An HTTP server to fetch data from a database, transform it via jq, and serve it over an API.

02a66fb Upgrade minimum Go version to 1.18

~nilium pushed to ~nilium/chisel git

4 months ago

New mailing list added

11 months ago


Chisel is a database frontend with support for transactions across multiple databases and query transformations via jq. Its main use is as a basic JSON API frontend to databases and for prototyping API work without the need to write a great deal of boilerplate to get started.

Currently, Chisel is a work-in-progress and only its basic functionality is implemented. This includes:

  • Listening on multiple addresses, and limiting endpoints to specific addresses.
  • Connecting to multiple databases for use in endpoints.
  • Defining HTTP endpoints, with support for both GET- and POST-style requests.
  • Executing a sequence of queries across one or more database transactions.
  • Parsing request query and path parameters using jq.
  • Mapping of database results, and passing database results to subsequent queries in transactions.


$ go install go.spiff.io/chisel@latest

Chisel does not currently have version tags, so this will always build and install the latest version to your GOBIN (defaults to ~/go/bin).

To build it locally, you can use bmake or pmake, which will build chisel with its default options as bin/chisel. This includes additional SQLite options that are not present when using go install on its own.

Otherwise, if you don't care for any of the default options, you can use go build in the root of the project, which will produce a chisel binary for you to run.


Chisel has very few CLI options currently, limited to the following:

Usage of chisel:

  • -C - Print the parsed program config as JSON and exit.
  • -c=config.json - The path to load program config JSON from. (default "config.json")
  • -v=level - Set the log level. May be one of info (default), warn, error, fatal, panic, debug, or trace.


Chisel is configured using JSON or YAML. Example configuration below uses YAML to provide comments alongside the data. The use of either is subject to change, possibly to codf later, so bear that in mind. Currently, the top-level elements of a Chisel configuration are:

  • bind ([]sockaddr): A list of addresses and ports to listen for connections on. If none are given, it defaults to (over IPv4 only). Each address is assumed to be IPv4, IPv6, or a Unix domain socket path. Unix domain socket paths must begin with / or . to identify them as socket paths.

      -    # IPv4 address and port.
      - '[::1]:8080'      # IPv6 address and port.
      - /var/run/chisel.s # Unix domain socket.

    Sockets in this list may be referred to by index later. Indices are zero-based, so 0 would refer to above, and 1 would refer to the Unix domain socket /var/run/chisel.s.

    This may change to provide named socket groups or treat all addresses as dual-stack where possible.

  • databases ([string]database): A mapping of database names to their configurations. See Databases below for the values these are configured with.

  • endpoints ([]endpoint): A list of endpoint definitions. See Endpoints below for the values these are configured with.


Every database has a name and a URL. Beyond that, all other values for a database are optional, and some are only configurable by setting values in the URL (these options are DBMS-specific and handed off to the driver).

A fully-define database connection is provided below for reference:

    url: sqlite://test.db # sqlite://, mysql://, postgres://
    # Connection limits:
    max_idle: 2      # Maximum idle connections.
    max_idle_time: 0 # Maximum idle connection lifespan.
    max_open: 0      # Maximum open connections.
    max_life_time: 0 # Maximum connection lifespan.
    # Query options:
      try_json: true       # Whether to try parsing values as JSON.
      skip_json: false     # Whether to skip all JSON parsing, even for JSON columns.
      time_format: rfc3339 # The format that times are parsed and rendered in.
      # Or:
      time_format: layout
      time_layout: ''

The following options from above are configurable:

  • url (string): A postgres://, mysql://, or sqlite:// database URL. Only limited support is available for SQLite and MySQL. MySQL has the least support due to it lacking an insert-returning statement, while Chisel's support for SQLite does not yet have locking and is experimental. If using MySQL, it is recommended you use MariaDB instead, which has support for an insert-returning statement.

    The format of a database URL is as follows, with optional parts wrapped in square brackets:

  • max_idle and max_open (int): These control max number of idle and open connections, respectively, for a database. By default, the maximum idle number is 2 and the maximum open is unlimited. These defaults are from Go, and may not always follow Go.

  • max_idle_time and max_life_time (duration string): These control the maximum lifespan of idle and open connections. By default, neither connection has a lifespan and will be open as long as is possible. These are formatted as Go duration strings, such as 5h4m3s2ms1us.

  • try_json (bool): If true, Chisel will attempt to parse all retrieved database values as JSON where it looks like it can. This applies to all columns with a text-like type, not only those with a JSON column type. Defaults to false.

  • skip_json (bool): If true, Chisel will skip parsing JSON on all retrieved database values, including those with JSON column types. This overrides try_json and forces it to false. Defaults to false.

  • time_format (enum): Sets the format of times both parsed from the database and returned in JSON. Time values are not parsed from JSON in HTTP request bodies. May be one of the following values:

    • rfc3339 (default) - format times as RFC 3339 date-time strings.
    • fsec - format times as Unix timestamps with sub-second fractional components.
    • unixns - format times as Unix timestamps in nanoseconds.
    • unixus - format times as Unix timestamps in microseconds.
    • unixms - format times as Unix timestamps in milliseconds.
    • unix - format times as Unix timestamps in seconds.
    • layout - format times using the layout string from time_layout.
  • time_layout (string): Sets the Go time layout string to parse and render times when time_format is set to layout.


