quasar-analytics / quasar

Quasar Analytics is a general-purpose compiler for translating data processing and analytics over semi-structured data into efficient plans that run 100% in the target infrastructure.

Website Github

Build status Coverage Status Latest version Join the chat at https://gitter.im/quasar-analytics/quasar

Thanks to Sponsors

YourKit supports open source projects with its full-featured Java Profiler. YourKit, LLC is the creator of YourKit Java Profiler and YourKit .NET Profiler, innovative and intelligent tools for profiling Java and .NET applications.


Quasar is an open source NoSQL analytics engine that can be used as a library or through a REST API to power advanced analytics across a growing range of data sources and databases, including MongoDB.


SQL² is the dialect of SQL that Quasar understands.

SQL² is a superset of standard SQL. Therefore, in the following documentation SQL² will be used interchangeably with SQL.

See the SQL² tutorial for more info on SQL².

SQL² supports variables inside queries (SELECT * WHERE pop < :cutoff). Values for these variables, which can be any expression, should be specified as additional parameters in the url, using the variable name prefixed by var. (e.g. var.cutoff=1000). Failure to specify valid values for all variables used inside a query will result in an error. These values use the same syntax as the query itself; notably, strings should be surrounded by double quotes. Some acceptable values are 123, "CO", and DATE("2015-07-06").

Using the Pre-Built JARs

In Github Releases, you can find pre-built JARs for all the subprojects in this repository.

See the instructions below for running and configuring these JARs.

Building from Source

Note: This requires Java 8 and Bash (Linux, Mac, or Cygwin on Windows).


The following sections explain how to build and run the various subprojects.

Basic Compile & Test

To compile the project and run tests, first clone the quasar repo and then execute the following command:

./sbt test

Note: please note that we are not using here a system wide sbt, but our own copy of it (under ./sbt). This is primarily done for determinism. In order to have a reproducible build, the helper script needs to be part of the repo.

Running the full test suite can be done using docker containers for various backends:

Full Testing (prerequisite: docker and docker-compose)

In order to run integration tests for various backends the docker/scripts are provided to easily create dockerized backend data stores.

Of particular interest are the following two scripts:

  1. docker/scripts/setupContainers
  2. docker/scripts/assembleTestingConf

Quasar supports the following datastores:


Knowing which backend datastores are supported you can create and configure docker containers using setupContainers. For example if you wanted to run integration tests with mongo, postgresql, marklogic, and couchbsase you would use:

./setupContainers -u quasar_metastore,quasar_mongodb_3_0,quasar_postgresql,quasar_marklogic_xml,quasar_couchbase

Note: quasar_metastore is always needed to run integration tests.

This command will pull docker images, create containers running the specified backends, and configure them appropriately for Quasar testing.

Once backends are ready we need to configure the integrations tests in order to inform Quasar about where to find the backends to test. This information is conveyed to Quasar using the file it/testing.conf. Using the assembleTestingConf script you can generate a testing.conf file based on the currently running containerizd backends using the following command:

./assembleTestingConf -a

After running this command your testing.conf file should look similar to this:

> cat it/testing.conf

IP's will vary depending on your docker environment. In addition the scripts assume you have docker and docker-compose installed. You can find information about installing docker here.


To build a JAR for the REPL, which allows entering commands at a command-line prompt, execute the following command:

./sbt 'repl/assembly'

The path of the JAR will be ./repl/target/scala-2.11/quasar-repl-assembly-[version].jar, where [version] is the Quasar version number.

To run the JAR, execute the following command:

java -jar [<path to jar>] [-c <config file>]

As a command-line REPL user, to work with a fully functioning REPL you will need the metadata store and a mount point. See here for instructions on creating the metadata store backend using docker. To add a mount you can start the web server mentioned below and issue a curl command like:

curl -v -X PUT http://localhost:8080/mount/fs/cb/ -d '{ "couchbase": { "connectionUri":"couchbase://" } }'

You can find examples of connectionUri values here.


To build a JAR containing a lightweight HTTP server that allows you to programmatically interact with Quasar, execute the following command:

./sbt 'web/assembly'

The path of the JAR will be ./web/target/scala-2.11/quasar-web-assembly-[version].jar, where [version] is the Quasar version number.

