frostfs-http-gw/README.md
Alex Vanin d0f6baa44b [#2] Return shields
Signed-off-by: Alex Vanin <a.vanin@yadro.com>
2023-06-07 15:36:16 +00:00

23 KiB

FrostFS logo

FrostFS is a decentralized distributed object storage integrated with the NEO Blockchain.


Report Release License

FrostFS HTTP Gateway

FrostFS HTTP Gateway bridges FrostFS internal protocol and HTTP standard.

  • you can download one file per request from the FrostFS Network
  • you can upload one file per request into the FrostFS Network

See available routes in specification.

Installation

go install git.frostfs.info/TrueCloudLab/frostfs-http-gw

Or you can call make to build it from the cloned repository (the binary will end up in bin/frostfs-http-gw). To build frostfs-http-gw binary in clean docker environment, call make docker/bin/frostfs-http-gw.

Other notable make targets:

dep          Check and ensure dependencies
image        Build clean docker image
dirty-image  Build dirty docker image with host-built binaries
fmt          Format the code
lint         Run linters
version      Show current version

Or you can also use a Docker image provided for the released (and occasionally unreleased) versions of the gateway (:latest points to the latest stable release).

Execution

HTTP gateway itself is not a FrostFS node, so to access FrostFS it uses node's gRPC interface and you need to provide some node that it will connect to. This can be done either via -p parameter or via HTTP_GW_PEERS_<N>_ADDRESS and HTTP_GW_PEERS_<N>_WEIGHT environment variables (the gate supports multiple FrostFS nodes with weighted load balancing).

If you launch HTTP gateway in bundle with frostfs-dev-env, you can get the IP address of the node in the output of make hosts command (with s0*.frostfs.devenv name).

These two commands are functionally equivalent, they run the gate with one backend node (and otherwise default settings):

$ frostfs-http-gw -p 192.168.130.72:8080
$ HTTP_GW_PEERS_0_ADDRESS=192.168.130.72:8080 frostfs-http-gw

It's also possible to specify uri scheme (grpc or grpcs) when using -p:

$ frostfs-http-gw -p grpc://192.168.130.72:8080
$ HTTP_GW_PEERS_0_ADDRESS=grpcs://192.168.130.72:8080 frostfs-http-gw

Configuration

In general, everything available as CLI parameter can also be specified via environment variables (see example), so they're not specifically mentioned in most cases (see --help also). If you prefer a config file you can use it in yaml format.

Nodes: weights and priorities

You can specify multiple -p options to add more FrostFS nodes, this will make gateway spread requests equally among them (using weight 1 and priority 1 for every node):

$ frostfs-http-gw -p 192.168.130.72:8080 -p 192.168.130.71:8080

If you want some specific load distribution proportions, use weights and priorities:

$ HTTP_GW_PEERS_0_ADDRESS=192.168.130.71:8080 HTTP_GW_PEERS_0_WEIGHT=1 HTTP_GW_PEERS_0_PRIORITY=1 \
  HTTP_GW_PEERS_1_ADDRESS=192.168.130.72:8080 HTTP_GW_PEERS_1_WEIGHT=9 HTTP_GW_PEERS_1_PRIORITY=2 \
  HTTP_GW_PEERS_2_ADDRESS=192.168.130.73:8080 HTTP_GW_PEERS_2_WEIGHT=1 HTTP_GW_PEERS_2_PRIORITY=2 \
  frostfs-http-gw

This command will make gateway use 192.168.130.71 while it is healthy. Otherwise, it will make the gateway use 192.168.130.72 for 90% of requests and 192.168.130.73 for remaining 10%.

Keys

You can provide a wallet via --wallet or -w flag. You can also specify the account address using --address (if no address provided default one will be used). If wallet is used, you need to set HTTP_GW_WALLET_PASSPHRASE variable to decrypt the wallet. If no wallet provided, the gateway autogenerates a key pair it will use for FrostFS requests.

