Below is a list of main features that any EVE-powered APIs can expose. Most of these features can be experienced live by consuming the Demo API (see Live demo).

Emphasis on REST

The Eve project aims to provide the best possible REST-compliant API implementation. Fundamental REST principles like separation of concerns, stateless and layered system, cacheability, uniform interface have been kept into consideration while designing the core API.

Full range of CRUD operations

APIs can support the full range of CRUD operations. Within the same API, you can have a read-only resource accessible at one endpoint, along with a fully editable resource at another endpoint. The following table shows Eve’s implementation of CRUD via REST:

Action HTTP Verb Context
Create POST Collection
Create PUT Document
Replace PUT Document
Read GET, HEAD Collection/Document
Update PATCH Document
Delete DELETE Collection/Document

Overriding HTTP Methods

As a fallback for the odd client not supporting any of these methods, the API will gladly honor X-HTTP-Method-Override requests. For example a client not supporting the PATCH method could send a POST request with a X-HTTP-Method-Override: PATCH header. The API would then perform a PATCH, overriding the original request method.

Customizable resource endpoints

By default, Eve will make known database collections available as resource endpoints (persistent identifiers in REST idiom). So a database people collection will be available at the example.com/people API endpoint. You can customize the URIs though, so the API endpoint could become, say, example.com/customers/overseas. Consider the following request:

$ curl -i http://eve-demo.herokuapp.com/people
HTTP/1.1 200 OK

The response payload will look something like this:

    "_items": [
            "firstname": "Mark",
            "lastname": "Green",
            "born": "Sat, 23 Feb 1985 12:00:00 GMT",
            "role": ["copy", "author"],
            "location": {"city": "New York", "address": "4925 Lacross Road"},
            "_id": "50bf198338345b1c604faf31",
            "_updated": "Wed, 05 Dec 2012 09:53:07 GMT",
            "_created": "Wed, 05 Dec 2012 09:53:07 GMT",
            "_etag": "ec5e8200b8fa0596afe9ca71a87f23e71ca30e2d",
            "_links": {
                "self": {"href": "people/50bf198338345b1c604faf31", "title": "person"},
    "_meta": {
        "max_results": 25,
        "total": 70,
        "page": 1
    "_links": {
        "self": {"href": "people", "title": "people"},
        "parent": {"href": "/", "title": "home"}

The _items list contains the requested data. Along with its own fields, each item provides some important, additional fields:

Field Description
_created item creation date.
_updated item last updated on.
_etag ETag, to be used for concurrency control and conditional requests.
_id unique item key, also needed to access the individual item endpoint.

These additional fields are automatically handled by the API (clients don’t need to provide them when adding/editing resources).

The _meta field provides pagination data and will only be there if Pagination has been enabled (it is by default) and there is at least one document being returned. The _links list provides HATEOAS directives.

Sub Resources

Endpoints support sub-resources so you could have something like: people/<contact_id>/invoices. When setting the url rule for such an endpoint you would use a regex and assign a field name to it:

invoices = {
    'url': 'people/<regex("[a-f0-9]{24}"):contact_id>/invoices'

Then, a GET to the following endpoint:


would cause the underlying database to be queried like this:

{'contact_id': '51f63e0838345b6dcd7eabff'}

And this one:

people/51f63e0838345b6dcd7eabff/invoices?where={"number": 10}

would be queried like:

{'contact_id': '51f63e0838345b6dcd7eabff', "number": 10}

Please note that when designing your API, most of the time you can get away without resorting to sub-resources. In the example above the same result would be achieved by simply exposing an invoices endpoint that clients could query this way:

invoices?where={"contact_id": 51f63e0838345b6dcd7eabff}


invoices?where={"contact_id": 51f63e0838345b6dcd7eabff, "number": 10}

It’s mostly a design choice, but keep in mind that when it comes to enabling individual document endpoints you might incur performance hits. This otherwise legit GET request:


would cause a two fields lookup on the database. This is not ideal and also not really needed, as <invoice_id> is a unique field. By contrast, if you had a simple resource endpoint the document lookup would happen on a single field:


Endpoints that supports sub-resources will have a specific behavior on DELETE operations. A DELETE to the following endpoint:


would cause the deletion of all the documents that match follow query:

{'contact_id': '51f63e0838345b6dcd7eabff'}

Therefore, for sub-resource endpoints, only the documents satisfying the endpoint semantic will be deleted. This differs from the standard behavior, whereas a delete operation on a collection enpoint will cause the deletion of all the documents in the collection.

Another example. A DELETE to the following item endpoint:


would cause the deletion all the documents matched by the follow query:

{'contact_id': '51f63e0838345b6dcd7eabff', "<invoice_id>": 1}

This behaviour enables support for typical tree structures, where the id of the resource alone is not necessarily a primary key by itself.

Customizable, multiple item endpoints

Resources can or cannot expose individual item endpoints. API consumers could get access to people, people/<ObjectId> and people/Doe, but only to /works. When you do grant access to item endpoints, you can define up to two lookups, both defined with regexes. The first will be the primary endpoint and will match your database primary key structure (i.e., an ObjectId in a MongoDB database).

$ curl -i http://eve-demo.herokuapp.com/people/521d6840c437dc0002d1203c
HTTP/1.1 200 OK
Etag: 28995829ee85d69c4c18d597a0f68ae606a266cc
Last-Modified: Wed, 21 Nov 2012 16:04:56 GMT

The second, which is optional and read-only, will match a field with unique values since Eve will retrieve only the first match anyway.

$ curl -i http://eve-demo.herokuapp.com/people/Doe
HTTP/1.1 200 OK
Etag: 28995829ee85d69c4c18d597a0f68ae606a266cc
Last-Modified: Wed, 21 Nov 2012 16:04:56 GMT

Since we are accessing the same item, in both cases the response payload will look something like this:

    "firstname": "John",
    "lastname": "Doe",
    "born": "Thu, 27 Aug 1970 14:37:13 GMT",
    "role": ["author"],
    "location": {"city": "Auburn", "address": "422 South Gay Street"},
    "_id": "50acfba938345b0978fccad7"
    "_updated": "Wed, 21 Nov 2012 16:04:56 GMT",
    "_created": "Wed, 21 Nov 2012 16:04:56 GMT",
    "_etag": "28995829ee85d69c4c18d597a0f68ae606a266cc",
    "_links": {
        "self": {"href": "people/50acfba938345b0978fccad7", "title": "person"},
        "parent": {"href": "/", "title": "home"},
        "collection": {"href": "people", "title": "people"}

As you can see, item endpoints provide their own HATEOAS directives.

Please Note

According to REST principles resource items should only have one unique identifier. Eve abides by providing one default endpoint per item. Adding a secondary endpoint is a decision that should be pondered carefully.

Consider our example above. Even without the people/<lastname> endpoint, a client could always retrieve a person by querying the resource endpoint by last name: people/?where={"lastname": "Doe"}. Actually the whole example is fubar, as there could be multiple people sharing the same last name, but you get the idea.


Resource endpoints allow consumers to retrieve multiple documents. Query strings are supported, allowing for filtering and sorting. Both native Mongo queries and Python conditional expressions are supported.

Here we are asking for all documents where lastname value is Doe:

http://eve-demo.herokuapp.com/people?where={"lastname": "Doe"}

With curl you would go like this:

$ curl -i -g http://eve-demo.herokuapp.com/people?where={%22lastname%22:%20%22Doe%22}
HTTP/1.1 200 OK

Filtering on embedded document fields is possible:

http://eve-demo.herokuapp.com/people?where={"location.city": "San Francisco"}

Date fields are also easy to query on:

http://eve-demo.herokuapp.com/people?where={"born": {"$gte":"Wed, 25 Feb 1987 17:00:00 GMT"}}

Date values should conform to RFC1123. Should you need a different format, you can change the DATE_FORMAT setting.

