In this tutorial, you create a blockchain with a module that lets you write to and read data from the blockchain. This module implements create, read and update functionalities for a blog-like application. The end user will be able to submit new blog posts and show a list of blog posts on the blockchain. Furthermore will the user be able to upvote and downvote the posts.
This tutorial builds on knowlege and skills developed in the earlier tutorials in the Ignite CLI Developer Tutorials. Before you start this building your blog chain, we recommend that you complete these foundational tutorials:
The purpose of this tutorial is to guide you through the implementation of a complete feedback loop: submitting data and reading this data back from the blockchain. By completing this tutorial, you will learn about:
- Scaffolding a Cosmos SDK message
- Defining new types in protocol buffer files
- Implementing keeper methods to write data to the store
- Reading data from the store and return it as a result of a query
- Using the blockchain's CLI to broadcast transactions and query the blockchain
Note: All the functions in this chapter can be scaffolded with a single command but instead you will learn how to add each functionality individually.
This series of blog tutorials is based on a specific version of Ignite CLI, so to install Ignite CLI v0.22.2 use the following command:
curl https://get.ignite.com/[email protected]! | bash
First, create a new blockchain.
Open a terminal and navigate to a directory where you have permissions to create files. To create your Cosmos SDK blockchain, run this command:
ignite scaffold chain blog
The blog
directory is created with the default directory structure.
So far, you have learned how to modify proto files to define a new API endpoint and modify a keeper query function to return static data back to the user. Of course, a keeper can do more than return a string of data. Its purpose is to manage access to the state of the blockchain.
You can think of the state as being a collection of key-value stores. Each module is responsible for its own store. Changes to the store are triggered by transactions that are signed and broadcasted by users. Each transaction contains Cosmos SDK messages (not to be confused with proto message
). When a transaction is processed, each message gets routed to its module. A module has message handlers that process messages. Processing a message can trigger changes in the state.
A Cosmos SDK message contains information that can trigger changes in the state of a blockchain.
Now, you are ready to implement these Cosmos SDK messages to achieve the desired functionality for your blog chain:
createPost
Allow users to create a post with some title and body.updatePost
Allow post writers to update the body of their posts with a new body.voteOnPost
Allow useres to upvote and downvote on posts.
First, change into the blog
directory:
cd blog
Use the ignite scaffold message
command to scaffold new messages for your module.
ignite scaffold message createPost title body
The ignite scaffold message
command accepts the message name (createPost
) as the first argument and a list of fields (title
and body
) as arguments.
The message
command has created and modified several files:
modify proto/blog/tx.proto
modify x/blog/client/cli/tx.go
create x/blog/client/cli/tx_create_post.go
modify x/blog/handler.go
create x/blog/keeper/msg_server_create_post.go
modify x/blog/module_simulation.go
create x/blog/simulation/create_post.go
modify x/blog/types/codec.go
create x/blog/types/message_create_post.go
create x/blog/types/message_create_post_test.go
🎉 Created a message `createPost`.
To create the updatePost
message for the blog module use:
ignite scaffold message updatePost index body
where:
- updatePost is the message name
- index is the post which the user wants to change
- body is the new text of the post
This ignite scaffold message
command modifies and creates the same set of files as the previous scaffold.
To create the voteOnPost
message for the blog module use:
ignite scaffold message voteOnPost index upvotes downvotes
where:
- voteOnPost is the message name
- index is the post on which the user wants to vote
- upvotes is the amount of upvotes the post gets form the user
- downvotes is the amount of downvotes the post gets form the user
This ignite scaffold message
command modifies and creates the same set of files as the previous scaffolds.
As always, start with a proto file. Inside the proto/blog/tx.proto
file, the MsgCreatePost
message has been created. Edit the file to add the line that defines the id
for message MsgCreatePostResponse
:
message MsgCreatePost {
string creator = 1;
string title = 2;
string body = 3;
}
message MsgCreatePostResponse {
uint64 id = 1;
}
Furthermore the same sets have been created for the updatePost
and voteOnPost
messages.
