search

kbase / blobstore

Simple S3 backed file storage.
MIT License
0 stars 3 forks source link

readme

KBase Blob store

Build status (master): Build Status codecov

The blob store is a simple file storage service backed by an S3 compatible storage system such as Minio. Storing a file provides a key - currently a UUID - that allows retrival of the file when provided along with proper credentials. Once stored, files are immutable other than deletion.

The user is responsible for saving the key for use later - in the context of KBase, that means creating a handle for the file via the handle service and saving an object to the workspace containing that handle in an @id handle annotation, or saving the key directly in the workspace object in an @id bytestream annotation. See the workspace documentation for details; also the DataFileUtil module can assist with these functions in the context of KBase applications.

The API is nominally compatible with a minimal subset of the KBase fork of Shock's API. The vast majority of functions are not supported; only those required for the KBase codebase are included.

API Data structures

Node

This data structure is a subset of Shock's node data structure.

{
  "data": {
    "attributes": null,                               # DEPRECATED
    "created_on": "2019-05-30T23:50:19.000Z",
    "file": {
      "checksum": {
        "md5": "1b9554867d35f0d59e4705f6b2712cd1"
      },
      "name": "foo",                                  # Provided filename (see below)
      "size": 8
    },
    "format": "bar",                                  # Provided file format (see below)
    "id": "c39192c7-45b1-4fec-b196-5976d8e628f7",     # The node ID generated by the blobstore.
    "last_modified": "2019-05-30T23:50:19.000Z"
  },
  "error": null,
  "status": 200
}

attributes is deprecated, always null and is only provided for backwards compatibility reasons.

last_modified is always the same as created_on and is only included for backwards compatibility reasons. Unlike Shock, the blobstore does not take ACL modifications into account when setting the last_modified date.

ACL

This data structure is a subset of Shock's ACL data structure.

{
  "data": {
    "delete": [User],
    "owner": User,
    "public": {
      "delete": false,
      "read:" <true if the node is publically readable, false otherwise>,
      "write": false
    },
    "read": [User...],
    "write": [User],
  },
  "error": null,
  "status": 200
}

delete and write ACLs are deprecated and only provided for backwards compatibility reasons. They are always false for public access or contain only the node owner for standard ACLs.

A User is usually just the UUID assigned to the user by the blobstore, but when full verbosity (see below) is requested, the User data structure is:

{
  "uuid": <the user's UUID assigned by the blobstore>,
  "username": <the user's KBase account name>
}

Error

This data structure is identical to Shock's error data structure.

{
  "data": null,
  "error": [<error string>],
  "status": <http status code as an integer>
}

API

Authentication

Requests are authenticated by including the header Authorization: OAuth <kbase token> or including a cookie with the value of <kbase token> in the request.

The names of cookies that the server will check are set in the deployment configuration file.

The header takes precedence, then each cookie in the list in the configuration file in order.

Note that for backwards compatibility, incorrect or invalid authentication headers respond with a 400 HTTP code. Invalid cookies respond with the appropriate 401 code.

Root

GET /
{
  "deprecationwarning": "The id and version fields are deprecated.",
  "id": "Shock",
  "servername": "blobstore",
  "servertime": <server time in epoch milliseconds>,
  "serverversion": <server version>,
  "version": "0.9.6"
  "gitcommit": <git commit from which the server was built>
}

The id and version fields are deprecated and present only for backwards compatibility with Shock. The version field will not change.

Upload a file / create a node

AUTHORIZATION REQUIRED
Content-Length header required
POST /node[?filename=<filename>&format=<file format>]
<file content>

RETURNS: a Node.

The Content-Length header must be present and accurate.

PUT is also supported - but is not idempotent - in order to ease using the curl -T option:

curl -H "Authorization: OAuth $KBASE_TOKEN" -T mylittlefile
  "http://<host>/node?filename=mylittlefile&format=text"

filename can be at most 256 characters consisting of only unicode alphanumerics, space, and the characters [ ] ( ) = . - _.
format can be at most 100 characters consisting of only unicode alphanumerics and the characters - _.

Copy a node

AUTHORIZATION REQUIRED
POST /node/<id>/copy

RETURNS: a Node.

Get a node

AUTHORIZATION OPTIONAL
GET /node/<id>

RETURNS: a Node.

Get a node's ACLs

AUTHORIZATION OPTIONAL
GET /node/<id>/acl[?verbosity=full]

RETURNS: an ACL.

Download a file from a node

AUTHORIZATION OPTIONAL
GET /node/<id>?download[_raw][&seek=#][&length=#][&del]

RETURNS: the file content.

?download_raw, as opposed to ?download, causes the Content-Disposition header to be omitted.

seek causes the first # bytes of the file to be skipped. A seek value greater than or equal to the file size is an error. Defaults to 0.

length determines the number of bytes of the file to return after skipping seek bytes. length may be greater than the remaining file length. Defaults to 0, which indicates that the remainder of the file should be returned.

del causes the node to be deleted once the file contents have been streamed. The user must be the node owner or a service administrator. Note this is playing very fast and loose with the semantics of an HTTP GET.

