Compare commits

...

2 Commits

Author SHA1 Message Date
Alexander Chan b353640b8e ft: add gcp design doc 2018-04-03 22:55:08 -07:00
Alexander Chan 7de1841a3e ft: add gcp instructions to public cloud doc 2018-04-03 21:23:54 -07:00
2 changed files with 291 additions and 2 deletions

View File

@ -11,7 +11,8 @@ This section of the documentation shows you how to set up our currently
supported public cloud backends:
- `Amazon S3 <#aws-s3-as-a-data-backend>`__ ;
- `Microsoft Azure <#microsoft-azure-as-a-data-backend>`__ .
- `Microsoft Azure <#microsoft-azure-as-a-data-backend>`__ ;
- `Google Cloud Storage <#google-cloud-storage-as-a-data-backend>`__ .
For each public cloud backend, you will have to edit your CloudServer
:code:`locationConfig.json` and do a few setup steps on the applicable public
@ -362,6 +363,199 @@ For more informations, refer to our template `~/.s3cfg <./CLIENTS/#s3cmd>`__ .
Pre-existing objects in your MS Azure container can unfortunately not be
accessed by CloudServer at this time.
Google Cloud Storage as a data backend
--------------------------------------
From the Google Cloud Console
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
From the Google Cloud Store, create a two buckets for this new location
constraint: one bucket where you will host your data and the other for
performing multipart upload.
You will also need to get one of your Interoperability Credentials and provide
it to CloudServer.
This can be found in the Google Cloud Storage "Settings" tab then under
"Interopability".
In this example, our buckets will be ``zenkobucket`` and ``zenkompubucket``.
From the CloudServer repository
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
locationConfig.json
^^^^^^^^^^^^^^^^^^^
Edit this file to add a new location constraint. This location constraint will
constain the information for the Google Cloud Storage bucket to which you will
be writing your data whenever you create a CloudServer bucket in this location.
There are a few configurable options here:
- :code:`type` : set to :code:`gcp` to indicate this location constraint is
writing data to Google Cloud Storage;
- :code:`legacyAwsBehavior` : set to :code:`true` to indicate this region should
behave like AWS S3 :code:`us-east-1` region, set to :code:`false` to indicate
this region should behave like any other AWS S3 region;
- :code:`bucketName` : set to an *existing bucket* in your Google Cloud Storage
Account; this is the bucket in which your data will be stored for this
location constraint;
- :code:`mpuBucketName` : set to an *existing bucket* in your Google Cloud
Storage Account; this is the bucket in which parts for multipart uploads will
be stored for this location constraint;
- :code:`gcpEndpoint` : set to your bucket's endpoint, usually :code:`storage.googleapis.com`;
- :code:`bucketMatch` : set to :code:`true` if you want your object name to be same
in your local bucket and your Google Cloud Storage bucket; set to :code:`false`
if you want your object name to be of the form :code:`{{localBucketName}}/{{objectname}}`
in your Google Cloud Storage hosted bucket;
- :code:`credentialsProfile` and :code:`credentials` are two ways to provide
your Google Cloud Storage Interoperability credentials for that bucket,
*use only one of them* :
- :code:`credentialsProfile` : set to the profile name allowing you to access
your Google Cloud Storage bucket from your :code:`~/.aws/credentials` file;
- :code:`credentials` : set the two fields inside the object (:code:`accessKey`
and :code:`secretKey`) to their respective values from your Google Cloud Storage
Interoperability credentials.
.. code:: json
(...)
"gcp-test": {
"type": "gcp",
"legacyAwsBehavior": true,
"details": {
"awsEndpoint": "storage.googleapis.com",
"bucketName": "zenkobucket",
"mpuBucketName": "zenkompubucket",
"bucketMatch": true,
"credentialsProfile": "zenko"
}
},
(...)
.. code:: json
(...)
"gcp-test": {
"type": "gcp",
"legacyAwsBehavior": true,
"details": {
"awsEndpoint": "storage.googleapis.com",
"bucketName": "zenkobucket",
"bucketMatch": true,
"mpuBucketName": "zenkompubucket",
"credentials": {
"accessKey": "WHDBFKILOSDDVF78NPMQ",
"secretKey": "87hdfGCvDS+YYzefKLnjjZEYstOIuIjs/2X72eET"
}
}
},