In general you will find that most MongoDB queries “just work”. Should you need it, MONGO_QUERY_BLACKLIST allows you to blacklist unwanted operators.

Native Python syntax works like this:

$ curl -i http://eve-demo.herokuapp.com/people?where=lastname=="Doe"
HTTP/1.1 200 OK

Both syntaxes allow for conditional and logical And/Or operators, however nested and combined.

Filters are enabled by default on all document fields. However, the API maintainer can choose to disable them all and/or whitelist allowed ones (see ALLOWED_FILTERS in Global Configuration). If scraping, or fear of DB DoS attacks by querying on non-indexed fields is a concern, then whitelisting allowed filters is the way to go.

You also have the option to validate the incoming filters against the resource’s schema and refuse to apply the filtering if any filters are invalid, by using the VALIDATE_FILTERING system setting (see Global Configuration)

Pretty Printing

You can pretty print the response by specifying a query parameter named pretty:

$ curl -i http://eve-demo.herokuapp.com/people?pretty
HTTP/1.1 200 OK

    "_items": [
            "_updated": "Tue, 19 Apr 2016 08:19:00 GMT",
            "firstname": "John",
            "lastname": "Doe",
            "born": "Thu, 27 Aug 1970 14:37:13 GMT",
            "role": [
            "location": {
                "city": "Auburn",
                "address": "422 South Gay Street"
            "_links": {
                "self": {
                    "href": "people/5715e9f438345b3510d27eb8",
                    "title": "person"
            "_created": "Tue, 19 Apr 2016 08:19:00 GMT",
            "_id": "5715e9f438345b3510d27eb8",
            "_etag": "86dc6b45fe7e2f41f1ca53a0e8fda81224229799"


Sorting is supported as well:

$ curl -i http://eve-demo.herokuapp.com/people?sort=city,-lastname
HTTP/1.1 200 OK

Would return documents sorted by city and then by lastname (descending). As you can see you simply prepend a minus to the field name if you need the sort order to be reversed for a field.

The MongoDB data layer also supports native MongoDB syntax:

http://eve-demo.herokuapp.com/people?sort=[("lastname", -1)]

which translates to the following curl request:

$ curl -i http://eve-demo.herokuapp.com/people?sort=[(%22lastname%22,%20-1)]
HTTP/1.1 200 OK

Would return documents sorted by lastname in descending order.

Sorting is enabled by default and can be disabled both globally and/or at resource level (see SORTING in Global Configuration and sorting in Domain Configuration). It is also possible to set the default sort at every API endpoints (see default_sort in Domain Configuration).

Please note

Always use double quotes to wrap field names and values. Using single quotes will result in 400 Bad Request responses.


Hypermedia as the Engine of Application State (HATEOAS) is enabled by default. Each GET response includes a _links section. Links provide details on their relation relative to the resource being accessed, and a title. Relations and titles can then be used by clients to dynamically updated their UI, or to navigate the API without knowing its structure beforehand. An example:

    "_links": {
        "self": {
            "href": "people",
            "title": "people"
        "parent": {
            "href": "/",
            "title": "home"
        "next": {
            "href": "people?page=2",
            "title": "next page"
        "last": {
            "href": "people?page=10",
            "title": "last page"

A GET request to the API home page (the API entry point) will be served with a list of links to accessible resources. From there, any client could navigate the API just by following the links provided with every response.

HATEOAS links are always relative to the API entry point, so if your API home is at examples.com/api/v1, the self link in the above example would mean that the people endpoint is located at examples.com/api/v1/people.

Please note that next, previous, last and related items will only be included when appropriate.

Disabling HATEOAS

HATEOAS can be disabled both at the API and/or resource level. Why would you want to turn HATEOAS off? Well, if you know that your client application is not going to use the feature, then you might want to save on both bandwidth and performance.


Eve responses are automatically rendered as JSON (the default) or XML, depending on the request Accept header. Inbound documents (for inserts and edits) are in JSON format.

$ curl -H "Accept: application/xml" -i http://eve-demo.herokuapp.com
HTTP/1.1 200 OK
Content-Type: application/xml; charset=utf-8
    <link rel="child" href="people" title="people" />
    <link rel="child" href="works" title="works" />

Default renderers might be changed by editing RENDERERS value in the settings file.


You can create your own renderer by subclassing eve.render.Renderer. Each renderer should set valid mime attr and have .render() method implemented. Please note that at least one renderer must always be enabled.

Conditional Requests

Each resource representation provides information on the last time it was updated (Last-Modified), along with an hash value computed on the representation itself (ETag). These headers allow clients to perform conditional requests by using the If-Modified-Since header:

$ curl -H "If-Modified-Since: Wed, 05 Dec 2012 09:53:07 GMT" -i http://eve-demo.herokuapp.com/people/521d6840c437dc0002d1203c
HTTP/1.1 200 OK

or the If-None-Match header:

$ curl -H "If-None-Match: 1234567890123456789012345678901234567890" -i http://eve-demo.herokuapp.com/people/521d6840c437dc0002d1203c
HTTP/1.1 200 OK

Data Integrity and Concurrency Control

API responses include a ETag header which also allows for proper concurrency control. An ETag is a hash value representing the current state of the resource on the server. Consumers are not allowed to edit (PATCH or PUT) or delete (DELETE) a resource unless they provide an up-to-date ETag for the resource they are attempting to edit. This prevents overwriting items with obsolete versions.

Consider the following workflow:

$ curl -H "Content-Type: application/json" -X PATCH -i http://eve-demo.herokuapp.com/people/521d6840c437dc0002d1203c -d '{"firstname": "ronald"}'

We attempted an edit (PATCH), but we did not provide an ETag for the item so we got a 428 PRECONDITION REQUIRED back. Let’s try again:

$ curl -H "If-Match: 1234567890123456789012345678901234567890" -H "Content-Type: application/json" -X PATCH -i http://eve-demo.herokuapp.com/people/521d6840c437dc0002d1203c -d '{"firstname": "ronald"}'

What went wrong this time? We provided the mandatory If-Match header, but it’s value did not match the ETag computed on the representation of the item currently stored on the server, so we got a 412 PRECONDITION FAILED. Again!

$ curl -H "If-Match: 80b81f314712932a4d4ea75ab0b76a4eea613012" -H "Content-Type: application/json" -X PATCH -i http://eve-demo.herokuapp.com/people/50adfa4038345b1049c88a37 -d '{"firstname": "ronald"}'
HTTP/1.1 200 OK

Finally! And the response payload looks something like this:

    "_status": "OK",
    "_updated": "Fri, 23 Nov 2012 08:11:19 GMT",
    "_id": "50adfa4038345b1049c88a37",
    "_etag": "372fbbebf54dfe61742556f17a8461ca9a6f5a11"
    "_links": {"self": "..."}

This time we got our patch in, and the server returned the new ETag. We also get the new _updated value, which eventually will allow us to perform subsequent conditional requests.

Concurrency control applies to all edition methods: PATCH (edit), PUT (replace), DELETE (delete).

Disabling concurrency control

If your use case requires, you can opt to completely disable concurrency control. ETag match checks can be disabled by setting the IF_MATCH configuration variable to False (see Global Configuration). When concurrency control is disabled no ETag is provided with responses. You should be careful about disabling this feature, as you would effectively open your API to the risk of older versions replacing your documents. Alternatively, ETag match checks can be made optional by the client if ENFORCE_IF_MATCH is disabled. When concurrency check enforcement is disabled, requests with the If-Match header will be processed as conditional requests, and requests made without the If-Match header will not be processed as conditional.