Set a node to be publicly readable

AUTHORIZATION REQUIRED
PUT /node/<id>/acl/public_read[?verbosity=full]

RETURNS: an ACL.

Set a node to be privately readable

AUTHORIZATION REQUIRED
DELETE /node/<id>/acl/public_read[?verbosity=full]

RETURNS: an ACL.

Add users to a node's read ACL

AUTHORIZATION REQUIRED
PUT /node/<id>/acl/read?users=<comma separated list of KBase user names>[&verbosity=full]

RETURNS: an ACL.

Remove users from a node's read ACL

AUTHORIZATION REQUIRED
DELETE /node/<id>/acl/read?users=<comma separated list of KBase user names>[&verbosity=full]

RETURNS: an ACL.

Change a node's owner

AUTHORIZATION REQUIRED
PUT /node/<id>/acl/owner?users=<KBase user name>[&verbosity=full]

RETURNS: an ACL.

The users parameter must contain a single user name.

Upload a file / create a node via a MIME multipart form

This upload method is provided for Shock compatibilty. It is recommended that the prior upload method is used rather than this one.

AUTHORIZATION REQUIRED
POST /node
<multipart form>

RETURNS: a Node.

The form MUST contain a part called upload where the part contents are the file to be uploaded. The part MUST have an accurate Content-Length header specifing the size of the file, not the entire multipart form.

The form may contain a part called format where the part contents are the format of the file, equivalent to the format query parameter for the standard upload method and with the same restrictions. The format part MUST come before the upload part.

Any file name provided in the Content-Disposition header has the same restrictions as the filename parameter for the standard upload method.

Curl example

curl -H "Authorization: OAuth $KBASE_TOKEN" \
  -F "upload=@mydata.fasta;headers=\"Content-Length: 67452\"" \
  http://<host>/node

Python example

import os
import requests
from requests_toolbelt.multipart.encoder import MultipartEncoder

df = open(filename, 'rb')
files = {'upload': (filename, df, None, {'Content-Length': os.path.getsize(filename)})}
mpe = MultipartEncoder(fields=files)
headers = {'Content-Type': mpe.content_type,
           'authorization': 'OAuth ' + token}

res = requests.post('http://<host>/node', headers=headers, data=mpe, stream=True)
res.json()

Java example

package blobstoreclienttest;

import java.io.IOException;
import java.io.InputStream;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.Paths;

import org.apache.commons.io.IOUtils;
import org.apache.http.client.ClientProtocolException;
import org.apache.http.client.methods.CloseableHttpResponse;
import org.apache.http.client.methods.HttpPost;
import org.apache.http.entity.mime.FormBodyPartBuilder;
import org.apache.http.entity.mime.MultipartEntityBuilder;
import org.apache.http.entity.mime.content.InputStreamBody;
import org.apache.http.impl.client.CloseableHttpClient;
import org.apache.http.impl.client.HttpClients;

public class blobstoreclient {

    public static void main(final String[] args) throws ClientProtocolException, IOException {
        final String fileName = args[0];
        final String token = args[1];
        // probably don't want to use the default client for most applications
        final CloseableHttpClient cli = HttpClients.createDefault();

        final HttpPost htp = new HttpPost("http://<host>/node");
        htp.setHeader("authorization", "OAuth " + token);

        final Path p = Paths.get(fileName);
        final MultipartEntityBuilder mpeb = MultipartEntityBuilder.create();
        final InputStream in = Files.newInputStream(p);

        mpeb.addPart(FormBodyPartBuilder.create()
                .setName("upload")
                .addField("Content-Length", "" + Files.size(p))
                .setBody(new InputStreamBody(in, p.getFileName().toString())).build());

        htp.setEntity(mpeb.build());

        final CloseableHttpResponse response = cli.execute(htp);
        in.close();
        IOUtils.copy(response.getEntity().getContent(), System.out);
        response.close();
    }
}

Copy a node via a MIME multipart form

This copy method is provided for Shock compatibilty. It is recommended that the prior copy method is used rather than this one.

AUTHORIZATION REQUIRED
POST /node
<multipart form>

RETURNS: a Node.

The multipart form must have exactly one part with the name copy_data and the value the id of the node to copy.

Curl example:

curl -H "Authorization: OAuth $KBASE_TOKEN" -F "copy_data=<node id>" http://<host>/node/

Requirements:

Running the server:

To build the git commit into the server:

export GIT_COMMIT=$(git rev-list -1 HEAD) 
    && go build -ldflags "-X main.gitCommit=$GIT_COMMIT" app/blobstore.go

Developers

Testing

Copy test.cfg.example to test.cfg and adjust the values as necessary.

BLOBSTORE_TEST_CFG=[absolute path to test.cfg] go test ./...

Each package gets its own working directory during tests so the path to the test.cfg file cannot be relative.

Mocks are generated with https://github.com/vektra/mockery v1.0.0.

Known issues

TODO

S3/Minio experimental server

While exploring upload speeds with various upload methods, this server was generated.