(...)
.. WARNING::
If you set :code:`bucketMatch` to :code:`true`, we strongly advise that you
only have one local bucket per Google Cloud Storage location.
Without :code:`bucketMatch` set to :code:`false`, your object names in your
Google Cloud Storage bucket will not be prefixed with your Cloud Server
bucket name. This means that if you put an object :code:`foo` to your
CloudServer bucket :code:`zenko1` and you then put a different :code:`foo` to
your CloudServer bucket :code:`zenko2` and both :code:`zenko1` and
:code:`zenko2` point to the same Google Cloud Storage bucket, the second
:code:`foo` will overwrite the first :code:`foo`.
~/.aws/credentials
^^^^^^^^^^^^^^^^^^
.. TIP::
If you explicitly set your :code:`accessKey` and :code:`secretKey` in the
:code:`credentials` object of your :code:`gcp` location in your
:code:`locationConfig.json` file, you may skip this section
Make sure your :code:`~/.aws/credentials` file has a profile matching the one
defined in your :code:`locationConfig.json`. Following our previous example, it
would look like:
.. code:: shell
[zenko]
aws_access_key_id=WHDBFKILOSDDVF78NPMQ
aws_secret_access_key=87hdfGCvDS+YYzefKLnjjZEYstOIuIjs/2X72eET
Start the server with the ability to write to Google Cloud Storage
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Inside the respository, once all the files have been edited, you shoul dbe able
to start the server and start writing data to Google Cloud Storage through
CloudServer.
.. code:: shell
# Start the server locally
$> S3DATA=multiple npm start
Run the server as a docker container with the ability to write to Google Cloud Storage
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
.. TIP::
If you set the :code:`credentials` object in you
:code:`locationConfig.json` file, you don't need to mount your
:code:`.aws/credentials` file
Mount all the files that have been edited to override defaults, and do a
standard Docker run; then you can start wiriting to Google Cloud Storage through
CloudServer.
.. code:: shell
# Start the server in a Docker container
$> sudo docker run -d --name CloudServer \
-v $(pwd)/data:/usr/src/app/localData \
-v $(pwd)/metadata:/usr/src/app/localMetadata \
-v $(pwd)/locationConfig.json:/usr/src/app/locationConfig.json \
-v $(pwd)/conf/authdata.json:/usr/src/app/conf/authdata.json \
-v ~/.aws/credentials:/root/.aws/credentials \
-e S3DATA=multiple -e ENDPOINT=http://localhost -p 8000:8000
-d scality/s3server
Testing: put an object to Google Cloud Storage using CloudServer
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In order to start testing pushing to AWS S3, you will need to create a local
bucket in the AWS S3 location constraint - this local bucket will only store the
metadata locally, while both the data and any user metadata (:code:`x-amz-meta`
headers sent with a PUT object, and tags) will be stored on AWS S3.
This example is based on all our previous steps.
.. code:: shell
# Create a local bucket storing data in AWS S3
$> s3cmd --host=127.0.0.1:8000 mb s3://zenkobucket --region=gcp-test
# Put an object to Google Cloud Storage, and store the metadata locally
$> s3cmd --host=127.0.0.1:8000 put /etc/hosts s3://zenkobucket/testput
upload: '/etc/hosts' -> 's3://zenkobucket/testput' [1 of 1]
330 of 330 100% in 0s 380.87 B/s done
# List locally to check you have the metadata
$> s3cmd --host=127.0.0.1:8000 ls s3://zenkobucket
2017-10-23 10:26 330 s3://zenkobucket/testput
Then, from the Google Cloud Console, if you go into your bucket, you should see
your newly uploaded object:
.. figure:: ../res/gcp-console-successful-put.png
:alt: Google Cloud Storage Console upload example
Troubleshooting
~~~~~~~~~~~~~~~
Make sure your :code:`~/.s3cfg` file has credentials matching your local
CloudServer credentials defined in :code:`conf/authdata.json`. By default, the
access key is :code:`accessKey1` and the secret key is :code:`verySecretKey1`.
For more informations, refer to our template `~/.s3cfg <./CLIENTS/#s3cmd>`__ .
Pre-existing objects in your Google Cloud Storage hosted bucket can
unfortunately not be accessed by CloudServer at this time.
For any data backend
--------------------

95
lib/data/external/GCP/DESIGN_GCP.md vendored Normal file