Bulk Inserts

A client may submit a single document for insertion:

$ curl -d '{"firstname": "barack", "lastname": "obama"}' -H 'Content-Type: application/json' http://eve-demo.herokuapp.com/people
HTTP/1.1 201 OK

In this case the response payload will just contain the relevant document metadata:

    "_status": "OK",
    "_updated": "Thu, 22 Nov 2012 15:22:27 GMT",
    "_id": "50ae43339fa12500024def5b",
    "_etag": "749093d334ebd05cf7f2b7dbfb7868605578db2c"
    "_links": {"self": {"href": "people/50ae43339fa12500024def5b", "title": "person"}}

When a 201 Created is returned following a POST request, the Location header is also included with the response. Its value is the URI to the new document.

In order to reduce the number of loopbacks, a client might also submit multiple documents with a single request. All it needs to do is enclose the documents in a JSON list:

$ curl -d '[{"firstname": "barack", "lastname": "obama"}, {"firstname": "mitt", "lastname": "romney"}]' -H 'Content-Type: application/json' http://eve-demo.herokuapp.com/people
HTTP/1.1 201 OK

The response will be a list itself, with the state of each document:

    "_status": "OK",
    "_items": [
            "_status": "OK",
            "_updated": "Thu, 22 Nov 2012 15:22:27 GMT",
            "_id": "50ae43339fa12500024def5b",
            "_etag": "749093d334ebd05cf7f2b7dbfb7868605578db2c"
            "_links": {"self": {"href": "people/50ae43339fa12500024def5b", "title": "person"}}
            "_status": "OK",
            "_updated": "Thu, 22 Nov 2012 15:22:27 GMT",
            "_id": "50ae43339fa12500024def5c",
            "_etag": "62d356f623c7d9dc864ffa5facc47dced4ba6907"
            "_links": {"self": {"href": "people/50ae43339fa12500024def5c", "title": "person"}}

When multiple documents are submitted the API takes advantage of MongoDB bulk insert capabilities which means that not only there’s just one request traveling from the client to the remote API, but also that a single loopback is performed between the API server and the database.

In case of successful multiple inserts, keep in mind that the Location header only returns the URI of the first created document.

Data Validation

Data validation is provided out-of-the-box. Your configuration includes a schema definition for every resource managed by the API. Data sent to the API to be inserted/updated will be validated against the schema, and a resource will only be updated if validation passes.

$ curl -d '[{"firstname": "bill", "lastname": "clinton"}, {"firstname": "mitt", "lastname": "romney"}]' -H 'Content-Type: application/json' http://eve-demo.herokuapp.com/people
HTTP/1.1 201 OK

The response will contain a success/error state for each item provided in the request:

    "_status": "ERR",
    "_error": "Some documents contains errors",
    "_items": [
            "_status": "ERR",
            "_issues": {"lastname": "value 'clinton' not unique"}
            "_status": "OK",

In the example above, the first document did not validate so the whole request has been rejected.

When all documents pass validation and are inserted correctly the response status is 201 Created. If any document fails validation the response status is 422 Unprocessable Entity, or any other error code defined by VALIDATION_ERROR_STATUS configuration.

For more information see Data Validation.

Extensible Data Validation

Data validation is based on the Cerberus validation system and therefore it is extensible, so you can adapt it to your specific use case. Say that your API can only accept odd numbers for a certain field value; you can extend the validation class to validate that. Or say you want to make sure that a VAT field actually matches your own country VAT algorithm; you can do that too. As a matter of fact, Eve’s MongoDB data-layer itself extends Cerberus validation by implementing the unique schema field constraint. For more information see Data Validation.

Resource-level Cache Control

You can set global and individual cache-control directives for each resource.

$ curl -i http://eve-demo.herokuapp.com
HTTP/1.1 200 OK
Content-Type: application/json
Content-Length: 131
Cache-Control: max-age=20
Expires: Tue, 22 Jan 2013 09:34:34 GMT
Server: Eve/0.0.3 Werkzeug/0.8.3 Python/2.7.3
Date: Tue, 22 Jan 2013 09:34:14 GMT

The response above includes both Cache-Control and Expires headers. These will minimize load on the server since cache-enabled consumers will perform resource-intensive request only when really needed.

API Versioning

I’m not too fond of API versioning. I believe that clients should be intelligent enough to deal with API updates transparently, especially since Eve-powered API support HATEOAS. When versioning is a necessity, different API versions should be isolated instances since so many things could be different between versions: caching, URIs, schemas, validation, and so on. URI versioning (http://api.example.com/v1/…) is supported.

Document Versioning

Eve supports automatic version control of documents. By default, this setting is turned off, but it can be turned globally or configured individually for each resource. When enabled, Eve begins automatically tracking changes to documents and adds the fields _version and _latest_version when retrieving documents.

Behind the scenes, Eve stores document versions in shadow collections that parallels the collection of each primary resource that Eve defines. New document versions are automatically added to this collection during normal POST, PUT, and PATCH operations. A special new query parameter is available when GETing an item that provides access to the document versions. Access a specific version with ?version=VERSION, access all versions with ?version=all, and access diffs of all versions with ?version=diffs. Collection query features like projections, pagination, and sorting work with all and diff except for sorting which does not work on diff.

It is important to note that there are a few non-standard scenarios which could produce unexpected results when versioning is turned on. In particular, document history will not be saved when modifying collections outside of the Eve generated API. Also, if at anytime the VERSION field gets removed from the primary document (which cannot happen through the API when versioning is turned on), a subsequent write will re-initialize the VERSION number with VERSION = 1. At this time there will be multiple versions of the document with the same version number. In normal practice, VERSIONING can be enable without worry for any new collection or even an existing collection which has not previously had versioning enabled.

Additionally, there are caching corner cases unique to document versions. A specific document version includes the _latest_version field, the value of which will change when a new document version is created. To account for this, Eve determines the time _latest_version changed (the timestamp of the last update to the primary document) and uses that value to populate the Last-Modified header and check the If-Modified-Since conditional cache validator of specific document version queries. Note that this will be different from the timestamp in the version’s last updated field. The etag for a document version does not change when _latest_version changes, however. This results in two corner cases. First, because Eve cannot determine if the client’s _latest_version is up to date from an ETag alone, a query using only If-None-Match for cache validation of old document versions will always have its cache invalidated. Second, a version fetched and cached in the same second that multiple new versions are created can receive incorrect Not Modified responses on ensuing GET queries due to Last-Modified values having a resolution of one second and the static etag values not providing indication of the changes. These are both highly unlikely scenarios, but an application expecting multiple edits per second should account for the possibility of holing stale _latest_version data.

For more information see and Global Configuration and Domain Configuration.


Customizable Basic Authentication (RFC-2617), Token-based authentication and HMAC-based Authentication are supported. OAuth2 can be easily integrated. You can lockdown the whole API, or just some endpoints. You can also restrict CRUD commands, like allowing open read-only access while restricting edits, inserts and deletes to authorized users. Role-based access control is supported as well. For more information see Authentication and Authorization.

CORS Cross-Origin Resource Sharing

Eve-powered APIs can be accessed by the JavaScript contained in web pages. Disabled by default, CORS allows web pages to work with REST APIs, something that is usually restricted by most browsers ‘same domain’ security policy. The X_DOMAINS setting allows to specify which domains are allowed to perform CORS requests. A list of regular expressions may be defined in X_DOMAINS_RE, which is useful for websites with dynamic ranges of subdomains. Make sure to anchor and escape the regexes properly, for example X_DOMAINS_RE = ['^http://sub-\d{3}\.example\.com$'].

JSONP Support

In general you don’t really want to add JSONP when you can enable CORS instead:

There have been some criticisms raised about JSONP. Cross-origin resource sharing (CORS) is a more recent method of getting data from a server in a different domain, which addresses some of those criticisms. All modern browsers now support CORS making it a viable cross-browser alternative (source.)

There are circumstances however when you do need JSONP, like when you have to support legacy software (IE6 anyone?)

To enable JSONP in Eve you just set JSONP_ARGUMENT. Then, any valid request with JSONP_ARGUMENT will get back a response wrapped with said argument value. For example if you set JSON_ARGUMENT = 'callback':

$ curl -i http://localhost:5000/?callback=hello
hello(<JSON here>)

Requests with no callback argument will be served with no JSONP.