$ frostfs-http-gw -p $FROSTFS_NODE -w $WALLET_PATH --address $ACCOUNT_ADDRESS

Example:

$ frostfs-http-gw -p 192.168.130.72:8080 -w wallet.json --address NfgHwwTi3wHAS8aFAN243C5vGbkYDpqLHP

Binding and TLS

You can make the gateway listen on specific address using the --listen_address option.

It can also provide TLS interface for its users, just specify paths to the key and certificate files via --tls_key and --tls_certificate parameters. Note that using these options makes gateway TLS-only. If you need to serve both TLS and plain text HTTP, you either have to run two gateway instances or use some external redirecting solution.

Example to bind to 192.168.130.130:443 and serve TLS there:

$ frostfs-http-gw -p 192.168.130.72:8080 --listen_address 192.168.130.130:443 \
  --tls_key=key.pem --tls_certificate=cert.pem

HTTP parameters

You can tune HTTP read and write buffer sizes as well as timeouts with HTTP_GW_WEB_READ_BUFFER_SIZE, HTTP_GW_WEB_READ_TIMEOUT, HTTP_GW_WEB_WRITE_BUFFER_SIZE and HTTP_GW_WEB_WRITE_TIMEOUT environment variables.

Note: to allow upload and download of big data streams, disable read and write timeouts correspondingly. To do that, set HTTP_GW_WEB_READ_TIMEOUT=0 and HTTP_GW_WEB_WRITE_TIMEOUT=0. Otherwise, HTTP Gateway will terminate request with data stream after timeout.

HTTP_GW_WEB_STREAM_REQUEST_BODY environment variable can be used to disable request body streaming (effectively it'll make the gateway accept the file completely first and only then try sending it to FrostFS).

HTTP_GW_WEB_MAX_REQUEST_BODY_SIZE controls maximum request body size limiting uploads to files slightly lower than this limit.

FrostFS parameters

Gateway can automatically set timestamps for uploaded files based on local time source, use HTTP_GW_UPLOAD_HEADER_USE_DEFAULT_TIMESTAMP environment variable to control this behavior.

Monitoring and metrics

Pprof and Prometheus are integrated into the gateway. To enable them use --pprof and --metrics flags or HTTP_GW_PPROF/HTTP_GW_METRICS environment variables.

Timeouts

You can tune gRPC interface parameters with --connect_timeout (for connection to a node) and --request_timeout (for request processing over established connection) options.

gRPC-level checks allow the gateway to detect dead peers, but it declares them unhealthy at pool level once per --rebalance_timer interval, so check for it if needed.

All timing options accept values with suffixes, so "15s" is 15 seconds and "2m" is 2 minutes.

Zip streaming

The gateway supports downloading files by common prefix (like dir) in zip format. You can enable compression using config or HTTP_GW_ZIP_COMPRESSION=true environment variable.

Logging

You can specify logging level using variable:

HTTP_GW_LOGGER_LEVEL=debug

Yaml file

Configuration file is optional and can be used instead of environment variables/other parameters. It can be specified with --config parameter:

$ frostfs-http-gw --config your-config.yaml

See config and defaults for example.

Multiple configs

You can use several config files when running application. It allows you to split configuration into parts. For example, you can use separate yaml file for pprof and prometheus section in config (see config examples). You can either provide several files with repeating --config flag or provide path to the dir that contains all configs using --config-dir flag. Also, you can combine these flags:

$ frostfs-http-gw --config ./config/config.yaml --config /your/partial/config.yaml --config-dir ./config/dir

Note: next file in --config flag overwrites values from the previous one. Files from --config-dir directory overwrite values from --config files. So the command above run frostfs-http-gw to listen on 0.0.0.0:8080 address (value from ./config/config.yaml), applies parameters from /your/partial/config.yaml, enable pprof (value from ./config/dir/pprof.yaml) and prometheus (value from ./config/dir/prometheus.yaml).