Edit the file to add the line that defines the upvotes
and downvotes
for message MsgVoteOnPostResponse
:
message MsgVoteOnPost {
string creator = 1;
string index = 2;
uint64 upvotes = 3;
uint64 downvotes = 4;
}
message MsgVoteOnPostResponse {
uint64 upvotes = 1;
uint64 downvotes = 2;
}
In the same file proto/blog/tx.proto
review the Cosmos SDK message type with proto message
. The MsgCreatePost
has three fields: creator, title, and body. Since the purpose of the MsgCreatePost
message is to create new posts in the store, the only thing the message needs to return is an ID of a created post. The CreatePost
rpc was already added to the Msg
service. The same goes for the UpdatePost
and VoteOnPost
rpc:
service Msg {
rpc CreatePost(MsgCreatePost) returns (MsgCreatePostResponse);
rpc UpdatePost(MsgUpdatePost) returns (MsgUpdatePostResponse);
rpc VoteOnPost(MsgVoteOnPost) returns (MsgVoteOnPostResponse);
}
Next, look at the x/blog/handler.go
file. Ignite CLI has added three case
statements to the switch
statement inside the NewHandler
function. This switch statement is responsible for routing messages and calling specific keeper methods based on the type of the message:
func NewHandler(k keeper.Keeper) sdk.Handler {
//...
return func(ctx sdk.Context, msg sdk.Msg) (*sdk.Result, error) {
//...
switch msg := msg.(type) {
case *types.MsgCreatePost:
res, err := msgServer.CreatePost(sdk.WrapSDKContext(ctx), msg)
return sdk.WrapServiceResult(ctx, res, err)
case *types.MsgUpdatePost:
res, err := msgServer.UpdatePost(sdk.WrapSDKContext(ctx), msg)
return sdk.WrapServiceResult(ctx, res, err)
case *types.MsgVoteOnPost:
res, err := msgServer.VoteOnPost(sdk.WrapSDKContext(ctx), msg)
return sdk.WrapServiceResult(ctx, res, err)
//...
}
}
}
The case *types.MsgCreatePost
statement handles messages of type MsgCreatePost
, calls the CreatePost
method, and returns back the response.
The case *types.MsgUpdatePost
statement handles messages of type MsgUpdatePost
, calls the UpdatePost
method, and returns back the response. The case *types.MsgVoteOnPost
statement handles messages of type MsgVoteOnPost
, calls the VoteOnPost
method, and returns back the response.
Every module has a handler function like this to process messages and call keeper methods.
In the newly scaffolded x/blog/keeper/msg_server_create_post.go
file, you can see a placeholder implementation of the CreatePost
function. Right now it does nothing and returns an empty response. For your blog chain, you want the contents of the message (title and body) to be written to the state as a new post.
You need to do two things:
- Create a variable of type
Post
with title and body from the message. Add the initial 0 upvotes and downvotes to the post. - Append this
Post
to the store
func (k msgServer) CreatePost(goCtx context.Context, msg *types.MsgCreatePost) (*types.MsgCreatePostResponse, error) {
// Get the context
ctx := sdk.UnwrapSDKContext(goCtx)
// Create variable of type Post
var post = types.Post{
Creator: msg.Creator,
Title: msg.Title,
Body: msg.Body,
Upvotes: 0,
Downvotes: 0,
}
// Add a post to the store and get back the ID
id := k.AppendPost(ctx, post)
// Return the ID of the post
return &types.MsgCreatePostResponse{Id: id}, nil
}
Define the Post
type and keeper methods.
When you define the Post
type in a proto file, Ignite CLI (with the help of protoc
) takes care of generating the required Go files.
Create the proto/blog/post.proto
file and define the Post
message:
syntax = "proto3";
package blog.blog;
option go_package = "blog/x/blog/types";
message Post {
string creator = 1;
uint64 id = 2;
string title = 3;
string body = 4;
uint64 upvotes = 5;
uint64 downvotes = 6;
}
The contents of the post.proto
file are standard. The file defines:
- A package name
blog.blog
that is used to identify messages - The Go package
go_package = "blog/x/blog/types"
where new files are generated - The message
message Post
Continue developing your blog chain.
The next step is to define the AppendPost
keeper method.
Create the x/blog/keeper/post.go
file and start thinking about the logic of the function and what you want to call the prefixes. The file will be empty for now.
-
To implement
AppendPost
you must first understand how the key store works. You can think of a store as a key-value database where keys are lexicographically ordered. You can loop through keys and useGet
andSet
to retrieve and set values based on keys. To distinguish between different types of data that a module can keep in its store, you can use prefixes likeproduct-
orpost-
. -
To keep a list of posts in what is essentially a key-value store, you need to keep track of the index of the posts you insert. Since both post values and post count (index) values are kept in the store, you can use different prefixes:
Post-value-
andPost-count-
.