Read-only by default

If all you need is a read-only API, then you can have it up and running in a matter of minutes.

Default and Nullable Values

Fields can have default values and nullable types. When serving POST (create) requests, missing fields will be assigned the configured default values. See default and nullable keywords in Schema Definition for more information.

Predefined Database Filters

Resource endpoints will only expose (and update) documents that match a predefined filter. This allows for multiple resource endpoints to seamlessly target the same database collection. A typical use-case would be a hypothetical people collection on the database being used by both the /admins and /users API endpoints.


This feature allows you to create dynamic views of collections and documents, or more precisely, to decide what fields should or should not be returned, using a ‘projection’. Put another way, Projections are conditional queries where the client dictates which fields should be returned by the API.

$ curl -i http://eve-demo.herokuapp.com/people?projection={"lastname": 1, "born": 1}
HTTP/1.1 200 OK

The query above will only return lastname and born out of all the fields available in the ‘people’ resource. You can also exclude fields:

$ curl -i http://eve-demo.herokuapp.com/people?projection={"born": 0}
HTTP/1.1 200 OK

The above will return all fields but born. Please note that key fields such as ID_FIELD, DATE_CREATED, DATE_UPDATED etc. will still be included with the payload. Also keep in mind that some database engines, Mongo included, do not allow for mixing of inclusive and exclusive selections.

Embedded Resource Serialization

If a document field is referencing a document in another resource, clients can request the referenced document to be embedded within the requested document.

Clients have the power to activate document embedding on per-request basis by means of a query parameter. Suppose you have a emails resource configured like this:

     'emails': {
         'schema': {
             'author': {
                 'type': 'objectid',
                 'data_relation': {
                     'resource': 'users',
                     'field': '_id',
                     'embeddable': True
             'subject': {'type': 'string'},
             'body': {'type': 'string'},

A GET like this: /emails?embedded={"author":1} would return a fully embedded users document, whereas the same request without the embedded argument would just return the user ObjectId. Embedded resource serialization is available at both resource and item (/emails/<id>/?embedded={"author":1}) endpoints.

Embedding can be enabled or disabled both at global level (by setting EMBEDDING to either True or False) and at resource level (by toggling the embedding value). Furthermore, only fields with the embeddable value explicitly set to True will allow the embedding of referenced documents.

Embedding also works with a data_relation to a specific version of a document, but the schema looks a little bit different. To enable the data_relation to a specific version, add 'version': True to the data_relation block. You’ll also want to change the type to dict and add the schema definition shown below.

     'emails': {
         'schema': {
             'author': {
                 'type': 'dict',
                 'schema': {
                     '_id': {'type': 'objectid'},
                     '_version': {'type': 'integer'}
                 'data_relation': {
                     'resource': 'users',
                     'field': '_id',
                     'embeddable': True,
                     'version': True,
             'subject': {'type': 'string'},
             'body': {'type': 'string'},

As you can see, 'version': True changes the expected value of a data_relation field to a dictionary with fields names data_relation['field'] and VERSION. With 'field': '_id' in the data_relation definition above and VERSION = '_version' in the Eve config, the value of the data_relation in this scenario would be a dictionary with fields _id and _version.

Predefined Resource Serialization

It is also possible to elect some fields for predefined resource serialization. If the listed fields are embeddable and they are actually referencing documents in other resources (and embedding is enabled for the resource), then the referenced documents will be embedded by default. Clients can still opt out from field that are embedded by default:

$ curl -i http://example.com/people/?embedded={"author": 0}
HTTP/1.1 200 OK


Currently we support embedding of documents by references located in any subdocuments (nested dicts and lists). For example, a query /invoices/?embedded={"user.friends":1} will return a document with user and all his friends embedded, but only if user is a subdocument and friends is a list of reference (it could be a list of dicts, nested dict, etc.). This feature is about serialization on GET requests. There’s no support for POST, PUT or PATCH of embedded documents.

Document embedding is enabled by default.

Please note

When it comes to MongoDB, what embedded resource serialization deals with is document references (linked documents), something different from embedded documents, also supported by Eve (see MongoDB Data Model Design). Embedded resource serialization is a nice feature that can really help with normalizing your data model for the client. However, when deciding whether to enable it or not, especially by default, keep in mind that each embedded resource being looked up will require a database lookup, which can easily lead to performance issues.

Soft Delete

Eve provides an optional “soft delete” mode in which deleted documents continue to be stored in the database and are able to be restored, but still act as removed items in response to API requests. Soft delete is disabled by default, but can be enabled globally using the SOFT_DELETE configuration setting, or individually configured at the resource level using the domain configuration soft_delete setting. See Global Configuration and Domain Configuration for more information on enabling and configuring soft delete.

When soft deletion is enabled, callbacks attached to on_delete_resource_originals and on_delete_resource_originals_<resource_name> events will receive both deleted and not deleted documents via the originals argument (see Event Hooks).


With soft delete enabled, DELETE requests to individual items and resources respond just as they do for a traditional “hard” delete. Behind the scenes, however, Eve does not remove deleted items from the database, but instead patches the document with a _deleted meta field set to true. (The name of the _deleted field is configurable. See Global Configuration.) All requests made when soft delete is enabled filter against or otherwise account for the _deleted field.

The _deleted field is automatically added and initialized to false for all documents created while soft delete is enabled. Documents created prior to soft delete being enabled and which therefore do not define the _deleted field in the database will still include _deleted: false in API response data, added by Eve during response construction. PUTs or PATCHes to these documents will add the _deleted field to the stored documents, set to false.

Responses to GET requests for soft deleted documents vary slightly from responses to missing or “hard” deleted documents. GET requests for soft deleted documents will still respond with 404 Not Found status codes, but the response body will contain the soft deleted document with _deleted: true. Documents embedded in the deleted document will not be expanded in the response, regardless of any default settings or the contents of the request’s embedded query param. This is to ensure that soft deleted documents included in 404 responses reflect the state of a document when it was deleted, and do not to change if embedded documents are updated.

By default, resource level GET requests will not include soft deleted items in their response. This behavior matches that of requests after a “hard” delete. If including deleted items in the response is desired, the show_deleted query param can be added to the request. (the show_deleted param name is configurable. See Global Configuration) Eve will respond with all documents, deleted or not, and it is up to the client to parse returned documents’ _deleted field. The _deleted field can also be explicitly filtered against in a request, allowing only deleted documents to be returned using a ?where={"_deleted": true} query.

Soft delete is enforced in the data layer, meaning queries made by application code using the app.data.find_one and app.data.find methods will automatically filter out soft deleted items. Passing a request object with req.show_deleted == True or a lookup dictionary that explicitly filters on the _deleted field will override the default filtering.

Restoring Soft Deleted Items

PUT or PATCH requests made to a soft deleted document will restore it, automatically setting _deleted to false in the database. Modifying the _deleted field directly is not necessary (or allowed). For example, using PATCH requests, only the fields to be changed in the restored version would be specified, or an empty request would be made to restore the document as is. The request must be made with proper authorization for write permission to the soft deleted document or it will be refused.

Be aware that, should a previously soft deleted document be restored, there is a chance that an eventual unique field might end up being now duplicated in two different documents: the restored one, and another which might have been stored with the same field value while the original (now restored) was in ‘deleted’ state. This is because soft deleted documents are ignored when validating the unique rule for new or updated documents.


Soft deleting a versioned document creates a new version of that document with _deleted set to true. A GET request to the deleted version will receive a 404 Not Found response as described above, while previous versions will continue to respond with 200 OK. Responses to ?version=diff or ?version=all will include the deleted version as if it were any other.