To run the JAR, execute the following command:

java -jar [<path to jar>] [-c <config file>]

Web jar users, will also need the metadata store. See here for getting up and running with one using docker.


The various REPL JARs can be configured by using a command-line argument to indicate the location of a JSON configuration file. If no config file is specified, it is assumed to be quasar-config.json, from a standard location in the user's home directory.

The JSON configuration file must have the following format:

  "server": {
    "port": 8080
  "metastore": {
    "database": {

Metadata Store

Configuration for the metadata store consists of providing connection information for a supported database. Currently the H2 and PostgreSQL (9.5+) databases are supported.

To easily get up and running with a PostgreSQL metastore backend using docker see Full Testing section.

If no metastore configuration is specified, the default configuration will use an H2 database located in the default quasar configuration directory for your operating system.

An example H2 configuration would look something like

"h2": {
  "file": "<path/to/database/file>"

A PostgreSQL configuration looks something like

"postgresql": {
  "host": "<hostname>",
  "port": "<port>",
  "database": "<database name>",
  "userName": "<database user>",
  "password": "<password for database user>",
  "parameters": <an optional JSON object of parameter key:value pairs>

The contents of the optional parameters object correspond to the various driver configuration parameters available for PostgreSQL. One example for a value of the parameters object may be a loglevel:

"parameters": {
  "loglevel": 1

Initializing and updating Schema

Before the server can be started, the metadata store schema must be initialized. To do so utilize the "initUpdateMetaStore" command with a web or repl quasar jar.

If mounts are already defined in the config file, initialization will migrate those to the metadata store.

Database mounts

If the mount's key is "mongodb", then the connectionUri is a standard MongoDB connection string. Only the primary host is required to be present, however in most cases a database name should be specified as well. Additional hosts and options may be included as specified in the linked documentation.

For example, say a MongoDB instance is running on the default port on the same machine as Quasar, and contains databases test and students, the students database contains a collection cs101, and the connectionUri is mongodb://localhost/test. Then the filesystem will contain the paths /local/test/ and /local/students/cs101, among others.

A database can be mounted at any directory path, but database mount paths must not be nested inside each other.


To connect to MongoDB using TLS/SSL, specify ?ssl=true in the connection string, and also provide the following via system properties when launching either JAR (i.e. java -Djavax.net.ssl.trustStore=/home/quasar/ssl/certs.ts):

  • javax.net.ssl.trustStore: path specifying a file containing the certificate chain for verifying the server.
  • javax.net.ssl.trustStorePassword: password for the trust store.
  • javax.net.ssl.keyStore: path specifying a file containing the client's private key.
  • javax.net.ssl.keyStorePassword: password for the key store.
  • javax.net.debug: (optional) use all for very verbose but sometimes helpful output.
  • invalidHostNameAllowed: (optional) use true to disable host name checking, which is less secure but may be needed in test environments using self-signed certificates.


To connect to Couchbase use the following connectionUri format:



  • Couchbase Server 4.5.1 or greater
  • A "default" bucket with anonymous access
  • Documents must have a "type" field to be listed
  • Primary index on queried buckets
  • Secondary index on "type" field for queried buckets
  • Additional indices and tuning as recommended by Couchbase for proper N1QL performance

Known Limitations

  • Slow queries — query optimization hasn't been applied
  • Join unimplemented — future support planned
  • Open issues

HDFS using Apache Spark

To connect to HDFS using Apache Spark use the following connectionUri format:


e.g "spark://spark_master:7077|hdfs://primary_node:9000|/hadoop/users/"


To connect to MarkLogic, specify an XCC URL with a format query parameter and an optional root directory as the connectionUri:


the mount will query either JSON or XML documents based on the value of the format parameter. For backwards-compatibility, if the format parameter is omitted then XML is assumed.

If a root directory path is specified, all operations and queries within the mount will be local to the MarkLogic directory at the specified path.


  • MarkLogic 8.0+
  • Documents must to be organized under directories to be found by Quasar.
  • Namespaces used in queries must be defined on the server.
  • Loading schema definitions into the server, while not required, will improve sorting and other operations on types other than xs:string. Otherwise, non-string fields may require casting in queries using SQL² conversion functions.