HTTP API provided

This gateway intentionally provides limited feature set and doesn't try to substitute (or completely wrap) regular gRPC FrostFS interface. You can download and upload objects with it, but deleting, searching, managing ACLs, creating containers and other activities are not supported and not planned to be supported.

Preparation

Before uploading or downloading a file make sure you have a prepared container. You can create it with instructions below.

Also, in case of downloading, you need to have a file inside a container.

NNS

In all download/upload routes you can use container name instead of its id ($CID).

Steps to start using name resolving:

  1. Enable NNS resolving in config (rpc_endpoint must be a valid neo rpc node, see configs for other examples):
rpc_endpoint: http://morph-chain.frostfs.devenv:30333
resolve_order:
  - nns
  1. Make sure your container is registered in NNS contract. If you use frostfs-dev-env you can check if your container (e.g. with container-name name) is registered in NNS:
$ curl -s --data '{"id":1,"jsonrpc":"2.0","method":"getcontractstate","params":[1]}' \
    http://morph-chain.frostfs.devenv:30333 | jq -r '.result.hash'

0x8e6c3cd4b976b28e84a3788f6ea9e2676c15d667

$ docker exec -it morph_chain neo-go \
    contract testinvokefunction \
    -r http://morph-chain.frostfs.devenv:30333 0x8e6c3cd4b976b28e84a3788f6ea9e2676c15d667 \
    resolve string:container-name.container int:16 \
    | jq -r '.stack[0].value | if type=="array" then .[0].value else . end' \
    | base64 -d && echo

7f3vvkw4iTiS5ZZbu5BQXEmJtETWbi3uUjLNaSs29xrL
  1. Use container name instead of its $CID. For example:
$ curl http://localhost:8082/get_by_attribute/container-name/FileName/object-name

Create a container

You can create a container via frostfs-cli:

$ frostfs-cli -r $FROSTFS_NODE -w $WALLET container create --policy $POLICY --basic-acl $ACL

where $WALLET is a path to user wallet,
$ACL -- hex encoded basic ACL value or keywords 'private, 'public-read', 'public-read-write' and
$POLICY -- QL-encoded or JSON-encoded placement policy or path to file with it

For example:

$ frostfs-cli -r 192.168.130.72:8080 -w ./wallet.json container create --policy "REP 3" --basic-acl public --await

If you have launched nodes via frostfs-dev-env, you can get the key value from wallets/wallet.json or write the path to the file wallets/wallet.key.

Prepare a file in a container

To create a file via frostfs-cli, run a command below:

$ frostfs-cli -r $FROSTFS_NODE -k $KEY object put --file $FILENAME --cid $CID

where $KEY -- the key, please read the information above, $CID -- container ID.

For example:

$ frostfs-cli -r 192.168.130.72:8080 -w ./wallet.json object put --file cat.png --cid Dxhf4PNprrJHWWTG5RGLdfLkJiSQ3AQqit1MSnEPRkDZ --attributes img_type=cat,my_attr=cute

Downloading

Requests

The following requests support GET/HEAD methods.

By IDs

Basic downloading involves container ID and object ID and is done via GET requests to /get/$CID/$OID path, where $CID is a container ID or its name if NNS is enabled, $OID is an object's (i.e. your file's) ID.

For example:

$ wget http://localhost:8082/get/Dxhf4PNprrJHWWTG5RGLdfLkJiSQ3AQqit1MSnEPRkDZ/2m8PtaoricLouCn5zE8hAFr3gZEBDCZFe9BEgVJTSocY

or if container has a name:

$ wget http://localhost:8082/get/container-name/2m8PtaoricLouCn5zE8hAFr3gZEBDCZFe9BEgVJTSocY
By attributes

There is also more complex interface provided for attribute-based downloads, it's usually used to retrieve files by their names, but any other attribute can be used as well. The generic syntax for it looks like this:

/get_by_attribute/$CID/$ATTRIBUTE_NAME/$ATTRIBUTE_VALUE

where $CID is a container ID or its name if NNS is enabled, $ATTRIBUTE_NAME is the name of the attribute we want to use, $ATTRIBUTE_VALUE is the value of this attribute that the target object should have.