View File

@ -0,0 +1,95 @@
## Google Cloud Storage Backend
### Overall Design
The Google Cloud Storage backend is implemented using the `aws-sdk` service
class for AWS compatible methods. The structure of these methods are
described in the `gcp-2017-11-01.api.json` file: request inputs, response
outputs, and required parameters. For non-compatible methods, helper methods are
implemented to perform the requests; these can be found under the `GcpApis`
directory.
The implement GCP Service is designed to work as close as possible to the AWS
service.
### Object Tagging
Google Cloud Storage does not have object-level tagging methods.
To be compatible with S3, object tags will be stored as metadata on
Google Cloud Storage.
### Multipart Upload
Google Cloud Storage does not have AWS S3 multipart upload methods, but there
are methods for merging multiple objects into a single composite object.
Utilizing these available methods, GCP is able to perform parallel uploads for
large uploads; however, due to limits set by Google Cloud Storage, the maximum
number of parts possible for a single upload is 1024 (AWS limit is 10000).
As Google Cloud Storage does not have methods for managing mutlipart uploads,
each part is uploaded as a single object in a Google Cloud Bucket.
Because of this, a secondary bucket for handling MPU parts is required for
a GCP multipart upload. The MPU bucket will serve to hide uploaded parts from
being listed as items of the main bucket as well as handling parts of multiple
in-progress mutlipart uploads.
<!--
<p style='font-size: 12'>
** The Google Cloud Storage method used for combining multipart objects into a
single object is the `compose` methods.<br/>
** <a>https://cloud.google.com/storage/docs/xml-api/put-object-compose</a>
</p>
-->
#### Multipart Upload Methods Design:
+ **inititateMultipartUpload**:
In `initiateMultipartUpload`, new multipart uploads will generate a prefix with
the scheme of `${objectKeyName}-${uploadI}` and each object related to an MPU
will be prefixed with it. This method will also create an `init` file that will
store the metadata related to an MPU for later assignment to the completed
object.
+ **uploadPart**:
`uploadPart` will prefix the upload with the MPU prefix then perform a
`putObject` request to Google Cloud Storage
+ **uploadPartCopy**:
`uploadPartCopy` will prefix the copy upload with the MPU prefix then perform a
`copyObject` request to Google Cloud Storage
+ **abortMultipartUpload**:
`abortMultipartUpload` will perform the action of removing all objects related
to a multipart upload from the MPU bucket. It does this by first making a
`listObjectVersions` request to GCP to list all parts with the
related MPU-prefix then performing a `deleteObject` request on each of the
objects received.
+ **completeMultipartUpload**:
`completeMultipartUpload` will perform the action of combining the given parts
to be create the single composite object. This method consists of multiple
steps, due to the limitations of the Google Cloud Storage `compose` method:
+ compose round 1: multiple compose calls to merge, at max, 32 objects into
a single subpart.
+ compose round 2: multiple compose calls to merge the subpart generated
in compose round 1 to create the final completed object
+ generate MPU ETag: generate the multipart etag that will be returned as
part of the completeMultipartUpload response
+ copy to main: retrieve the metadata stored in the `init` file created in
`initiateMultipartUpload` to be assigned the completed object and copy the
the completed object from the MPU bucket to the Main bucket
+ cleanUp: remove all objects related to a MPU
### Limitations
+ GCP multipart uploads are limited to 1024 parts
+ Each `compose` can merge up to 32 objects per request
+ As Google Cloud Storage doesn't have AWS style MPU methods, GCP MPU will
require a secondary bucket to perform multipart uploads
+ GCP doesn't not have object-level tagging methods; AWS style tags are stored
as metadata on Google Cloud Storage
More information can be found at:
+ https://cloud.google.com/storage/docs/xml-api/overview;
+ https://cloud.google.com/storage/quotas;
+ https://cloud.google.com/storage/docs/xml-api/put-object-compose;