Known Limitations

  • Field aliases when working with XML must currently be valid XML QNames (#1642).
  • "Default" numeric field names are prefixed with an underscore ("_") when working with XML in order to make them valid QNames. For example, select count((1, 2, 3, 4)) will result in {"_1": 4} (#1642).
  • It is not possible to query both JSON and XML documents from a single mount, a separate mount with the appropriate format value must be created for each type of document.
  • Index usage is currently poor, so performance may degrade on large directories and/or complex queries and joins. This should improve as optimizations are applied both to the MarkLogic connector and the QScript compiler.

Quasar's data model is JSON-ish and thus there is a bit of translation required when applying it to XML. The current mapping aims to be intuitive while still taking advantage of the XDM as much as possible. Take note of the following:

  • Projecting a field will result in the child element(s) having the given name. If more than one element matches, the result will be an array.
  • As the children of an element form a sequence, they may be treated both as a mapping from element names to values and as an array of values. That is to say, given a document like <foo><bar>1</bar><baz>2</baz></foo>, foo.bar and foo[0] both refer to <bar>1</bar>.
  • XML document results are currently serialized to JSON with an emphasis on producting idiomatic JSON:
    • An element is serialized to a singleton object with the element name as the only key and an object representing the children as its value. The child object will contain an entry for each child element with repeated elements collected into an array.
    • An element without attributes containing only text content will be serialized as a singleton object with the element name as the only key and the text content as its value.
    • Element attributes are serialized to an object at the _attributes key.
    • Text content of elements containing mixed text and element children or attributes will be available at the _text key.

View mounts

If the mount's key is "view" then the mount represents a "virtual" file, defined by a SQL² query. When the file's contents are read or referred to, the query is executed to generate the current result on-demand. A view can be used to create dynamic data that combines analysis and formatting of existing files without creating temporary results that need to be manually regenerated when sources are updated.

For example, given the above MongoDB mount, an additional view could be defined with a connectionUri of sql2:///?q=select%20_id%20as%20zip%2C%20city%2C%20state%20from%20%60%2Flocal%2Ftest%2Fzips%60%20where%20pop%20%3C%20%3Acutoff&var.cutoff=1000

A view can be mounted at any file path. If a view's path is nested inside the path of a database mount, it will appear alongside the other files in the database. A view will "shadow" any actual file that would otherwise be mapped to the same path. Any attempt to write data to a view will result in an error.

Build Quasar for Apache Spark

In order for Quasar to work with Apache Spark based connectors (like spark-hdfs or spark-local) you need to build sparkcore.jar and move it to same location where your quasar-web.jar is placed. To build sparkcore.jar:

./sbt 'set every sparkDependencyProvided := true' sparkcore/assembly

REPL Usage

The interactive REPL accepts SQL SELECT queries.

First, choose the database to be used. Here, a MongoDB instance is mounted at the root, and it contains a database called test:

💪 $ cd test

The "tables" in SQL queries refer to collections in the database by name:

💪 $ select * from zips where state="CO" limit 3
    { "$match": { "state": "CO" } },
    { "$limit": NumberLong(3) },
    { "$out": "tmp.gen_0" }],
  { "allowDiskUse": true });

Query time: 0.1s
 city    | loc[0]       | loc[1]     | pop    | state |
 ARVADA  |  -105.098402 |  39.794533 |  12065 | CO    |
 ARVADA  |  -105.065549 |  39.828572 |  32980 | CO    |
 ARVADA  |   -105.11771 |  39.814066 |  33260 | CO    |

💪 $ select city from zips limit 3
 city     |
 BARRE    |

You may also store the result of a SQL query:

💪 $ out1 := select * from zips where state="CO" limit 3

The location of a collection may be specified as an absolute path by surrounding the path with double quotes:

select * from `/test/zips`

Type help for information on other commands.

API Usage

The server provides a simple JSON API.

GET /query/fs/[path]?q=[query]&offset=[offset]&limit=[limit]&var.[foo]=[value]

Executes a SQL² query, contained in the required q parameter, on the backend responsible for the request path.

Optional offset and limit parameters can be specified to page through the results, and are interpreted the same way as for GET /data requests.

The result is returned in the response body. The Accept header may be used in order to specify the desired format in which the client wishes to receive results.

For compressed output use Accept-Encoding: gzip.