NB! The attribute key and value should be url encoded, i.e., if you want to download an object with the attribute value a cat, the value in the request must be a+cat. In the same way with the attribute key. If you don't escape such values everything can still work (for example you can use d@ta without encoding) but it's HIGHLY RECOMMENDED to encode all your attributes.

If multiple objects have specified attribute with specified value, then the first one of them is returned (and you can't get others via this interface).

Example for file name attribute:

$ wget http://localhost:8082/get_by_attribute/88GdaZFTcYJn1dqiSECss8kKPmmun6d6BfvC4zhwfLYM/FileName/cat.jpeg

Or when the filename includes special symbols:

$ wget http://localhost:8082/get_by_attribute/88GdaZFTcYJn1dqiSECss8kKPmmun6d6BfvC4zhwfLYM/FileName/cat+jpeg # means 'cat jpeg'
$ wget http://localhost:8082/get_by_attribute/88GdaZFTcYJn1dqiSECss8kKPmmun6d6BfvC4zhwfLYM/FileName/cat%25jpeg # means 'cat%jpeg'

Some other user-defined attributes:

$ wget http://localhost:8082/get_by_attribute/Dxhf4PNprrJHWWTG5RGLdfLkJiSQ3AQqit1MSnEPRkDZ/Ololo/100500

Or when the attribute includes special symbols:

$ wget http://localhost:8082/get_by_attribute/Dxhf4PNprrJHWWTG5RGLdfLkJiSQ3AQqit1MSnEPRkDZ/Olo%2Blo/100500 # means Olo+lo

An optional download=true argument for Content-Disposition management is also supported (more on that below):

$ wget http://localhost:8082/get/Dxhf4PNprrJHWWTG5RGLdfLkJiSQ3AQqit1MSnEPRkDZ/2m8PtaoricLouCn5zE8hAFr3gZEBDCZFe9BEgVJTSocY?download=true

Zip

You can download some dir (files with the same prefix) in zip (it will be compressed if config contains appropriate param):

$ wget http://localhost:8082/zip/Dxhf4PNprrJHWWTG5RGLdfLkJiSQ3AQqit1MSnEPRkDZ/common/prefix

Note: the objects must have a valid FilePath attribute (it should not contain trailing /), otherwise they will not be in the zip archive. You can upload file with this attribute using curl:

$ curl -F 'file=@cat.jpeg;filename=cat.jpeg' -H 'X-Attribute-FilePath: common/prefix/cat.jpeg' http://localhost:8082/upload/Dxhf4PNprrJHWWTG5RGLdfLkJiSQ3AQqit1MSnEPRkDZ

Replies

You get object contents in the reply body (if GET method was used), but at the same time you also get a set of reply headers generated using the following rules:

  • Content-Length is set to the length of the object
  • Content-Type is autodetected dynamically by gateway
  • Content-Disposition is inline for regular requests and attachment for requests with download=true argument, filename is also added if there is FileName attribute set for this object
  • Last-Modified header is set to Timestamp attribute value if it's present for the object
  • x-container-id contains container ID
  • x-object-id contains object ID
  • x-owner-id contains owner address
  • all the other FrostFS attributes are converted to X-Attribute-* headers (but only if they can be safely represented in HTTP header), for example FileName attribute becomes X-Attribute-FileName header
Caching strategy

HTTP Gateway doesn't control caching (doesn't anything with the Cache-Control header). Caching strategy strictly depends on application use case. So it should be carefully done by proxy server.

Uploading

You can POST files to /upload/$CID path where $CID is a container ID or its name if NNS is enabled. The request must contain multipart form with mandatory filename parameter. Only one part in multipart form will be processed, so to upload another file just issue a new POST request.