Then, add these prefixes to the x/blog/types/keys.go
file in the const
and add a comment that describes the keys:
const (
//...
// Keep track of the index of posts
PostKey = "Post-value-"
PostCountKey = "Post-count-"
)
Your blog is now updated to take these actions when a Post
message is sent to the AppendPost
function:
- Get the number of posts in the store (count)
- Add a post by using the count as an ID
- Increment the count
- Return the count
Now, after the import
section in the x/blog/keeper/post.go
file, draft the AppendPost
function. You can add these comments to help you visualize what you do next:
// func (k Keeper) AppendPost() uint64 {
// count := k.GetPostCount()
// store.Set()
// k.SetPostCount()
// return count
// }
First, implement GetPostCount
:
func (k Keeper) GetPostCount(ctx sdk.Context) uint64 {
// Get the store using storeKey (which is "blog") and PostCountKey (which is "Post-count-")
store := prefix.NewStore(ctx.KVStore(k.storeKey), []byte(types.PostCountKey))
// Convert the PostCountKey to bytes
byteKey := []byte(types.PostCountKey)
// Get the value of the count
bz := store.Get(byteKey)
// Return zero if the count value is not found (for example, it's the first post)
if bz == nil {
return 0
}
// Convert the count into a uint64
return binary.BigEndian.Uint64(bz)
}
Now that GetPostCount
returns the correct number of posts in the store, implement SetPostCount
:
func (k Keeper) SetPostCount(ctx sdk.Context, count uint64) {
// Get the store using storeKey (which is "blog") and PostCountKey (which is "Post-count-")
store := prefix.NewStore(ctx.KVStore(k.storeKey), []byte(types.PostCountKey))
// Convert the PostCountKey to bytes
byteKey := []byte(types.PostCountKey)
// Convert count from uint64 to string and get bytes
bz := make([]byte, 8)
binary.BigEndian.PutUint64(bz, count)
// Set the value of Post-count- to count
store.Set(byteKey, bz)
}
Now that you have implemented functions for getting the number of posts and setting the post count, at the top of the same x/blog/keeper/post.go
file, implement the logic behind the AppendPost
function:
package keeper
import (
"encoding/binary"
"github.com/cosmos/cosmos-sdk/store/prefix"
sdk "github.com/cosmos/cosmos-sdk/types"
"blog/x/blog/types"
)
func (k Keeper) AppendPost(ctx sdk.Context, post types.Post) uint64 {
// Get the current number of posts in the store
count := k.GetPostCount(ctx)
// Assign an ID to the post based on the number of posts in the store
post.Id = count
// Get the store
store := prefix.NewStore(ctx.KVStore(k.storeKey), []byte(types.PostKey))
// Convert the post ID into bytes
byteKey := make([]byte, 8)
binary.BigEndian.PutUint64(byteKey, post.Id)
// Marshal the post into bytes
appendedValue := k.cdc.MustMarshal(&post)
// Insert the post bytes using post ID as a key
store.Set(byteKey, appendedValue)
// Update the post count
k.SetPostCount(ctx, count+1)
return count
}
🎉 Congratulations. By following these steps, you have implemented all of the code required to create new posts and store them on-chain. Now, when a transaction that contains a message of type MsgCreatePost
is broadcast, the message is routed to your blog module.
x/blog/handler.go
callsk.CreatePost
which in turn callsAppendPost
AppendPost
gets the number of posts from the store, adds a post using the count as an ID, increments the count, and returns the ID
Use the ignite chain build
command to compile your newly implemented keeper.
´´´bash ignite chain build ´´´
In the newly scaffolded x/blog/keeper/msg_server_update_post.go
file, you can see a placeholder implementation of the UpdatePost
function. Right now it does nothing and returns an empty response. For your blog chain, you want the contents of the message (body) to be written to the correct post.
You need to do three things:
- Check if the index already exists.
- Check if the sender of the message is the owner of the post.
