This guide is deliberately written at a high level and with minimal details because it is assumed that VSP operators will already have a level of familiarity with Decred software and a level of sysadmin experience.
A vspd deployment consists of a single front-end server which handles web requests, and a number of remote servers which host voting wallets.
%%{init:{
"theme": "neutral",
"themeVariables": {
"fontSize": "16px",
"clusterBkg": "#E9F8FE",
"clusterBorder": "#3D5873",
"mainBkg": "#2DD8A3",
"primaryColor": "#2970FF"
}
}}%%
graph LR
subgraph Front-end Server
nginx(nginx) --> vspd(vspd)
vspd --> dcrd0(dcrd)
end
subgraph voting3 [Voting Server 3]
dcrwallet3(dcrwallet) --> dcrd3(dcrd)
end
subgraph voting2 [Voting Server 2]
dcrwallet2(dcrwallet) --> dcrd2(dcrd)
end
subgraph voting1 [Voting Server 1]
dcrwallet1(dcrwallet) --> dcrd1(dcrd)
end
vspd ----> dcrwallet1 & dcrwallet2 & dcrwallet3
web(Internet<br />Traffic) ---> nginx
Compiled binaries are not provided for vspd - VSP operators are expected to build vspd from source.
When deploying vspd to production, always build all binaries from release branches. Neither vspd nor its dependencies (dcrd/dcrwallet) should be built from master when handling mainnet tickets.
A wallet should be created to collect VSP fees. Ideally this would be a cold
wallet which is not used for any other purpose, and it should be completely
separate from the vspd infrastructure. The dcrwallet getmasterpubkey
RPC
should be used to export an extended public (xpub) key from one of the wallet
accounts. This xpub key will be provided to vspd through a CLI flag, and it will
be used to derive addresses for receiving fee payments.
A vspd deployment should have a minimum of three remote voting wallets. The servers hosting these wallets should ideally be in geographically separate locations.
Each voting server should be running an instance of dcrd and dcrwallet. The
wallet on these servers should be completely empty and not used for any purpose
other than voting tickets added by vspd.
dcrwallet should be permanently unlocked and have voting enabled
(--enablevoting
). dcrwallet is also required to have the manual tickets
option (--manualtickets
) enabled which disables dcrwallet adding tickets
arriving over the network.
This prevents a user from reusing a voting address and the VSP voting multiple
tickets with only a single fee payment.
vspd on the front-end server must be able to reach each instance of dcrwallet
over RPC.
The front-end server is where vspd will be running. A webserver (eg. nginx) should be configured with SSL and used to proxy requests from the internet to vspd. Do not run a voting wallet on your webserver.
-
Start an instance of dcrd on this server with transaction index enabled (
--txindex
). dcrd is used for fishing ticket details out of the chain, for receivingblockconnected
notifications, and for broadcasting and checking the status of fee transactions. -
Use vspadmin to write a config file containing default values. Modify the config file to set your dcrd and dcrwallet connection details, and any other required customization.
$ go run ./cmd/vspadmin writeconfig
-
Use vspadmin to initialize a vpsd database. The xpub key to be used for collecting fees must be passed in as an argument.
$ go run ./cmd/vspadmin createdatabase tpubVppjaMjp8GEW...
-
Once the database is initialized and required fields in the config file have been entered, vspd can be started for normal operation.
-
Configure nginx with SSL and set up reverse proxy to forward requests to the vspd process. nginx must also set the
X-Forwarded-For
header to make vspd aware of the IP address of clients. Client IPs are used for logging and rate limiting.server { location / { proxy_pass http://127.0.0.1:8800/; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; } }
A monitoring system with alerting should be pointed at vspd and tested/verified to be operating properly. An ideal solution would monitor the following:
-
Front-end host:
- vspd and dcrd processes are running.
- No errors in vspd or dcrd logs.
- Both dcrd and vspd are keeping up to date with new blocks.
- Web front-end is accessible from the internet.
- vspd
/admin/status
endpoint indicates no issues.
-
Voting wallet hosts:
- dcrwallet and dcrd processes are running.
- No errors in dcrwallet or dcrd logs.
- dcrwallet has voting enabled and is unlocked.
- Both dcrd and dcrwallet are keeping up to date with new blocks.
Any event logged at the [ERR]
level is worthy of immediate investigation.
Examples include configuration or deployment issues, RPC failures and database
errors.
The [WRN]
level is used to indicate events which are of interest, but do not
necessarily require investigation (eg. bad requests from clients, recoverable
errors).
The current status of the VSP is displayed in a table on the /admin
page, and the same information can be retrieved as a JSON object from
/admin/status
for automated monitoring. This endpoint requires Basic HTTP
Authentication with the username admin
and the password set in vspd
configuration. A 200 HTTP status will be returned if the VSP seems
healthy, or a 500 status will be used to indicate something is wrong.
$ curl --user admin:12345 http://localhost:8800/admin/status
{
"dcrd": {
"host": "wss://127.0.0.1:19109/ws",
"connected": true,
"bestblockerror": false,
"bestblockheight": 802572
},
"wallets": {
"wss://127.0.0.1:20111/ws": {
"connected": true,
"infoerror": false,
"daemonconnected": true,
"voteversion": 10,
"unlocked": true,
"voting": true,
"bestblockerror": false,
"bestblockheight": 802572
}
}
}
The bbolt database file used by vspd is stored in the process home directory, at
the path {homedir}/data/{network}/vspd.db
. vspd keeps a file lock on this
file, so it cannot be opened by any other processes while vspd is running.
To facilitate back-ups, vspd will periodically write a copy of the bbolt
database to the path {homedir}/data/{network}/vspd.db-backup
.
The backup file will also be written when vspd shuts down.
This file should be backed up often and regularly (probably at least hourly).
Backups should be transferred off-site, ideally to a server which is not part of
the vspd deployment.
It is also possible to generate and download a database backup on demand from the admin page of the vspd web front-end.
To recover from a total failure of a wallet or wallet server (or to add a new wallet to the vspd deployment):
- Set up a new empty wallet. Ensure voting is enabled and the wallet is unlocked.
- Update the vspd config file to include the new wallet (and remove the old wallet if necessary).
- Restart vspd and it will connect to the new wallet and automatically insert all required data to bring it up to date.
The vspd database file contains everything needed to restore a vspd deployment from scratch. Simply place the database file into the vspd data directory and start vspd as normal. vspd will use the database to insert all necessary data into all of the voting wallets.
The process for listing a new VSP on decred.org, and consequently in Decrediton, is detailed in listing.md.