Data Relations

The Eve data_relation validator will not allow references to documents that have been soft deleted. Attempting to create or update a document with a reference to a soft deleted document will fail just as if that document had been hard deleted. Existing data relations to documents that are soft deleted remain in the database, but requests requiring embedded document serialization of those relations will resolve to a null value. Again, this matches the behavior of relations to hard deleted documents.

Versioned data relations to a deleted document version will also fail to validate, but relations to versions prior to deletion or after restoration of the document are allowed and will continue to resolve successfully.


Disabling soft delete after use in an application requires database maintenance to ensure your API remains consistent. With soft delete disabled, requests will no longer filter against or handle the _deleted field, and documents that were soft deleted will now be live again on your API. It is therefore necessary when disabling soft delete to perform a data migration to remove all documents with _deleted == True, and recommended to remove the _deleted field from documents where _deleted == False. Enabling soft delete in an existing application is safe, and will maintain documents deleted from that point on.

Event Hooks

Pre-Request Event Hooks

When a GET/HEAD, POST, PATCH, PUT, DELETE request is received, both a on_pre_<method> and a on_pre_<method>_<resource> event is raised. You can subscribe to these events with multiple callback functions.

>>> def pre_get_callback(resource, request, lookup):
...  print('A GET request on the "%s" endpoint has just been received!' % resource)

>>> def pre_contacts_get_callback(request, lookup):
...  print('A GET request on the contacts endpoint has just been received!')

>>> app = Eve()

>>> app.on_pre_GET += pre_get_callback
>>> app.on_pre_GET_contacts += pre_contacts_get_callback

>>> app.run()

Callbacks will receive the resource being requested, the original flask.request object and the current lookup dictionary as arguments (only exception being the on_pre_POST hook which does not provide a lookup argument).

Dynamic Lookup Filters

Since the lookup dictionary will be used by the data layer to retrieve resource documents, developers may choose to alter it in order to add custom logic to the lookup query.

def pre_GET(resource, request, lookup):
    # only return documents that have a 'username' field.
    lookup["username"] = {'$exists': True}

app = Eve()

app.on_pre_GET += pre_GET

Altering the lookup dictionary at runtime would have similar effects to applying Predefined Database Filters via configuration. However, you can only set static filters via configuration whereas by hooking to the on_pre_<METHOD> events you are allowed to set dynamic filters instead, which allows for additional flexibility.

Post-Request Event Hooks

When a GET, POST, PATCH, PUT, DELETE method has been executed, both a on_post_<method> and on_post_<method>_<resource> event is raised. You can subscribe to these events with multiple callback functions. Callbacks will receive the resource accessed, original flask.request object and the response payload.

>>> def post_get_callback(resource, request, payload):
...  print('A GET on the "%s" endpoint was just performed!' % resource)

>>> def post_contacts_get_callback(request, payload):
...  print('A get on "contacts" was just performed!')

>>> app = Eve()

>>> app.on_post_GET += post_get_callback
>>> app.on_post_GET_contacts += post_contacts_get_callback

>>> app.run()

Database event hooks

Database event hooks work like request event hooks. These events are fired before and after a database action. Here is an example of how events are configured:

>>> def add_signature(resource, response):
...     response['SIGNATURE'] = "A %s from eve" % resource

>>> app = Eve()
>>> app.on_fetched_item += add_signature

You may use flask’s abort() to interrupt the database operation:

>>> from flask import abort

>>> def check_update_access(resource, updates, original):
...     abort(403)

>>> app = Eve()
>>> app.on_insert_item += check_update_access

The events are fired for resources and items if the action is available for both. And for each action two events will be fired:

  • Generic: on_<action_name>
  • With the name of the resource: on_<action_name>_<resource_name>

Let’s see an overview of what events are available:

Action What When Event name / method signature
Fetch Resource After
def event(resource_name, response)
def event(response)
Item After
def event(resource_name, response)
def event(response)
Diffs After
def event(resource_name, response)
def event(response)
Insert Items Before
def event(resource_name, items)
def event(items)
def event(resource_name, items)
def event(items)
Replace Item Before
def event(resource_name, item, original)
def event(item, original)
def event(resource_name, item, original)
def event(item, original)
Update Item Before
def event(resource_name, updates, original)
def event(updates, original)
def event(resource_name, updates, original)
def event(updates, original)
Delete Item Before
def event(resource_name, item)
def event(item)
def event(resource_name, item)
def event(item)
Resource Before
def event(resource_name)
def event()
def event(resource_name, originals, lookup)
def event(originals, lookup)
def event(resource_name, item)
def event(item)

Fetch Events

These are the fetch events with their method signature:

  • on_fetched_resource(resource_name, response)
  • on_fetched_resource_<resource_name>(response)
  • on_fetched_item(resource_name, response)
  • on_fetched_item_<resource_name>(response)
  • on_fetched_diffs(resource_name, response)
  • on_fetched_diffs_<resource_name>(response)

They are raised when items have just been read from the database and are about to be sent to the client. Registered callback functions can manipulate the items as needed before they are returned to the client.

>>> def before_returning_items(resource_name, response):
...  print('About to return items from "%s" ' % resource_name)

>>> def before_returning_contacts(response):
...  print('About to return contacts')

>>> def before_returning_item(resource_name, response):
...  print('About to return an item from "%s" ' % resource_name)

>>> def before_returning_contact(response):
...  print('About to return a contact')

>>> app = Eve()
>>> app.on_fetched_resource += before_returning_items
>>> app.on_fetched_resource_contacts += before_returning_contacts
>>> app.on_fetched_item += before_returning_item
>>> app.on_fetched_item_contacts += before_returning_contact

It is important to note that item fetch events will work with Document Versioning for specific document versions like ?version=5 and all document versions with ?version=all. Accessing diffs of all versions with ?version=diffs will only work with the diffs fetch events. Note that diffs returns partial documents which should be handled in the callback.

Insert Events

These are the insert events with their method signature:

  • on_insert(resource_name, items)
  • on_insert_<resource_name>(items)
  • on_inserted(resource_name, items)
  • on_inserted_<resource_name>(items)

When a POST requests hits the API and new items are about to be stored in the database, these events are fired:

  • on_insert for every resource endpoint.
  • on_insert_<resource_name> for the specific <resource_name> resource endpoint.

Callback functions could hook into these events to arbitrarily add new fields or edit existing ones.

After the items have been inserted, these two events are fired:

  • on_inserted for every resource endpoint.
  • on_inserted_<resource_name> for the specific <resource_name> resource endpoint.

Validation errors

Items passed to these events as arguments come in a list. And only those items that passed validation are sent.


>>> def before_insert(resource_name, items):
...  print('About to store items to "%s" ' % resource)

>>> def after_insert_contacts(items):
...  print('About to store contacts')

>>> app = Eve()
>>> app.on_insert += before_insert
>>> app.on_inserted_contacts += after_insert_contacts

Replace Events

These are the replace events with their method signature:

  • on_replace(resource_name, item, original)
  • on_replace_<resource_name>(item, original)
  • on_replaced(resource_name, item, original)
  • on_replaced_<resource_name>(item, original)

When a PUT request hits the API and an item is about to be replaced after passing validation, these events are fired:

  • on_replace for any resource item endpoint.
  • on_replace_<resource_name> for the specific resource endpoint.

item is the new item which is about to be stored. original is the item in the database that is being replaced. Callback functions could hook into these events to arbitrarily add or update item fields, or to perform other accessory action.

After the item has been replaced, these other two events are fired:

  • on_replaced for any resource item endpoint.
  • on_replaced_<resource_name> for the specific resource endpoint.

Update Events

These are the update events with their method signature:

  • on_update(resource_name, updates, original)
  • on_update_<resource_name>(updates, original)
  • on_updated(resource_name, updates, original)
  • on_updated_<resource_name>(updates, original)

When a PATCH request hits the API and an item is about to be updated after passing validation, these events are fired before the item is updated:

  • on_update for any resource endpoint.
  • on_update_<resource_name> is fired only when the <resource_name> endpoint is hit.