Endpoints define the HTTP endpoints served on one or more bind addresses. An endpoint has the following top-level values:

  • bind ([]int): A set of one or more indices from the list of bind addresses. The endpoint will only be served on the addresses corresponding to the indices. The format of this field is subject to change, but may be used to limit certain endpoints to internal interfaces.

  • method (string, required): An HTTP method, such as GET or POST.

  • path (string, required): The HTTP path, rooted at /. You may define variable elements of the path by declaring them as :name, such as /things/:id/name, where :id is a path parameter name. To capture all subsequent path elements as a parameter, you can declare the final path element as *name, such as /file/at/*path, where all path elements after /file/at/ are captured as the path parameter. Path routing is currently handled by httprouter, so its behavior determines how paths are currently handled.

  • body_type (enum): The type of body to expect if METHOD is not GET or HEAD. May be one of the following:

    • json (default): Parse request bodies as JSON. If parsing fails, reject the request.
    • string: Read the body without parsing it and treat it as a string.
    • form: Parse the body as a form. Currently unsupported.
    • none: Do not attempt to read or parse the request body.
  • query_params, path_params ([string][]mapping): Mappings for query and path parameters, respectively. Parameters named in this are transformed with one or more mappings, allowing you to parse parameters as numbers, check if they match a regexp, or other features using jq expressions. For example:

        - tonumber | if . <= 0 then error("id must be a positive number") else . end

    Note: although you can pass multiple mappings per parameter, this may not be supported in the future.

  • query (query): Defines the query associated with the endpoint, including transactions and steps. See Queries below for more detail.


Queries have two top-level keys: transactions, which defines a list of transactions that make up an endpoint's queries; and steps, which defines a list of individual queries against those transactions and the transformations and arguments to each.


Transactions define one or more transactions for an endpoint when accessing a database, as well as the isolation level of each transaction. This is used to ensure that queries run by chisel have appropriate isolation from other queries.

Every transaction defined in an endpoint must be used by at least one step in the query.

  # 0
  - db: test
    isolation: serializable # Has a DBMS-level transaction.
  # 1
  - db: test
    isolation: none         # Has no DBMS-level transaction.
  • db (string): The list of transactions defines the databases that an endpoint accesses and the isolation level of each transaction against the database. The databases are referred to by their names in the root-level databases mapping.

  • isolation (enum): The isolation level of a transaction determines the kind of isolation the database gives the transaction. Not all databases are guaranteed to support all isolation levels (or even support them correctly if they do). Every database has its own default isolation level, and you should consult the DBMS documentation for yours.

    Valid isolation levels are:

    • default (default): Use the DBMS's default isolation level.
    • none: Do not create a transaction and instead run a one-off query against the database. This is useful for single-query endpoints and those that do not need to do things like perform multiple updates or inserts followed by selects.
    • read_uncommitted
    • read_committed
    • write_committed
    • repeatable_read
    • snapshot
    • serializable
    • linearizable

Query steps are the individual query statements, their arguments, and the transformations applied to their results.

  - transaction: 0
    query: SELECT * FROM builds WHERE id = ? LIMIT 1
    - path: id
    map: # Output is the first row.
    - first

  - transaction: 0
    query: SELECT id AS artifact_id, url AS artifact_url FROM artifacts WHERE build_id = ?
    # Fetch build ID from previous output.
    - expr: '$context.outputs[0][0].id'
    map: # Merge outputs.
    - '{ data: ({ artifacts: . } * $context.outputs[0]) }'

A step is defined by the following fields:

  • transaction (int): An index into the transactions list defined in the parent query. If not set, defaults to the first transaction as a convenience for single-transaction queries.

  • query (string, required): The query to run against the transaction. This can use ? parameters as placeholders for anything the DBMS permits for parameterization. Because this uses sqlx for parameter binding, cases like col IN (?) are expanded when list arguments (below) are given.

  • args ([]arg): The arguments passed to the above query. If the query doesn't take parameters, this must be empty or undefined. Each argument is defined in one of four ways:

    • A literal value, such as 1, "foo", or a list of literal values.
    • { path: "key" } - A mapping binding the argument to the value of a path parameter, defined on the endpoint. If the parameter is not defined, the request fails.
    • { query: "key" } - A mapping binding the argument to the value of a query parameter. As with path parameters, this must be defined for the request, or the request fails.
    • { expr: "jq" } - A mapping binding the argument to the result value of a jq expression. Composite return types such as mappings are encoded as JSON, while lists are passed to the query for binding. To ensure that a list is encoded as JSON, the result of the expression should include a final | tojson pipeline. An example of this can be seen above in the second step's argument list.
  • map ([]jqexpr): A list of jq expressions, encoded as strings, to define transformations of the result set into the output of the query step. The output is captured and passed to the next steps for reuse. If this is the final step, the result of the mapping is used as the JSON response body.

    A special case is used to adjust the resulting HTTP status and headers on response, by returning an object of the form:

      "__response": {
        "status": 201,
        "headers": {
          "x-foobar": ["something"]
        "data_key": "actual_body"
      "actual_body": {
        "data": "foobar"

    Where, in the above case, all values of the __response object are optional. If status is undefined, it defaults to HTTP 200 (OK).

    If data_key is defined, instead of returning the object minus the __response key, the response body will be returned using the data found at the expression .[$data_key]. So, in the above example, the actual response returned in the response is {"data":"foobar"}.


Chisel is licensed under the Apache 2.0 license. A copy of this license is included with the source code of the project.