HTTP API

MARV provides an HTTP API for integration with other services. There is a dedicated API endpoint that can be used to run complex queries on the MARV database. For performing actions like tagging or commenting there are no dedicated endpoints at this time, and we recommend using the frontend API endpoints documented below for the time being.

To work with the API on the command line it is handy to have curl and jq.

sudo apt-get install curl jq

Set an environment variable for the API.

export MARV_API=http://localhost:8000/marv/api

Auth

For some API calls you need to be authenticated, let’s get a token, and set handy environment vars.

echo -n "Enter username: "; read -s MARV_USER && echo && \
echo -n "Enter password: "; read -s MARV_PASS && echo && \
TOKEN=$(curl -s -X POST -H "Content-Type: application/json" \
  -d '{"username": "'$MARV_USER'", "password": "'$MARV_PASS'"}' \
  $MARV_API/auth | jq -r .access_token) && \
echo $TOKEN

Query

Database queries can be performed by posting a corresponding request to v1/rpcs. The endpoint expects a json encoded object with an rpcs key and a list of actions to run. An empty request and its response would look like:

curl -X POST \
     -H "Authorization: Bearer $TOKEN" \
     -H "Content-Type: application/json" \
     -d '{"rpcs": []}' \
     $MARV_API/v1/rpcs

{
  "data": {}
}

Query actions are represented by dictionaries with a key query. The simplest query payload {"model": "dataset"} asks for all entries of a single entity, in this case dataset. The valid entities are dataset, file, comment, tag, user, and group. Additionally, the API supports querying collections using collection:<name>, where <name> denotes configured collection name. The complete HTTP request to get all datasets looks like:

curl -X POST \
     -H "Authorization: Bearer $TOKEN" \
     -H "Content-Type: application/json" \
     -d '{"rpcs": [{"query": {"model": "dataset"}}]}' \
     $MARV_API/v1/rpcs

{
  "data": {
    "dataset": [
      {
        "id": 1,
        "collection": "bags",
        "discarded": 0,
        "name": "laser",
        "status": 0,
        "time_added": 1568903426121,
        "timestamp": 1521328251000,
        "setid": "dbgjsro7vfazyrj2pavoouk3vq"
      },
      // and many more ...
    }
  }
}

Often not all attributes are required on the client side. To restrict the fields in the response the query payload can contain an attrs object. Its keys correspond to the database fields that should be returned, and its values are always set to true for real database columns.

{
  "model": "dataset",
  "attrs": {
    "name": true,
    "setid": true,
  },
}

{
  "data": {
    "dataset": [
      {
        "name": "laser",
        "setid": "dbgjsro7vfazyrj2pavoouk3vq",
        "id": 1
      },
      {
        "name": "map",
        "setid": "ob56ua4ztunqeh7iw6ywg2ntum",
        "id": 2
      },
      {
        "name": "navsatfix",
        "setid": "g5nnjxaejquqrph6dnsencj2mi",
        "id": 3
      },
      {
        "name": "odom",
        "setid": "gl5zdundbwwsr2qhduous5324y",
        "id": 4
      },
      {
        "name": "test",
        "setid": "o6yufr25xqhx7nz7tljdr4ogaa",
        "id": 5
      }
    ]
  }
}

The results can be filtered using the filters key. Its value is a list of filter objects to apply. Each filter object consists at least of an operator op. Most operators are applied to a field name and work with a value value. The following query payload gets the dataset where the field id equals 3:

{
  "model": "dataset",
  "filters": [
    {"op": "eq", "name": "id", "value": 3},
  ],
}

{
  "data": {
    "dataset": [
      {
        "id": 3,
        "name": "navsatfix",
        // other attrs
      }
    ]
  }
}

The filters list can contain multiple filter objects and by default the results have to match all. The following operators work on fields and values:

between: field is in the range of value ([low, high])
endswith: string field ends with value
eq: field equal to value, for comparison to null use is (see below)
gt: field greater than value
gte: field greater equal value
in: field is contained in list of values [val1, val2, …]
is: field is value
isnot: field is not value
lt: field is lower than value
lte: field is lower equal value
ne: field is not equal value
notbetween: field is not in the range of value ([low, high])
notin: field is not one of the values in value [val1, val2, …]
startswith: string field starts with value
substring: value is a substring of field

There are three additional operators that implement boolean logic to combine filters. Each of the operators works without a field name and operates on value only:

not: Negates val which is another filter object itself
and: “ANDs” all filter objects in a value list
or: “ORs” all filter objects in a value list

Filters can also operate on related tables by using a dotted path notation in field names. To search for datasets tagged important use:

{
  "model": "dataset",
  "filters": [
    {"op": "eq", "name": "tags.value", "value": "important"},
  ],
}

{
  "data": {
    "dataset": [
      {
        "id": 1,
        "name": "laser",
        // other attrs
      }
    ]
  }
}

A query can also embed related models into the response by setting its corresponding attrs key similarly to regular model fields. Set the value to true to embed complete objects or set the value to an object that shall be used as attrs for the related model:

{
  "model": "dataset",
  "attrs": {
    "name": true,
    "tags": {"value": true},  # populate tag objects only with value field
  },
  "filters": [
    {"op": "eq", "name": "tags.value", "value": "important"},
  ],
}

{
  "data": {
    "dataset": [
      {
        "name": "laser",
        "id": 1,
        "tags": [
          1, 4
        ]
      }
    ],
    "tag": [
      {
        "value": "autonomous",
        "id": 1
      }
      {
        "value": "important",
        "id": 4
      }
    ]
  }
}