Here updates stands for updates being applied to the item and original is the item in the database that is about to be updated. Callback functions could hook into these events to arbitrarily add or update fields in updates, or to perform other accessory action.

After the item has been updated:

  • on_updated is fired for any resource endpoint.
  • on_updated_<resource_name> is fired only when the <resource_name> endpoint is hit.

Please note

Please be aware that last_modified and etag headers will always be consistent with the state of the items on the database (they won’t be updated to reflect changes eventually applied by the callback functions).

Delete Events

These are the delete events with their method signature:

  • on_delete_item(resource_name, item)
  • on_delete_item_<resource_name>(item)
  • on_deleted_item(resource_name, item)
  • on_deleted_item_<resource_name>(item)
  • on_delete_resource(resource_name)
  • on_delete_resource_<resource_name>()
  • on_delete_resource_originals(originals, lookup)
  • on_delete_resource_originals_<resource_name>(originals, lookup)
  • on_deleted_resource(resource_name)
  • on_deleted_resource_<resource_name>()

When a DELETE request hits an item endpoint and before the item is deleted, these events are fired:

  • on_delete_item for any resource hit by the request.
  • on_delete_item_<resource_name> for the specific <resource_name> item endpoint hit by the DELETE.

After the item has been deleted the on_deleted_item(resource_name, item) and on_deleted_item_<resource_name>(item) are raised.

item is the item being deleted. Callback functions could hook into these events to perform accessory actions. And no you can’t arbitrarily abort the delete operation at this point (you should probably look at Data Validation, or eventually disable the delete command altogether).


If you were brave enough to enable the DELETE command on resource endpoints (allowing for wipeout of the entire collection in one go), then you can be notified of such a disastrous occurrence by hooking a callback function to the on_delete_resource(resource_name) or on_delete_resource_<resource_name>() hooks.

  • on_delete_resource_originals for any resource hit by the request after having retrieved the originals documents.
  • on_delete_resource_originals_<resource_name> for the specific <resource_name> resource endpoint hit by the DELETE after having retrieved the original document.

NOTE: those two event are useful in order to perform some business logic before the actual remove operation given the look up and the list of originals

Aggregation event hooks

You can also attach one or more callbacks to your aggregation endpoints. The before_aggregation event is fired when an aggregation is about to be performed. Any attached callback function will receive both the endpoint name and the aggregation pipeline as arguments. The pipeline can then be altered if needed.

>>> def on_aggregate(endpoint, pipeline):
...   pipeline.append({"$unwind": "$tags"})

>>> app = Eve()
>>> app.before_aggregation += on_aggregate

The after_aggregation event is fired when the aggregation has been performed. An attached callback function could leverage this event to modify the documents before they are returned to the client.

>>> def alter_documents(endpoint, documents):
...   for document in documents:
...     document['hello'] = 'well, hello!'

>>> app = Eve()
>>> app.after_aggregation += alter_documents

For more information on aggregation support, see MongoDB Aggregation Framework

Please note

To provide seamless event handling features Eve relies on the Events package.

Rate Limiting

API rate limiting is supported on a per-user/method basis. You can set the number of requests and the time window for each HTTP method. If the requests limit is hit within the time window, the API will respond with 429 Request limit exceeded until the timer resets. Users are identified by the Authentication header or (when missing) by the client IP. When rate limiting is enabled, appropriate X-RateLimit- headers are provided with every API response. Suppose that the rate limit has been set to 300 requests every 15 minutes, this is what a user would get after hitting a endpoint with a single request:

X-RateLimit-Remaining: 299
X-RateLimit-Limit: 300
X-RateLimit-Reset: 1370940300

You can set different limits for each one of the supported methods (GET, POST, PATCH, DELETE).

Please Note

Rate Limiting is disabled by default, and needs a Redis server running when enabled. A tutorial on Rate Limiting is forthcoming.

Custom ID Fields

Eve allows to extend its standard data type support. In the Handling custom ID fields tutorial we see how it is possible to use UUID values instead of MongoDB default ObjectIds as unique document identifiers.

File Storage

Media files (images, pdf, etc.) can be uploaded as media document fields. Upload is done via POST, PUT and PATCH as usual, but using the multipart/form-data content-type.

Let us assume that the accounts endpoint has a schema like this:

accounts = {
    'name': {'type': 'string'},
    'pic': {'type': 'media'},

With curl we would POST like this:

$ curl -F "name=john" -F "[email protected]" http://example.com/accounts

For optimized performance files are stored in GridFS by default. Custom MediaStorage classes can be implemented and passed to the application to support alternative storage systems. A FileSystemMediaStorage class is in the works, and will soon be included with the Eve package.

As a proper developer guide is not available yet, you can peek at the MediaStorage source if you are interested in developing custom storage classes.

Serving media files as Base64 strings

When a document is requested media files will be returned as Base64 strings,

     '_items': [
             '_updated':'Sat, 05 Apr 2014 15:52:53 GMT',

However, if the EXTENDED_MEDIA_INFO list is populated (it isn’t by default) the payload format will be different. This flag allows passthrough from the driver of additional meta fields. For example, using the MongoDB driver, fields like content_type, name and length can be added to this list and will be passed-through from the underlying driver.

When EXTENDED_MEDIA_INFO is used the field will be a dictionary whereas the file itself is stored under the file key and other keys are the meta fields. Suppose that the flag is set like this:

EXTENDED_MEDIA_INFO = ['content_type', 'name', 'length']

Then the output will be something like

    '_items': [
            '_updated':'Sat, 05 Apr 2014 15:52:53 GMT',
            'pic': {
                'file': 'iVBORw0KGgoAAAANSUhEUgAAA4AAAAOACA...',
                'content_type': 'text/plain',
                'name': 'test.txt',
                'length': 8129

For MongoDB, further fields can be found in the driver documentation.

If you have other means to retrieve the media files (custom Flask endpoint for example) then the media files can be excluded from the payload by setting to False the RETURN_MEDIA_AS_BASE64_STRING flag. This takes into account if EXTENDED_MEDIA_INFO is used.

Serving media files at a dedicated endpoint

While returning files embedded as Base64 fields is the default behaviour, you can opt for serving them at a dedicated media endpoint. You achieve that by setting RETURN_MEDIA_AS_URL to True. When this feature is enabled document fields contain urls to the correspondent files, which are served at the media endpoint.

You can change the default media endpoint (media) by updating the MEDIA_BASE_URL and MEDIA_ENDPOINT setting. Suppose you are storing your images on Amazon S3 via a custom MediaStorage subclass. You would probably set your media endpoint like so:

# disable default behaviour

# return media as URL instead

# set up the desired media endpoint
MEDIA_BASE_URL = 'https://s3-us-west-2.amazonaws.com'

Setting MEDIA_BASE_URL is optional. If no value is set, then the API base address will be used when building the URL for MEDIA_ENDPOINT.

Partial media downloads

When files are served at a dedicated endpoint, clients can request partial downloads. This allows them to provide features such as optimized pause/resume (with no need to restart the download). To perform a partial download, make sure the Range header is added the the client request.

$ curl http://localhost/media/yourfile -i -H "Range: bytes=0-10"
Date: Sun, 20 Aug 2017 14:26:42 GMT
Content-Type: audio/mp4
Content-Length: 11
Connection: keep-alive
Content-Range: bytes 0-10/23671
Last-Modified: Sat, 19 Aug 2017 03:25:36 GMT
Accept-Ranges: bytes


In the snippet above, we see curl requesting the first chunk of a file.

Leveraging Projections to optimize the handling of media files

Clients and API maintainers can exploit the Projections feature to include/exclude media fields from response payloads.