- Update the new body on the post
func (k msgServer) UpdatePost(goCtx context.Context, msg *types.MsgUpdatePost) (*types.MsgUpdatePostResponse, error) {
//get the context
ctx := sdk.UnwrapSDKContext(goCtx)
// Check if the value exists
post, isFound := k.GetPost(
ctx,
msg.Index,
)
if !isFound {
return nil, sdkerrors.Wrap(sdkerrors.ErrKeyNotFound, "index not set")
}
// Check if the the msg creator is the same as the current owner
if msg.Creator != post.Creator {
return nil, sdkerrors.Wrap(sdkerrors.ErrUnauthorized, "incorrect owner")
}
//update the body
post.Body = msg.Body
//store the new post
k.SetPost(ctx, post)
return &types.MsgUpdatePostResponse{}, nil
}
Add sdkerrors "github.com/cosmos/cosmos-sdk/types/errors"
to the imports, which provides further context and meaning to a failed execution:
import (
//...
sdkerrors "github.com/cosmos/cosmos-sdk/types/errors"
)
First, implement GetPost
in the x/blog/keeper/post.go
file:
// GetPost returns a post from its index
func (k Keeper) GetPost(
ctx sdk.Context,
index string,
) (post types.Post, found bool) {
store := prefix.NewStore(ctx.KVStore(k.storeKey), []byte(types.PostKey))
// Convert the string index to uint64 and then to byte
n, err := strconv.ParseUint(index, 10, 64)
if err == nil {
fmt.Printf("%d of type %T", n, n)
}
byteKey := make([]byte, 8)
binary.BigEndian.PutUint64(byteKey, n)
// Get the value of the count
postBytes := store.Get(byteKey)
//retrun an empty post if the index does not exists
if postBytes == nil {
return post, false
}
//Unmarshal the bytes of the post
k.cdc.MustUnmarshal(postBytes, &post)
return post, true
}
Add strconv
and fmt
to the imports at the top of the file, which handles the string conversion and enables printing in the terminal:
import (
//...
"fmt"
"strconv"
)
Now that GetPost
returns the correct post in the store at the given index, implement SetPost
in the same x/blog/keeper/post.go
file:
// SetPost updates a post at its index
func (k Keeper) SetPost(ctx sdk.Context, post types.Post) {
// Get the store
store := prefix.NewStore(ctx.KVStore(k.storeKey), []byte(types.PostKey))
// Convert the post ID into bytes
byteKey := make([]byte, 8)
binary.BigEndian.PutUint64(byteKey, post.Id)
// Marshal the post into bytes
postBytes := k.cdc.MustMarshal(&post)
// Insert the post bytes using post ID as a key
store.Set(byteKey, postBytes)
}
🎉 Congratulations. By following these steps, you have implemented all of the code required to update posts and store them on-chain. Now, when a transaction that contains a message of type MsgUpdatePost
is broadcast, the message is routed to your blog module.
x/blog/handler.go
callsk.UpdatePost
which in turn callsGetPost
andSetPost
.k.UpdatePost
checks if the post exists and if the caller is allowed to change the body. It then saves the new body on chain.
Use the ignite chain build
command to compile your newly implemented keeper.
´´´bash ignite chain build ´´´
In the newly scaffolded x/blog/keeper/msg_server_vote_on_post.go
file, you can see a placeholder implementation of the VoteOnPost
function. Right now it does nothing and returns an empty response. For your blog chain, you want the contents of the message (upvotes, downvotes) to be added to the upvotes and downvotes of the correct post.
You need to do two things:
- Check if the index already exists.
- Add the new upvotes and downvotes to the post
package keeper
import (
"context"
"fmt"
"strconv"
"blog/x/blog/types"
sdk "github.com/cosmos/cosmos-sdk/types"
sdkerrors "github.com/cosmos/cosmos-sdk/types/errors"
)
func (k msgServer) VoteOnPost(goCtx context.Context, msg *types.MsgVoteOnPost) (*types.MsgVoteOnPostResponse, error) {
//get the context
ctx := sdk.UnwrapSDKContext(goCtx)
// Check if the value exists
post, isFound := k.GetPost(
ctx,
msg.Index,
)
if !isFound {
return nil, sdkerrors.Wrap(sdkerrors.ErrKeyNotFound, "index not set")
}
//Convert upvotes and downvotes from string to unit64
upvotes, err := strconv.ParseUint(msg.Upvotes, 10, 64)
if err == nil {
fmt.Printf("%d of type %T", upvotes, upvotes)
}
downvotes, err := strconv.ParseUint(msg.Downvotes, 10, 64)
if err == nil {
fmt.Printf("%d of type %T", downvotes, downvotes)
}
//add the new upvotes and downvotes on top of the old ones
post.Upvotes += upvotes
post.Downvotes += downvotes
//update the new post
k.SetPost(ctx, post)
return &types.MsgVoteOnPostResponse{Upvotes: post.Upvotes, Downvotes: post.Downvotes}, nil
}
🎉 Congratulations. By following these steps, you have implemented all of the code required to vote on posts. Now, when a transaction that contains a message of type MsgVoteOnPost
is broadcast, the message is routed to your blog module. You already implemented GetPost
and SetPost
in the previous steps.
x/blog/handler.go
callsk.VoteOnPost
which in turn callsGetPost
andSetPost
.k.VoteOnPost
checks if the post exists and then adds the votes on chain.
