grpcui
is a command-line tool that lets you interact with gRPC servers via a browser.
It's sort of like Postman, but for gRPC APIs instead of
REST.
In some ways, this is like an extension to grpcurl.
Whereas grpcurl
is a command-line interface, grpcui
provides a web/browser-based
GUI. This lets you interactively construct requests to send to a gRPC server.
With this tool you can also browse the schema for gRPC services, which is presented as a
list of available endpoints. This is enabled either by querying a server that supports
server reflection,
by reading proto source files, or by loading in compiled "protoset" files (files that contain
encoded file descriptor protos).
In fact, the way the tool transforms JSON request data into a binary encoded protobuf
is using that very same schema. So, if the server you interact with does not support
reflection, you will either need the proto source files that define the service or need
protoset files that grpcui
can use.
This repo also provides two library packages
github.com/fullstorydev/grpcui
: This package contains the building blocks for embedding a gRPC web form into any Go HTTP server. It has functions for accessing the HTML form, the JavaScript code that powers it, as well as a sample CSS file, for styling the form.github.com/fullstorydev/grpcui/standalone
: This package goes a step further and supplies a single, simple HTTP handler that provides the entire gRPC web UI. You can just wire this handler into your HTTP server to embed a gRPC web page that looks exactly like the one you see when you use thegrpcui
command-line program. This single handler uses the above package but also supplies the enclosing HTML page, some other script dependencies (jQuery and jQuery-UI), and additional CSS and image resources.
grpcui
supports all kinds of RPC methods, including streaming methods. However, it requires
you to construct the entire stream of request messages all at once and then renders the entire
resulting stream of response messages all at once (so you can't interact with bidirectional
streams the way that grpcurl
can).
grpcui
supports both plain-text and TLS servers and has numerous options for TLS
configuration. It also supports mutual TLS, where the client is required to present a
client certificate.
As mentioned above, grpcui
works seamlessly if the server supports the reflection
service. If not, you can supply the .proto
source files or you can supply protoset
files (containing compiled descriptors, produced by protoc
) to grpcui
.
The web UI allows you to set request metadata in addition to defining the request message data.
When defining request message data, it uses a dynamic HTML form that supports data entry for
all possible kinds of protobuf messages, including rich support for well-known types (such as
google.protobuf.Timestamp
), one ofs, and maps.
In addition to entering the data via HTML form, you can also enter the data in JSON format, by typing or pasting the entire JSON request body into a text form.
Upon issuing an RPC, the web UI shows all gRPC response metadata, including both headers and trailers sent by the server. And, of course, it shows a human-comprehensible response body, in the form of an HTML table.
You can use the go
tool to install grpcui
:
go install github.com/fullstorydev/grpcui/cmd/grpcui@latest
This installs the command into the bin
sub-folder of wherever your $GOPATH
environment variable points. If this directory is already in your $PATH
, then
you should be good to go.
If you have already pulled down this repo to a location that is not in your
$GOPATH
and want to build from the sources, you can cd
into the repo and then
run make install
.
If you encounter compile errors, you could have out-dated versions of grpcui
's
dependencies. You can update the dependencies by running make updatedeps
.
The usage doc for the tool explains the numerous options:
grpcui -help
Most of the flags control how the program connects to the gRPC server that to which
requests will be sent. However, there is one flag that controls grpcui
itself: the
-port
flag controls what port the HTTP server should use to expose the web UI. If
no port is specified, an ephemeral port will be used (so likely a different port each
time it is run, allocated by the operating system).
When you run grpcui
, it will show you a URL to put into a browser in order to access
the web UI.
$ grpcui -plaintext localhost:12345
gRPC Web UI available at http://127.0.0.1:60551/...
When you navigate to this URL, you are presented with the user interface:
The top two listboxes allow you to select the service and method of the RPC to issue. Once a selection is made, the panel below will show a form that allows you to define an RPC request. The form is constructed, dynamically, based on the actual request message structure of the selected RPC.
You'll notice a second tab that lets you view (and edit) the raw JSON value for the request data. This can be useful to copy+paste a large request message, without having to point-and-click to define each field value, one at a time.
The third tab shows the response data. This tab is grayed out and disabled until you actually click the "Invoke" button, which can be found at the bottom of the page.
The first thing to note about the form is that it will generally be a table, where each row is a field. The table has three important columns:
- The first column shows the name and type of the field.