Suppose that a client stored a document with an image. The image field is called image and it is of media type. At a later time, the client wants to retrieve the same document but, in order to optimize for speed and since the image is cached already, it does not want to download the image along with the document. It can do so by requesting the field to be trimmed out of the response payload:

$ curl -i http://example.com/people/<id>?projection={"image": 0}
HTTP/1.1 200 OK

The document will be returned with all its fields except the image field.

Moreover, when setting the datasource property for any given resource endpoint it is possible to explicitly exclude fields (of media type, but also of any other type) from default responses:

people = {
    'datasource': {
        'projection': {'image': 0}

Now clients will have to explicitly request the image field to be included with response payloads by sending requests like this one:

$ curl -i http://example.com/people/<id>?projection={"image": 1}
HTTP/1.1 200 OK

See also

for details on the datasource setting.

Note on media files as multipart/form-data

If you are uploading media files as multipart/form-data all the additional fields except the file fields will be treated as strings for all field validation purposes. If you have already defined some of the resource fields to be of different type (boolean, number, list etc) the validation rules for these fields would fail, preventing you to successffully submit your resource.

If you still want to be able to perform field validation in this case, you will have to turn on MULTIPART_FORM_FIELDS_AS_JSON in your settings file in order to treat the incoming fields as JSON encoded strings and still be able to validate your fields.

Please note, that in case you indeed turn on MULTIPART_FORM_FIELDS_AS_JSON you will have to submit all resource fields as properly encoded JSON strings.

For example a number should be submited as 1234 (as you would normally expect). A boolean will have to be send as true (note the lowercase t). A list of strings as ["abc", "xyz"]. And finally a string, which is the thing that will most likely trip, you will have to be submitted as "'abc'" (note that it is surrounded with double quotes). If ever in doubt if what you are submitting is a valid JSON string you can try passing it from the JSON Validator at http://jsonlint.com/ to be sure that it is correct.

Using lists of media

When using lists of media, there is no way to submit these in the default configuration. Enable AUTO_COLLAPSE_MULTI_KEYS and AUTO_CREATE_LISTS to make this possible. This allows to send multiple values for one key in multipart/form-data requests and in this way upload a list of files.


The MongoDB data layer supports geographic data structures encoded in GeoJSON format. All GeoJSON objects supported by MongoDB are available:

  • Point
  • Multipoint
  • LineString
  • MultiLineString
  • Polygon
  • MultiPolygon
  • GeometryCollection

All these objects are implemented as native Eve data types (see Schema Definition) so they are are subject to the proper validation.

In the example below we are extending the people endpoint by adding a location field of type Point.

people = {
    'location': {
        'type': 'point'

Storing a contact along with its location is pretty straightforward:

$ curl -d '[{"firstname": "barack", "lastname": "obama", "location": {"type":"Point","coordinates":[100.0,10.0]}}]' -H 'Content-Type: application/json'
HTTP/1.1 201 OK

Eve also supports GeoJSON Feature and FeatureCollection objects, which are not explicitely mentioned in MongoDB documentation. GeoJSON specification allows object to contain any number of members (name/value pairs). Eve validation was implemented to be more strict, allowing only two members. This restriction can be disabled by setting ALLOW_CUSTOM_FIELDS_IN_GEOJSON to True.

Querying GeoJSON Data

As a general rule all MongoDB geospatial query operators and their associated geometry specifiers are supported. In this example we are using the $near operator to query for all contacts living in a location within 1000 meters from a certain point:

?where={"location": {"$near": {"$geometry": {"type":"Point", "coordinates": [10.0, 20.0]}, "$maxDistance": 1000}}}

Please refer to MongoDB documentation for details on geo queries.

Internal Resources

By default responses to GET requests to the home endpoint will include all the resources. The internal_resource setting keyword, however, allows you to make an endpoint internal, available only for internal data manipulation: no HTTP calls can be made against it and it will be excluded from the HATEOAS links.

An usage example would be a mechanism for logging all inserts happening in the system, something that can be used for auditing or a notification system. First we define an internal_transaction endpoint, which is flagged as an internal_resource:

 internal_transactions = {
     'schema': {
         'entities': {
             'type': 'list',
         'original_resource': {
             'type': 'string',
     'internal_resource': True

Now, if we access the home endpoint and HATEOAS is enabled, we won’t get the internal-transactions listed (and hitting the endpoint via HTTP will return a 404.) We can use the data layer to access our secret endpoint. Something like this:

 from eve import Eve

 def on_generic_inserted(self, resource, documents):
     if resource != 'internal_transactions':
         dt = datetime.now()
         transaction = {
             'entities':  [document['_id'] for document in documents],
             'original_resource': resource,
             config.LAST_UPDATED: dt,
             config.DATE_CREATED: dt,
         app.data.insert('internal_transactions', [transaction])

 app = Eve()
 app.on_inserted += self.on_generic_inserted


I admit that this example is as rudimentary as it can get, but hopefully it will get the point across.

Enhanced Logging

A number of events are available for logging via the default application logger. The standard LogRecord attributes are extended with a few request attributes:

clientip IP address of the client performing the request.
url Full request URL, eventual query parameters included.
method Request method (POST, GET, etc.)

You can use these fields when logging to a file or any other destination.

Callback functions can also take advantage of the builtin logger. The following example logs application events to a file, and also logs custom messages every time a custom function is invoked.

import logging

from eve import Eve

def log_every_get(resource, request, payload):
    # custom INFO-level message is sent to the log file
    app.logger.info('We just answered to a GET request!')

app = Eve()
app.on_post_GET += log_every_get

if __name__ == '__main__':

    # enable logging to 'app.log' file
    handler = logging.FileHandler('app.log')

    # set a custom log format, and add request
    # metadata to each log line
        '%(asctime)s %(levelname)s: %(message)s '
        '[in %(filename)s:%(lineno)d] -- ip: %(clientip)s, '
        'url: %(url)s, method:%(method)s'))

    # the default log level is set to WARNING, so
    # we have to explicitly set the logging level
    # to INFO to get our custom message logged.

    # append the handler to the default application logger

    # let's go

Currently only exceptions raised by the MongoDB layer and POST, PATCH and PUT methods are logged. The idea is to also add some INFO and possibly DEBUG level events in the future.

Operations Log

The OpLog is an API-wide log of all edit operations. Every POST, PATCH PUT and DELETE operation can be recorded to the oplog. At its core the oplog is simply a server log. What makes it a little bit different is that it can be exposed as a read-only endpoint, thus allowing clients to query it as they would with any other API endpoint.

Every oplog entry contains information about the document and the operation:

  • Operation performed
  • Unique ID of the document
  • Update date
  • Creation date
  • Resource endpoint URL
  • User token, if User-Restricted Resource Access is enabled for the endpoint
  • Optional custom data

Like any other API-maintained document, oplog entries also expose:

  • Entry ID
  • ETag
  • HATEOAS fields if that’s enabled.

If OPLOG_AUDIT is enabled entries also expose:

  • client IP
  • Username or token, if available
  • changes applied to the document (for DELETE the whole document is included).

A typical oplog entry looks like this:

    "o": "DELETE",
    "r": "people",
    "i": "542d118938345b614ea75b3c",
    "c": {...},
    "ip": "",
    "u": "admin",
    "_updated": "Fri, 03 Oct 2014 08:16:52 GMT",
    "_created": "Fri, 03 Oct 2014 08:16:52 GMT",
    "_etag": "e17218fbca41cb0ee6a5a5933fb9ee4f4ca7e5d6"
    "_id": "542e5b7438345b6dadf95ba5",
    "_links": {...},

To save a little space (at least on MongoDB) field names have been shortened:

  • o stands for operation performed
  • r stands for resource endpoint
  • i stands for document id
  • ip is the client IP
  • u stands for user (or token)
  • c stands for changes occurred
  • extra is an optional field which you can use to store custom data

_created and _updated are relative to the target document, which comes handy in a variety of scenarios (like when the oplog is available to clients, more on this later).