Example request:

$ curl -F 'file=@cat.jpeg;filename=cat.jpeg' http://localhost:8082/upload/Dxhf4PNprrJHWWTG5RGLdfLkJiSQ3AQqit1MSnEPRkDZ

Chunked encoding is supported by the server (but check for request read timeouts if you're planning some streaming). You can try streaming support with a large file piped through named FIFO pipe:

$ mkfifo pipe
$ cat video.mp4 > pipe &
$ curl --no-buffer -F 'file=@pipe;filename=catvideo.mp4' http://localhost:8082/upload/Dxhf4PNprrJHWWTG5RGLdfLkJiSQ3AQqit1MSnEPRkDZ

You can also add some attributes to your file using the following rules:

  • all "X-Attribute-*" headers get converted to object attributes with "X-Attribute-" prefix stripped, that is if you add "X-Attribute-Ololo: 100500" header to your request the resulting object will get "Ololo: 100500" attribute
  • "X-Attribute-SYSTEM-*" headers are special (-SYSTEM- part can also be -system- or-System- (and even legacy -Neofs- for some next releases)), they're used to set internal FrostFS attributes starting with __SYSTEM__ prefix, for these attributes all dashes get converted to underscores and all letters are capitalized. For example, you can use "X-Attribute-SYSTEM-Expiration-Epoch" header to set __SYSTEM__EXPIRATION_EPOCH attribute
  • FileName attribute is set from multipart's filename if not set explicitly via X-Attribute-FileName header
  • Timestamp attribute can be set using gateway local time if using HTTP_GW_UPLOAD_HEADER_USE_DEFAULT_TIMESTAMP option and if request doesn't provide X-Attribute-Timestamp header of its own

NOTE

There are some reserved headers type of X-Attribute-SYSTEM-* (headers are arranged in descending order of priority):

  1. X-Attribute-System-Expiration-Epoch: 100
  2. X-Attribute-System-Expiration-Duration: 24h30m
  3. X-Attribute-System-Expiration-Timestamp: 1637574797
  4. X-Attribute-System-Expiration-RFC3339: 2021-11-22T09:55:49Z

which transforms to X-Attribute-System-Expiration-Epoch. So you can provide expiration any convenient way.


For successful uploads you get JSON data in reply body with a container and object ID, like this:

{
        "object_id": "9ANhbry2ryjJY1NZbcjryJMRXG5uGNKd73kD3V1sVFsX",
        "container_id": "Dxhf4PNprrJHWWTG5RGLdfLkJiSQ3AQqit1MSnEPRkDZ"
}

Authentication

You can always upload files to public containers (open for anyone to put objects into), but for restricted containers you need to explicitly allow PUT operations for a request signed with your HTTP Gateway keys.

If your don't want to manage gateway's secret keys and adjust eACL rules when gateway configuration changes (new gate, key rotation, etc) or you plan to use public services, there is an option to let your application backend (or you) to issue Bearer Tokens ans pass them from the client via gate down to FrostFS level to grant access.

FrostFS Bearer Token basically is a container owner-signed ACL data (refer to FrostFS documentation for more details). There are two options to pass them to gateway:

  • "Authorization" header with "Bearer" type and base64-encoded token in credentials field
  • "Bearer" cookie with base64-encoded token contents

For example, you have a mobile application frontend with a backend part storing data in FrostFS. When a user authorizes in the mobile app, the backend issues a FrostFS Bearer token and provides it to the frontend. Then, the mobile app may generate some data and upload it via any available FrostFS HTTP Gateway by adding the corresponding header to the upload request. Accessing the ACL protected data works the same way.

Example

In order to generate a bearer token, you need to have wallet (which will be used to sign the token) and the address of the sender who will do the request to FrostFS (in our case, it's a gateway wallet address).

Suppose we have:

  • NhVtreTTCoqsMQV5Wp55fqnriiUCpEaKm3 (token owner (gateway address))

Firstly, we need to encode the container id and the sender address to base64 (now it's base58). So use base58 and base64 utils.