POST /query/fs/[path]?var.[foo]=[value]

Executes a SQL² query, contained in the request body, on the backend responsible for the request path.

The Destination header must specify the output path, where the results of the query will become available if this API successfully completes. If the output path already exists, it will be overwritten with the query results.

All paths referenced in the query, as well as the output path, are interpreted as relative to the request path, unless they begin with /.

This API method returns the name where the results are stored, as an absolute path, as well as logging information.

  "out": "/[path]/tmp231",
  "phases": [

If the query fails to compile, a 400 response is produced with a JSON body similar to the following:

  "status": "Bad Request",
  "detail": {
    "errors" [
      <all errors produced during compilation, each an object with `status` and `detail` fields>
    "phases": [
      <see the following sections>

If an error occurs while executing the query on a backend, a 500 response is produced, with this content:

  "status": <general error description>,
  "detail": {
    "message": <specific error description>,
    "phases": [
      <see the following sections>
    "logicalPlan": <tree of objects describing the logical plan the query compiled to>,
    "cause": <optional, backend-specific error>

the cause field is optional and the detail object may also contain additional, backend-specific fields.

The phases array contains a sequence of objects containing the result from each phase of the query compilation process. A phase may result in a tree of objects with type, label and (optional) children:

  "phases": [
      "name": "Logical Plan",
      "tree": {
        "type": "LogicalPlan/Let",
        "label": "'tmp0",
        "children": [
            "type": "LogicalPlan/Read",
            "label": "./zips"

Or a blob of text:

  "phases": [
      "name": "Mongo",
      "detail": "db.zips.aggregate([\n  { \"$sort\" : { \"pop\" : 1}}\n])\n"

Or an error (typically no further phases appear, and the error repeats the error at the root of the response):

  "phases": [
      "name": "Physical Plan",
      "error": "Cannot compile ..."

GET /compile/fs/[path]?q=[query]&var.[foo]=[value]

Compiles (but does not execute) a SQL² query, contained in the single, required query parameter. Returns a Json object with the following shape:

  "inputs": [<filePath>, ...],
  "physicalPlan": "Description of physical plan"

where `inputs` is a field containing a list of files that are referenced by the query.
where `physicalPlan` is a string description of the physical plan that would be executed by this query. `null` if no physical plan is required in order to execute this query. A query may not need a physical plan in order to be executed if the query is "constant", that is that no data needs to be read from a backend.

### GET /metadata/fs/[path]

Retrieves metadata about the files, directories, and mounts which are children of the specified directory path. If the path names a file, the result is empty.

  "children": [
    {"name": "foo", "type": "directory"},
    {"name": "bar", "type": "file"},
    {"name": "test", "type": "directory", "mount": "mongodb"},
    {"name": "baz", "type": "file", "mount": "view"}

GET /data/fs/[path]?offset=[offset]&limit=[limit]

Retrieves data from the specified path in the format specified in the Accept header. The optional offset and limit parameters can be used in order to page through results.


If the supplied path represents a directory (ends with a slash), this request produces a zip archive containing the contents of the named directory, database, etc. Each file in the archive is formatted as specified in the request query and/or Accept header.

PUT /data/fs/[path]

Replace data at the specified path. Uploaded data may be in any of the supported formats and the request must include the appropriate Content-Type header indicating the format used.

A successful upload will replace any previous contents atomically, leaving them unchanged if an error occurs.

If an error occurs when reading data from the request body, the response will contain a summary in the common error field and a separate array of error messages about specific values under details.

Fails if the path identifies an existing view.

Uploading multpile files

If the supplied path represents a directory (ends with a slash), the request body must contain a zip archive containing the contents of the named directory, database, etc., and a special file, /.quasar-metadata.json, which specifies the format for each file, as it would be provided in a Content-Type header if the file was individually uploaded:

  "/foo": {
    "Content-Type": "application/ldjson"
  "/foo/bar": {
    "Content-Type": "application/json; mode=precise"

Note: if the zip archive was created by downloading a directory from Quasar, then it will already have this hidden file.

Each file in the archive is written as if it was uploaded separately. The write is not atomic; if an error occurs after some files are written, the file system is not restored to its previous state.

POST /data/fs/[path]

Append data to the specified path. Uploaded data may be in any of the supported formats and the request must include the appropriate Content-Type header indicating the format used. This operation is not atomic and some data may have been written even if an error occurs. The body of an error response will describe what was done.