Please note that by default the c (changes) field is not included for POST operations. You can add POST to the OPLOG_CHANGE_METHODS setting (see Global Configuration) if you wish the whole document to be included on every insertion.

How is the oplog operated?

Seven settings are dedicated to the OpLog:

  • OPLOG switches the oplog feature on and off. Defaults to False.
  • OPLOG_NAME is the name of the oplog collection on the database. Defaults to oplog.
  • OPLOG_METHODS is a list of HTTP methods to be logged. Defaults to all of them.
  • OPLOG_ENDPOINT is the endpoint name. Defaults to None.
  • OPLOG_AUDIT if enabled, IP addresses and changes are also logged. Defaults to True.
  • OPLOG_CHANGE_METHODS determines which methods will log changes. Defaults to [‘PATCH’, ‘PUT’, ‘DELETE’].
  • OPLOG_RETURN_EXTRA_FIELD determines if the optional extra field should be returned by the OPLOG_ENDPOINT. Defaults to False.

As you can see the oplog feature is turned off by default. Also, since OPLOG_ENDPOINT defaults to None, even if you switch the feature on no public oplog endpoint will be available. You will have to explicitly set the endpoint name in order to expose your oplog to the public.

The Oplog endpoint

Since the oplog endpoint is nothing but a standard API endpoint, you can customize it. This allows for setting up custom authentication (you might want this resource to be only accessible for administrative purposes) or any other useful setting.

Note that while you can change most of its settings, the endpoint will always be read-only so setting either resource_methods or item_methods to something other than ['GET'] will serve no purpose. Also, unless you need to customize it, adding an oplog entry to the domain is not really necessary as it will be added for you automatically.

Exposing the oplog as an endpoint could be useful in scenarios where you have multiple clients (say phone, tablet, web and desktop apps) which need to stay in sync with each other and the server. Instead of hitting every single endpoint they could just access the oplog to learn all that’s happened since their last access. That’s a single request versus several. This is not always the best approach a client could take. Sometimes it is probably better to only query for changes on a certain endpoint. That’s also possible, just query the oplog for changes occured on that endpoint.

Extending Oplog entries

Every time the oplog is about to be updated the on_oplog_push event is fired. You can hook one or more callback functions to this event. Callbacks receive resource and entries as arguments. The former is the resource name while the latter is a list of oplog entries which are about to be written to disk.

Your callback can add an optional extra field to canonical oplog entries. The field can be of any type. In this example we are adding a custom dict to each entry:

def oplog_extras(resource, entries):
    for entry in entries:
        entry['extra'] = {'myfield': 'myvalue'}

app = Eve()

app.on_oplog_push += oplog_extras

Please note that unless you explicitly set OPLOG_RETURN_EXTRA_FIELD to True, the extra field will not be returned by the OPLOG_ENDPOINT.


Are you on MongoDB? Consider making the oplog a capped collection. Also, in case you are wondering yes, the Eve oplog is blatantly inspired by the awesome Replica Set Oplog.

The Schema Endpoint

Resource schema can be exposed to API clients by enabling Eve’s schema endpoint. To do so, set the SCHEMA_ENDPOINT configuration option to the API endpoint name from which you want to serve schema data. Once enabled, Eve will treat the endpoint as a read only resource containing JSON encoded Cerberus schema definitions, indexed by resource name. Resource visibility and authorization settings are honored, so internal resources or resources for which a request does not have read authentication will not be accessible at the schema endpoint. By default, SCHEMA_ENDPOINT is set to None.

MongoDB Aggregation Framework

Support for the MongoDB Aggregation Framework is built-in. In the example below (taken from PyMongo) we’ll perform a simple aggregation to count the number of occurrences for each tag in the tags array, across the entire collection. To achieve this we need to pass in three operations to the pipeline. First, we need to unwind the tags array, then group by the tags and sum them up, finally we sort by count.

As python dictionaries don’t maintain order you should use SON or collections OrderedDict where explicit ordering is required eg $sort:

posts = {
    'datasource': {
        'aggregation': {
            'pipeline': [
                {"$unwind": "$tags"},
                {"$group": {"_id": "$tags", "count": {"$sum": 1}}},
                {"$sort": SON([("count", -1), ("_id", -1)])}

The pipeline above is static. You have the option to allow for dynamic pipelines, whereas the client will directly influence the aggregation results. Let’s update the pipeline a little bit:

posts = {
    'datasource': {
        'aggregation': {
            'pipeline': [
                {"$unwind": "$tags"},
                {"$group": {"_id": "$tags", "count": {"$sum": "$value"}}},
                {"$sort": SON([("count", -1), ("_id", -1)])}

As you can see the count field is now going to sum the value of $value, which will be set by the client upon performing the request:

$ curl -i http://example.com/posts?aggregate={"$value": 2}

The request above will cause the aggregation to be executed on the server with a count field configured as if it was a static {"$sum": 2}. The client simply adds the aggregate query parameter and then passes a dictionary with field/value pairs. Like with all other keywords, you can change aggregate to a keyword of your liking, just set QUERY_AGGREGATION in your settings.

You can also set all options natively supported by PyMongo. For more information on aggregation see Advanced Datasource Patterns.

You can pass {} to fields which you want to ignore. Considering the following pipelines:

posts = {
    'datasource': {
        'aggregation': {
            'pipeline': [
                {"$match": { "name": "$name", "time": "$time"}}
                {"$unwind": "$tags"},
                {"$group": {"_id": "$tags", "count": {"$sum": 1}}},

If performing the following request:

$ curl -i http://example.com/posts?aggregate={"$name": {"$regex": "Apple"}, "$time": {}}

The stage {"$match": { "name": "$name", "time": "$time"}} in the pipeline will be executed as {"$match": { "name": {"$regex": "Apple"}}}. And for the following request:

$ curl -i http://example.com/posts?aggregate={"$name": {}, "$time": {}}

The stage {"$match": { "name": "$name", "time": "$time"}} in the pipeline will be completely skipped.

The request above will ignore "count": {"$sum": "$value"}}. A Custom callback functions can be attached to the before_aggregation and after_aggregation event hooks. For more information, see Aggregation event hooks.


HATEOAS is not available at aggregation endpoints. This should not be surprising as documents returned by these endpoints are aggregation results and do not reside on the database, so there is no static link available for them.

Client pagination (?page=2) is enabled by default. This is currently achieved by injecting two additional stages ($limit first, then $skip) to the very end of the aggregation pipeline. You can turn pagination off by setting pagination to False for the endpoint. Keep in mind that, when pagination is disabled, all aggregation results are included with every response. Disabling pagination might be appropriate (and actually advisable) only if the expected response payload is not huge.

Client sorting (?sort=field1) is not supported at aggregation endpoints. You can of course add one or more $sort stages to the pipeline, as we did with the example above. If you do add a $sort stage to the pipeline, consider adding it at the end of the pipeline. According to MongoDB’s $limit documentation (link):

When a $sort immediately precedes a $limit in the pipeline, the sort operation only maintains the top n results as it progresses, where n is the specified limit, and MongoDB only needs to store n items in memory.

As we just saw earlier, pagination adds a $limit stage to the end of the pipeline. So if pagination is enabled and $sort is the last stage of your pipeline, then the resulting combined pipeline should be optimized.

A single endpoint cannot serve both regular and aggregation results. However, since it is possible to setup multiple endpoints all serving from the same datasource (see Multiple API Endpoints, One Datasource), similar functionality can be easily achieved.

MongoDB and SQL Support

Support for single or multiple MongoDB database/servers comes out of the box. An SQLAlchemy extension provides support for SQL backends. Additional data layers can can be developed with relative ease. Visit the extensions page for a list of community developed data layers and extensions.

Powered by Flask

Eve is based on the Flask micro web framework. Actually, Eve itself is a Flask subclass, which means that Eve exposes all of Flask functionalities and niceties, like a built-in development server and debugger, integrated support for unittesting and an extensive documentation.