The API supports sorting and paging. To sort the results use the option order with a value of ["fieldname" "ORDER"] with ORDER being either ASC or DESC. Paging is achieved with the limit and offset integer options. An example query payload would look like:

{
  "model": "dataset",
  "order": ["setid", "ASC"],  # sort ascending by setid
  "limit": 5,                 # limit number of results to 5
  "offset": 10,               # skip the first 10 results
}

Collection

The filter fields of each collection are available on virtual models on the Query endpoint. Querying the virtual model of a collection is achieved by setting the model parameter to collection:<name>. The following rpc query payload gets the bags collection:

{
  "model": "collection:bags",
}

Since each collection can have individually configured filters, the returned models will have different keys for different collections. All filter field names from the marv config are prefixed with f_ in the query API. Depending on type of a filter its values are returned in one of two ways:

column on the virtual model, if the value is a single scalar per dataset,
relation to secondary model, if there can be multiple values per dataset (string[] or subset filters).

All fields of scalar type are directly embedded on the collection models in the query response.

{
  "data": {
    "collection:bags": [
      {
        "id": 1
        "f_name": "laser",
        "f_setid": "dbgjsro7vfazyrj2pavoouk3vq",
        # other filter fields
      },
      # other items
    ],
  },
}

Values of filter fields represented as a relation have to be requested specifically by name. If there is a field named topics, it can be requested using the attrs key:

{
  "model": "collection:bags",
  "attrs": {"f_topics": true},
}

The collection models in the response have a f_topics key, listing the related f_topic ids, and a top level f_topics list will contain the related values:

{
  "data": {
    "collection:bags": [
      {
        "id": 1
        "f_topics": [1, 2],
        # other filter fields
      },
      # other items
    ],
    "f_topics": [
      {
        "id": 1,
        "value": "/camA/jai/nir/camera_info",
      },
      {
        "id": 2,
        "value": "/camA/jai/nir/image_raw",
      },
      # other items
    ],
  },
}

Filtering works as with any other model. For example, to find a collection entry by setid use:

{
  "model": "collection:bags",
  "filters": [{"op": "eq", "name": "f_setid", "value": "..."}]
}

Multi-value fields can also be used in queries. For example, to find datasets including Image messages use:

{
  "model": "collection:bags",
  "filters": [{"op": "eq", "name": "f_msg_types.value", "value": "sensor_msgs/Image"}]
}

Each collection entry corresponds to one dataset. For filtering by and embedding of relations belonging to the dataset itself – namely file, comment, and tag – the API supports the dataset. prefix.

{
  "model": "collection:bags",
  "filters": [
    {"op": "substring", "name": "dataset.comments.text", "value": "failure"},
    {"op": "eq", "name": "dataset.tags.value", "value": "autonomous"},
  ]
  "attrs": {"dataset.files"},
}

List of dataset files

curl -X POST \
     -H "Content-Type: application/json" \
     -d "[42]" \
     $MARV_API/file-list

output:

{
  "paths": [
    "/scanroot/scan_odom_map_test.bag",
  ],
  "urls": [
    "dataset/h27zmwsdzcnmu6kqncwdhhvrva/0",
  ]
}

Download

First file of dataset

curl -OJ $MARV_API/dataset/h27zmwsdzcnmu6kqncwdhhvrva/0

Comment

curl -X POST \
     -H "Authorization: Bearer $TOKEN" \
     -H "Content-Type: application/json" \
     -d '{"42": {"add": ["comment 1", "comment 2"]}}' \
     $MARV_API/comment

output:

{}

Delete

curl -X DELETE \
     -H "Authorization: Bearer $TOKEN" \
     -H "Content-Type: application/json" \
     -d "[1,2]" \
     $MARV_API/dataset

output:

{}

Tag

curl -X POST \
     -H "Authorization: Bearer $TOKEN" \
     -H "Content-Type: application/json" \
     -d '{"bags": {"add": {"foo": [42]}, "remove": {"bar": [17,42]}}}' \
     $MARV_API/tag

{}

Trigger Actions (EE)

The trigger API allows to schedule actions. In case no action is running a triggered action will start immediately, otherwise right after the running action has finished. In both cases the HTTP request will remain open until the action has finished. There are two actions scan and run, which wrap the corresponding CLI commands and return a mapping with returncode, stdout, and stderr; returncode == 0 means success. In case of server-side errors only an error field is returned.

Scan

The scan action schedules a scan of all configured scanroots.

curl -X POST \
     -H "Authorization: Bearer $TOKEN" \
     -H "Content-Type: application/json" \
     -d '{"action": "scan"}' \
     $MARV_API/v1/trigger

Example output:

{"returncode": 0, "stdout": "...", "stderr": ""}

The stdout field will contain the same MARV log messages a scan on the CLI would produce.

Run

The run action schedules a run of all configured nodes on a specific dataset.

curl -X POST \
     -H "Authorization: Bearer $TOKEN" \
     -H "Content-Type: application/json" \
     -d '{"action": "run", "dataset": "xy5ba2hy2hr5453m6ftokk6kdq"}' \
     $MARV_API/v1/trigger

Example output:

{"returncode": 0, "stdout": "...", "stderr": ""}

If all nodes have already been run on this dataset the action does nothing and returns immediately indicating successful completion.