Use the ignite chain build
command to compile your newly implemented keeper.
´´´bash ignite chain build ´´´
Now that you have added the functionality to create posts and broadcast them to our chain, you can add querying.
To display posts, use the ignite query message
command to scaffold a new query for your module.
ignite scaffold query posts --response title,body,upvotes,downvotes
The ignite scaffold query
command accepts the query name (posts
) as the first argument and a list of response fields (title
, body
, upvotes
and downvotes
) as arguments.
The query
command has created and modified several files:
modify proto/blog/query.proto
modify x/blog/client/cli/query.go
create x/blog/client/cli/query_posts.go
create x/blog/keeper/grpc_query_posts.go
🎉 Created a query `posts`.
Two components are responsible for querying data:
- An rpc inside
service Query
in a proto file that defines data types and specifies the HTTP API endpoint - A keeper method that performs the querying from the key-value store
First, review the services and messages in proto/blog/query.proto
. The Posts
rpc accepts an empty request and returns an object with four fields: title, body, upvotes and downvotes. Now you can make changes so it can return a list of posts. The list of posts can be long, so add pagination. When pagination is added, the request and response include a page number so you can request a particular page when you know what page has been returned.
To define the types in proto files, make the following updates in proto/blog/query.proto
:
- Add the
import
:
import "blog/post.proto";
- Add pagination to the post request:
message QueryPostsRequest {
// Adding pagination to request
cosmos.base.query.v1beta1.PageRequest pagination = 1;
}
- Add pagination to the post response:
message QueryPostsResponse {
// Returning a list of posts
repeated Post Post = 1;
// Adding pagination to response
cosmos.base.query.v1beta1.PageResponse pagination = 2;
}
To implement post querying logic in the x/blog/keeper/grpc_query_posts.go
file, delete the contents of that file and replace it with:
package keeper
import (
"context"
"github.com/cosmos/cosmos-sdk/store/prefix"
sdk "github.com/cosmos/cosmos-sdk/types"
"github.com/cosmos/cosmos-sdk/types/query"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/status"
"blog/x/blog/types"
)
func (k Keeper) Posts(c context.Context, req *types.QueryPostsRequest) (*types.QueryPostsResponse, error) {
// Throw an error if request is nil
if req == nil {
return nil, status.Error(codes.InvalidArgument, "invalid request")
}
// Define a variable that will store a list of posts
var posts []*types.Post
// Get context with the information about the environment
ctx := sdk.UnwrapSDKContext(c)
// Get the key-value module store using the store key (in our case store key is "chain")
store := ctx.KVStore(k.storeKey)
// Get the part of the store that keeps posts (using post key, which is "Post-value-")
postStore := prefix.NewStore(store, []byte(types.PostKey))
// Paginate the posts store based on PageRequest
pageRes, err := query.Paginate(postStore, req.Pagination, func(key []byte, value []byte) error {
var post types.Post
if err := k.cdc.Unmarshal(value, &post); err != nil {
return err
}
posts = append(posts, &post)
return nil
})
// Throw an error if pagination failed
if err != nil {
return nil, status.Error(codes.Internal, err.Error())
}
// Return a struct containing a list of posts and pagination info
return &types.QueryPostsResponse{Post: posts, Pagination: pageRes}, nil
}
In the x/blog/module.go
file:
- Add
"context"
to the imports, don't save the file yet.
import (
"context"
// ... other imports
)
- Update the
RegisterGRPCGatewayRoutes
function to register the query handler client:
// RegisterGRPCGatewayRoutes registers the gRPC Gateway routes for the module.
func (AppModuleBasic) RegisterGRPCGatewayRoutes(clientCtx client.Context, mux *runtime.ServeMux) {
types.RegisterQueryHandlerClient(context.Background(), mux, types.NewQueryClient(clientCtx))
}
- Now that you've modified the file with the two updates, now it's safe to save the file.
Now that you have implemented logic for creating and querying posts, you can interact with your blog chain using the command line. The blog chain binary is blogd
.