- The second columns indicates the "cardinality" of the field. Typical fields are optional. The second column for optional fields is a checkbox indicating whether the field is present or not. If the field is not present, its default value is assumed. Repeated fields show buttons in this column for adding and deletig values. The green "+" allows you to add values to the repeated field. The red "x" next to a value will remove that value. Finally, if the field is required (only supported in syntax "proto2"), the column will contain only an asterisk.
- The third column shows the value of the field. If the field is absent, this will show the default value for the field. Fields that are nested messages show "unset", but scalar types show their default (usually the type's zero value, but default values are configurable when using syntax "proto2"). Absent fields are also not editable -- you must first check the box in column two to make the field present before editing its value. Repeated fields show a nested table that occupies columns two and three and allows for adding and removing values.
Fields whose type is a nested message will include a nested table in column three. This nested table has its own three columns and one row per field therein.
One-ofs are rendered a little differently. Instead of two columns indicating the presence and value of the field, they include a nested table showing all of the possible fields in the one-of. However, the middle column is a radio button instead of a checkbox, so that only one of the fields can be present at any given time. In addition to a row for each field in the one-of, there is also an option named None, which indicates a one-of where no value is set.
Here's an example form for a message that has two required fields (id
and name
), one
repeated field (labels
), a handful of normal optional fields, and a single one-of that
has four options. In the first image, no values are present (except, of course, for the
required fields at the top). In the second, several field values are present.
For RPCs that accept a stream of requests, the web form allows the user to define multiple messages in the stream. It defaults to a single request, but the user can remove it to send none or can send many. A stream resembles a repeated field, but the repeated "thing" is the entire request:
That last example also shows how well-known message types get special treatment. In that example,
the request type is google.protobuf.StringValue
. Instead of showing a form for a message with a
single field named value
with type string
, the UI is simple and the "boxing" ceremony is
elided. It instead just shows a simple textbox for entering the string value.
A more interesting example of how well-known message types are treated is google.protobuf.Timestamp
,
where a date picker is shown:
The second tab lets you view the JSON representation of the request data you have defined on the first tab. You can also directly edit the JSON data -- including pasting in an entire JSON message.
The JSON representation uses the standard JSON mapping for Protocol Buffers.
When working with an RPC that has a streaming request, the JSON data will be a JSON array, where each element is a single message in the stream.
When the "Invoke" button is pressed, the request data is sent to the server and the selected RPC method is invoked. The web form will then navigate to the third tab to show the server's response.
The response tab has three sections:
- Response Headers: Any response header metadata is shown here.
- Response Data: Any response messages are shown here as are any error messages. RPC methods with a streaming response may show both message data and an error. Error messages show the gRPC status code and the server-defined message text.
- Response Trailers: Finally, any response trailer metadata is shown.
Each of these three sections is a table of data. Response messages are the most interesting, and their structure closely resembles how messages are structured on the "Request Form" tab. Fields that have nested messages will include a nested table.
The grpcui
tool can operate on a variety of sources for descriptors. The descriptors
are required, in order for grpcui
to understand the RPC schema, translate inputs
into the protobuf binary format as well as translate responses from the binary format
into text. The sections below document the supported sources and what command-line flags
are needed to use them.
Without any additional command-line flags, grpcui
will try to use server reflection.
Examples for how to set up server reflection can be found here.
To use grpcui
on servers that do not support reflection, you can use .proto
source
files.
In addition to using -proto
flags to point grpcui
at the relevant proto source file(s),
you may also need to supply -import-path
flags to tell grpcui
the folders from which
dependencies can be imported.
Just like when compiling with protoc
, you do not need to provide an import path for the
location of the standard protos included with protoc
(which contain various "well-known
types" with a package definition of google.protobuf
). These files are "known" by grpcui
as a snapshot of their descriptors is built into the grpcui
binary.
You can also use compiled protoset files with grpcui
. Protoset files contain binary
encoded google.protobuf.FileDescriptorSet
protos. To create a protoset file, invoke
protoc
with the *.proto
files that define the service:
protoc --proto_path=. \
--descriptor_set_out=myservice.protoset \
--include_imports \
my/custom/server/service.proto
The --descriptor_set_out
argument is what tells protoc
to produce a protoset,
and the --include_imports
argument is necessary for the protoset to contain
everything that grpcui
needs to process and understand the schema.