If an error occurs when reading data from the request body, the response contains a summary in the common error field, and a separate array of error messages about specific values under details.

Fails if the path identifies an existing view.

DELETE /data/fs/[path]

Removes all data and views at the specified path. Single files are deleted atomically.

MOVE /data/fs/[path]

Moves data from one path to another within the same backend. The new path must be provided in the Destination request header. Single files are moved atomically.

GET /mount/fs/[path]

Retrieves the configuration for the mount point at the provided path. In the case of MongoDB, the response will look like

{ "mongodb": { "connectionUri": "mongodb://localhost/test" } }

The outer key is the backend in use, and the value is a backend-specific configuration structure.

POST /mount/fs/[path]

Adds a new mount point using the JSON contained in the body. The path is the containing directory, and an X-File-Name header should contain the name of the mount. This will return a 409 Conflict if the mount point already exists or if a database mount already exists above or below a new database mount.

PUT /mount/fs/[path]

Creates a new mount point or replaces an existing mount point using the JSON contained in the body. This will return a 409 Conflict if a database mount already exists above or below a new database mount.

DELETE /mount/fs/[path]

Deletes an existing mount point, if any exists at the given path. If no such mount exists, the request succeeds but the response has no content. Mounts that are nested within the mount being deleted (i.e. views) are also deleted.

MOVE /mount/fs/[path]

Moves a mount from one path to another. The new path must be provided in the Destination request header. This will return a 409 Conflict if a database mount is being moved above or below the path of an existing database mount. Mounts that are nested within the mount being moved (i.e. views) are moved along with it.

PUT /server/port

Takes a port number in the body, and attempts to restart the server on that port, shutting down the current instance which is running on the port used to make this http request.

DELETE /server/port

Removes any configured port, reverting to the default (20223) and restarting, as with PUT.

GET /invoke/fs/[path]

Where path is a file path. Invokes the function represented by the file path with the parameters supplied in the query string.

Error Responses

Error responses from the REST api have the following form

  "error": {
    "status": <succinct message>,
    "detail": {
      "field1": <JSON>,
      "field2": <JSON>,
      "fieldN": <JSON>

The status field will always be present and will contain a succinct description of the error in english, the same content will be used as the status message of the HTTP response itself. The detail field is optional and, if present, will contain a JSON object with additional information about the error.

Examples of detail fields would be a backend-specific error message, detailed type information for type errors in queries, the actual invalid arguments presented to a function, etc. These fields are error-specific, however, if the error is going to include a more detailed error message, it will found under the message field in the detail object.


Paths identify files and directories in Quasar's virtual file system. File and directory paths are distinct, so /foo and /foo/ represent a file and a directory, respectively.

Depending on the backend, some restrictions may apply:

  • it may be possible for a file and directory with the same name to exist side by side.
  • it may not be possible for an empty directory to exist. That is, deleting the only descendant file from a directory may cause the directory to disappear as well.
  • there may be limits on the overall length of paths, and/or the length of particular path segments. Any request that exceeds these limits will result in an error.

Any character can appear in a path, but when paths are embedded in character strings and byte-streams they are encoded in the following ways:

When a path appears in a request URI, or in a header such as Destination or X-FileName, it must be URL-encoded. Note: / characters that appear within path segments are encoded.

When a path appears in a JSON string value, / characters that appear within path segments are encoded as $sep$.

In both cases, the special names . and .. are encoded as $dot$ and $dotdot$`, but only if they appear as an entire segment.

When only a single path segment is shown, as in the response body of a /metadata request, no special encoding is done (beyond the normal JSON encoding of " and non-ASCII characters).

For example, a file called Plan 1/2 笑 in a directory mydata would appear in the following ways:

  • in a URL: http://<host>:<port>/data/fs/mydata/Plan%201%2F2%20%E7%AC%91
  • in a header: Destination: /mydata/Plan%201%2F2%20%E7%AC%91
  • in the response body of /metadata/fs/mydata/: { "type": "file", "name": "Plan 1/2 \u7b11" }
  • in an error:
  "error": {
    "status": "Path not found.",
    "detail": {
      "path": "/local/quasar-test/mydata/Plan 1$sep$2 \u7b11"

Request Headers

Request headers may be supplied via a query parameter in case the client is unable to send arbitrary headers (e.g. browsers, in certain circumstances). The parameter name is request-headers and the value should be a JSON-formatted string containing an object whose fields are named for the corresponding header and whose values are strings or arrays of strings. If any header appears both in the request-headers query parameter and also as an ordinary header, the query parameter takes precedence.

