. Note that jobs expire from Amazon Glacier after around 24 hours or so.

glacier checkpresent cannot always check for certain that an archive definitely exists within Glacier. Vault inventories take hours to retrieve, and even when retrieved do not necessarily represent an up-to-date state. For this reason and as a compromise, glacier checkpresent will confirm to git-annex that an archive exists if it is known to have existed less than 60 hours ago. You may override this permitted lag interval with the --max-age option to glacier checkpresent.

Commands

• glacier vault list
• glacier vault create vault-name
• glacier vault sync [--wait] [--fix] [--max-age hours] vault-name
• glacier archive list vault-name
• glacier archive upload [--name archive-name] vault-name filename
• glacier archive retrieve [--wait] [-o filename] [--multipart-size bytes] vault-name archive-name
• glacier archive retrieve [--wait] [--multipart-size bytes] vault-name archive-name [archive-name...]
• glacier archive delete vault-name archive-name
• glacier job list

Delayed Completion

If you request an archive retrieval, then this requires a job which will take some number of hours to complete. You have one of two options:

1. If the command fails with a temporary failure—printed to stderr and with an exit status of EX_TEMPFAIL (75)—then a job is pending, and you must retry the command until it succeeds.
2. If you prefer to just wait, then use --wait (or retry with --wait if you didn't use it the first time). This will just do everything and exit when it is done. Amazon Glacier jobs typically take around four hours to complete.

Without --wait, glacier-cli will follow this logic:

1. Look for a suitable existing archive retrieval job.
2. If such a job exists and it is pending, then exit with a temporary failure.
3. If such a job exists and it has finished, then retrieve the data and exit with success.
4. Otherwise, submit a new job to retrieve the archive and exit with a temporary failure. Subsequent calls requesting the same archive will find this job and follow these same four steps with it, resulting in a downloaded archive when the job is complete.

Cache Reconstruction

glacier-cli follows the XDG Base Directory Specification and keeps its cache in ${XDG_CACHE_HOME:-$HOME/.cache}/glacier-cli/db.

After a disaster, or if you have modified a vault from another machine, you can reconstruct your cache by running:

$ glacier vault sync example-vault

This will set off an inventory job if required. This command is subject to delayed completion semantics as above but will also respond to --wait as needed.

By default, existing inventory jobs that completed more than 24 hours ago are ignored, since they may be out of date. You can override this with --max-age=hours. Specify --max-age=0 to force a new inventory job request.

Note that there is a lag between creation or deletion of an archive and the archive's corresponding appearance or disappearance in a subsequent inventory, since Amazon only periodically regenerates vault inventories. glacier-cli will show you newer information if it knows about it, but if you perform vault operations that do not update the cache (eg. on another machine, as another user, or from another program), then updates may take a while to show up here. You will need to run a vault sync operation after Amazon have updated your vault's inventory, which could be a good day or two after the operation took place.

If something doesn't go as expected (eg. an archive that glacier-cli knows it created fails to appear in the inventory after a couple of days, or an archive disappears from the inventory after it showed up there), then vault sync will warn you about it. You can use --fix to accept the correction and update the cache to match the official inventory.

Addressing Archives

Normally, you can just address an archive by its name (which, from Amazon's perspective, is the Glacier archive description).

However, you may end up with multiple archives with the same name, or archives with no name, since Amazon allows this. In this case, you can refer to an archive by ID instead by prefixing your reference with id:.

To avoid ambiguity, prefixing a reference with name: works as you would expect. If you end up with archive names or IDs that start with name: or id:, then you must use a prefix to disambiguate.

Using Pipes

Use glacier archive upload <vault> --name=<name> - to upload data from standard input. In this case you must use --name to name your archive correctly.

Use glacier archive retrieve <vault> <name> -o- to download data to standard output. glacier-cli will not output any data to standard output apart from the archive data in order to prevent corrupting the output data stream.

Future Directions

• Add resume functionality for uploads and downloads

Contact

• For bugs or feature requests please create a glacier-cli github issue.
• Reach me on Twitter: @robiebasak, but please tweet me an email address (or a reCAPTCHA mailhide URL, or some other way for me to reply) if my reply is likely to take more than 140 characters!