First, start the chain on your development machine by running the following command in the blog
directory:
ignite chain serve
To create a post at the command line:
blogd tx blog create-post foo bar --from alice
where:
- blog is the module name
- create-post is the message name
- foo is the title
- bar is the body
- alice is the account that creates the post
The transaction is output to the terminal. You are prompted to confirm the transaction:
{"body":{"messages":[{"@type":"/blog.blog.MsgCreatePost","creator":"cosmos1ctxp3pfdtr3sw9udz2ptuh59ce9z0eaa2zvv6w","title":"foo","body":"bar"}],"memo":"","timeout_height":"0","extension_options":[],"non_critical_extension_options":[]},"auth_info":{"signer_infos":[],"fee":{"amount":[],"gas_limit":"200000","payer":"","granter":""}},"signatures":[]}
confirm transaction before signing and broadcasting [y/N]: y
Type y
to sign and broadcast the transaction.
🎉Congratulations, you used the blogd
binary CLI to create a blog post.
To query the list of all on-chain posts:
blogd q blog posts
The result:
Post:
- body: bar
creator: cosmos1a8chns8ay6vqfrgnq54cg7uzu8rlu2hk2uv06e
downvotes: "0"
id: "0"
title: foo
upvotes: "0"
pagination:
next_key: null
total: "1"
To update a post at the command line use:
blogd tx blog update-post 0 newbody --from alice
where:
- blog is the module name
- update-post is the message name
- 0 is the index of the post that gets the body changed
- newbody is the updated body
- alice is the account that updates the post. If bob tries to update a post from alice an error occurs.
The transaction is output to the terminal. You are prompted to confirm the transaction:
{"body":{"messages":[{"@type":"/blog.blog.MsgUpdatePost","creator":"cosmos1a8chns8ay6vqfrgnq54cg7uzu8rlu2hk2uv06e","index":"0","body":"newbody"}],"memo":"","timeout_height":"0","extension_options":[],"non_critical_extension_options":[]},"auth_info":{"signer_infos":[],"fee":{"amount":[],"gas_limit":"200000","payer":"","granter":""}},"signatures":[]}
confirm transaction before signing and broadcasting [y/N]:
Type y
to sign and broadcast the transaction.
Query the list of all on-chain posts to see the updated body:
blogd q blog posts
The result:
Post:
- body: newbody
creator: cosmos1a8chns8ay6vqfrgnq54cg7uzu8rlu2hk2uv06e
downvotes: "0"
id: "0"
title: foo
upvotes: "0"
pagination:
next_key: null
total: "1"
🎉Congratulations, you just used the blogd
binary CLI to update a blog post and see the result with a query.
To vote on a post at the command line use:
blogd tx blog vote-on-post 0 1 2 --from bob
where:
- blog is the module name
- vote-on-post is the message name
- 0 is the index of the post that gets the body changed
- 1 is the number of upvotes
- 2 is the number of downvote
- bob is the account that votes on the post
The transaction is output to the terminal. You are prompted to confirm the transaction:
{"body":{"messages":[{"@type":"/blog.blog.MsgVoteOnPost","creator":"cosmos1ww26u0md05sux2gvtznkurrw4ywrc0r06sqscq","index":"0","upvotes":"1","downvotes":"2"}],"memo":"","timeout_height":"0","extension_options":[],"non_critical_extension_options":[]},"auth_info":{"signer_infos":[],"fee":{"amount":[],"gas_limit":"200000","payer":"","granter":""}},"signatures":[]}
confirm transaction before signing and broadcasting [y/N]:
Type y
to sign and broadcast the transaction.
Query the list of all on-chain posts to see the updated votes:
blogd q blog posts
The result:
Post:
- body: newbody
creator: cosmos1a8chns8ay6vqfrgnq54cg7uzu8rlu2hk2uv06e
downvotes: "2"
id: "0"
title: foo
upvotes: "1"
pagination:
next_key: null
total: "1"
🎉Congratulations, you just used the blogd
binary CLI to vote on a blog post and see the result with a query.
🎉🎉🎉Congratulations🎉🎉🎉. You have built a blog blockchain!
You have successfully completed these steps:
- Write blog posts to your chain
- Update and vote on blog posts
- Read from blog posts
- Scaffold a Cosmos SDK message
- Define new types in protocol buffer files
- Write keeper methods to write data to the store
- Register query handlers
- Read data from the store and return it as a result a query
- Use the CLI to broadcast transactions