For example:

GET http://localhost:8080/data/fs/local/test/foo?request-headers=%7B%22Accept%22%3A+%22text%2Fcsv%22%7D

Note: that's the URL-encoded form of {"Accept": "text/csv"}.

Data Formats

Quasar produces and accepts data in two JSON-based formats or CSV (text/csv). Each JSON-based format can represent all the types of data that Quasar supports. The two formats are appropriate for different purposes.

Json can either be line delimited (application/ldjson/application/x-ldjson) or a single json value (application/json).

In the case of an HTTP request, it is possible to add the disposition extension to any media-type specified in an Accept header in order to receive a response with that value in the Content-Disposition header field.

Choosing between the two json formats is done using the "mode" content-type extension and by supplying either the "precise" or "readable" values. If no mode is supplied, quasar will default to the readable mode. If neither json nor csv is supplied, quasar will default to returning the results in json format. In the case of an upload request, the client MUST supply a media-type and requests without any media-type will result in an HTTP 415 error response.

Precise JSON

This format is unambiguous, allowing every value of every type to be specified. It's useful for entering data, and for extracting data to be read by software (as opposed to people.) Contains extra information that can make it harder to read.

Readable JSON

This format is easy to read and use with other tools, and contains minimal extra information. It does not always convey the precise type of the source data, and does not allow all values to be specified. For example, it's not possible to tell the difference between the string "12:34:56" and the time value equal to 34 minutes and 56 seconds after noon.


Type Readable Precise Notes
null null same
boolean true, false same
string "abc" same
int 1 same
decimal 2.1 same
object { "a": 1 } same
object { "$foo": 2 } { "$obj": { "$foo": 2 } } Requires a type-specifier if any key starts with $.
array [1, 2, 3] same
set [1, 2, 3] { "$set": [1, 2, 3] }
timestamp "2015-01-31T10:30:00Z" { "$timestamp": "2015-01-31T10:30:00Z" }
date "2015-01-31" { "$date": "2015-01-31" }
time "10:30:05" { "$time": "10:30:05" } HH:MM[:SS[:.SSS]]
interval "PT12H34M" { "$interval": "P7DT12H34M" } Note: year/month not currently supported.
binary "TE1OTw==" { "$binary": "TE1OTw==" } BASE64-encoded.
object id "abc" { "$oid": "abc" }


When Quasar produces CSV, all fields and array elements are "flattened" so that each column in the output contains the data for a single location in the source document. For example, the document { "foo": { "bar": 1, "baz": 2 } } becomes


Data is formatted the same way as the "Readable" JSON format, except that all values including null, true, false, and numbers are indistinguishable from their string representations.

It is possible to use the columnDelimiter, rowDelimiter quoteChar and escapeChar media-type extensions keys in order to customize the layout of the csv. If some or all of these extensions are not specified, they will default to the following values:

  • columnDelimiter: ,
  • rowDelimiter: \r\n
  • quoteChar: "
  • escapeChar: "

Note: Due to the following issue in one of our dependencies. The rowDelimiter extension will be ignored for any CSV being uploaded. The rowDelimiter extension will, however, be observed for downloaded data. Also due to this issue best to avoid non "standard" csv formats. See the MessageFormatGen.scala file for examples of which csv formats we test against.

When data is uploaded in CSV format, the headers are interpreted as field names in the same way. As with the Readable JSON format, any string that can be interpreted as another kind of value will be, so for example there's no way to specify the string "null".


First, make sure that the quasar-analytics/quasar Github repo is building correctly (the status is displayed at the top of the README).

Then, you can try the following command:

./sbt test

This will ensure that your local version is also passing the tests.

Check to see if the problem you are having is mentioned in the JIRA issues and, if it isn't, feel free to create a new issue.

You can also discuss issues on Gitter: quasar-analytics/quasar.


Copyright © 2014 - 2017 SlamData Inc.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at


Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.