  1. Encoding token owner id:
$ echo 'NhVtreTTCoqsMQV5Wp55fqnriiUCpEaKm3' | base58 --decode | base64
# output: NezFK4ujidF+X7bB88uzREQzRQeAvdj3Gg==
  1. Form a Bearer token (10000 is lifetime expiration in epoch) and save it to bearer.json:
{
    "body": {
        "allowImpersonate": true,
        "ownerID": {
            "value": "NezFK4ujidF+X7bB88uzREQzRQeAvdj3Gg=="
        },
        "lifetime": {
            "exp": "10000",
            "nbf": "0",
            "iat": "0"
        }
    },
    "signature": null
}
  1. Sign it with the wallet:
$ frostfs-cli util sign bearer-token --from bearer.json --to signed.json -w ./wallet.json
  1. Encode to base64 to use in header:
$ base64 -w 0 signed.json
# output: Ck4KKgoECAIQBhIiCiCZGdlbN7DPGPMg9rsWqV+p2XdMzUqknRiexewSFp8kmBIbChk17MUri6OJ0X5ftsHzy7NERDNFB4C92PcaGgMIkE4SZgohAxpsb7vfAso1F0X6hrm6WpRS14WsT3/Ct1SMoqRsT89KEkEEGxKi8GjKSf52YqhppgaOTQHbUsL3jn7SHLqS3ndAQ7NtAATnmRHleZw2V2xRRSRBQdjDC05KK83LhdSax72Fsw==

After that, the Bearer token can be used:

$ curl -F 'file=@cat.jpeg;filename=cat.jpeg' -H "Authorization: Bearer Ck4KKgoECAIQBhIiCiCZGdlbN7DPGPMg9rsWqV+p2XdMzUqknRiexewSFp8kmBIbChk17MUri6OJ0X5ftsHzy7NERDNFB4C92PcaGgMIkE4SZgohAxpsb7vfAso1F0X6hrm6WpRS14WsT3/Ct1SMoqRsT89KEkEEGxKi8GjKSf52YqhppgaOTQHbUsL3jn7SHLqS3ndAQ7NtAATnmRHleZw2V2xRRSRBQdjDC05KK83LhdSax72Fsw==" \
  http://localhost:8082/upload/BJeErH9MWmf52VsR1mLWKkgF3pRm3FkubYxM7TZkBP4K
# output:
# {
#	"object_id": "DhfES9nVrFksxGDD2jQLunGADfrXExxNwqXbDafyBn9X",
#	"container_id": "BJeErH9MWmf52VsR1mLWKkgF3pRm3FkubYxM7TZkBP4K"
# }
Note

For the token to work correctly, you need to create a container with a basic ACL that:

  1. Allow PUT operation to others
  2. Doesn't set "final" bit

For example:

$ frostfs-cli -w ./wallet.json --basic-acl 0x0FFFCFFF -r 192.168.130.72:8080 container create --policy "REP 3"  --await

To deny access to a container without a token, set the eACL rules:

$ frostfs-cli -w ./wallet.json -r 192.168.130.72:8080 container set-eacl --table eacl.json --await --cid BJeErH9MWmf52VsR1mLWKkgF3pRm3FkubYxM7TZkBP4K

File eacl.json:

{
    "version": {
        "major": 0,
        "minor": 0
    },
    "containerID": {
        "value": "mRnZWzewzxjzIPa7Fqlfqdl3TM1KpJ0YnsXsEhafJJg="
    },
    "records": [
        {
            "operation": "PUT",
            "action": "DENY",
            "filters": [],
            "targets": [
                {
                    "role": "OTHERS",
                    "keys": []
                }
            ]
        }
    ]
}

Metrics and Pprof

If enabled, Prometheus metrics are available at localhost:8084 endpoint and Pprof at localhost:8083/debug/pprof by default. Host and port can be configured. See configuration.

Credits

Please see CREDITS for details.