/cosmosd

Upgrade manager daemon for Cosmos blockchains

Primary LanguageGoApache License 2.0Apache-2.0

Cosmos Upgrade Manager

CircleCI codecov

This is a tiny little shim around Cosmos SDK binaries that use the upgrade module that allows for smooth and configurable management of upgrading binaries as a live chain is upgraded, and can be used to simplify validator devops while doing upgrades or to make syncing a full node for genesis simple. The upgrade manager will monitor the stdout of the daemon to look for messages from the upgrade module indicating a pending or required upgrade and act appropriately. (With better integrations possible in the future)

Arguments

upgrader is a shim around a native binary. All arguments passed to the upgrade manager command will be passed to the current daemon binary (as a subprocess). It will return stdout and stderr of the subprocess as it's own. Because of that, it cannot accept any command line arguments, nor print anything to output (unless it dies before executing a binary).

Configuration will be passed in the followingenvironmental variables:

  • DAEMON_HOME is the location where upgrade binaries should be kept (can be $HOME/.gaiad or $HOME/.xrnd)
  • DAEMON_NAME is the name of the binary itself (eg. xrnd, gaiad)
  • DAEMON_ALLOW_DOWNLOAD_BINARIES (optional) if set to on will enable auto-downloading of new binaries (for security reasons, this is intended for fullnodes rather than validators)
  • DAEMON_RESTART_AFTER_UPGRADE (optional) if set to on it will restart a the sub-process with the same args (but new binary) after a successful upgrade. By default, the manager dies afterwards and allows the supervisor to restart it if needed. Note that this will not auto-restart the child if there was an error.

Folder Layout

$DAEMON_HOME/upgrade_manager is expected to belong completely to the upgrade manager and subprocesses constrolled by it. Under this folder, we will see the following:

- genesis
  - bin
    - $DAEMON_NAME
- upgrades
  - <name>
    - bin
      - $DAEMON_NAME
- current -> upgrades/foo, genesis, etc

Each version of the chain is stored under either genesis or upgrades/<name>, which holds bin/$DAEMON_NAME along with any other needed files (maybe the cli client? maybe some dlls?). current is a symlink to the currently active folder (so current/bin/$DAEMON_NAME is the binary)

Note: the <name> after upgrades is the URI-encoded name of the upgrade as specified in the upgrade module plan.

Please note that $DAEMON_HOME/upgrade_manager just stores the binaries and associated program code. The upgrader binary can be stored in any typical location (eg /usr/local/bin). The actual blockchain program will store it's data under $GAIA_HOME etc, which is independent of the $DAEMON_HOME. You can choose to export GAIA_HOME=$DAEMON_HOME and then end up with a configuation like the following, but this is left as a choice to the admin for best directory layout.

- .gaiad
  - config
  - data
  - upgrade_manager

Usage

Basic Usage:

  • The admin is responsible for installing the upgrade_manager and setting it as a eg. systemd service to auto-restart, along with proper environmental variables
  • The admin is responsible for installing the genesis folder manually
  • The upgrade manager will set the current link to point to genesis at first start (when no current link exists)
  • The admin is (generally) responsible for installing the upgrades/<name> folders manually
  • The upgrade manager handles switching over the binaries at the correct points, so the admin can prepare days in advance and relax at upgrade time

Note that chains that wish to support upgrades may package up a genesis upgrade manager tar file with this info, just as they prepare the genesis binary tar file. In fact, they may offer a tar file will all upgrades up to current point for easy download for those who wish to sync a fullnode from start.

The DAEMON specific code, like the tendermint config, the application db, syncing blocks, etc is done as normal. The same eg. GAIA_HOME directives and command-line flags work, just the binary name is different.

Upgradeable Binary Specification

In the basic version, the upgrade_manager will read the stdout log messages to determine when an upgrade is needed. We are considering more complex solutions via signaling of some sort, but starting with the simple design:

  • when an upgrade is needed the binary will print a line that matches this regular expression: UPGRADE "(.*)" NEEDED at height (\d+):(.*).
  • the second match in the above regular expression can be a JSON object with a binaries key as described above

The name (first regexp) will be used to select the new binary to run. If it is present, the current subprocess will be killed, current will be upgraded to the new directory, and the new binary will be launched.

Question should we just kill the upgrade manager after it does the updates? so it gets a clean restart and just runs the new binary (under current). it should be safe to restart (as a service).

Auto-Download

Generally, the system requires that the administrator place all relevant binaries on the disk before the upgrade happens. However, for people who don't need such control and want an easier setup (maybe they are syncing a non-validating fullnode and want to do little maintenance), there is another option.

If you set DAEMON_ALLOW_DOWNLOAD_BINARIES=on then when an upgrade is triggered and no local binary can be found, the upgrade_manager will attempt to download and install the binary itself. The plan stored in the upgrade module has an info field for arbitrary json. This info is expected to be outputed on the halt log message. There are two valid format to specify a download in such a message:

  1. Store an os/architecture -> binary URI map in the upgrade plan info field as JSON under the "binaries" key, eg:
{
  "binaries": {
    "linux/amd64":"https://example.com/gaia.zip?checksum=sha256:aec070645fe53ee3b3763059376134f058cc337247c978add178b6ccdfb0019f"
  }
}
  1. Store a link to a file that contains all information in the above format (eg. if you want to specify lots of binaries, changelog info, etc without filling up the blockchain).

e.g https://example.com/testnet-1001-info.json?checksum=sha256:deaaa99fda9407c4dbe1d04bd49bab0cc3c1dd76fa392cd55a9425be074af01e

This file contained in link will be retrieved by go-getter and the "binaries" field will be parsed as above.

If there is no local binary, DAEMON_ALLOW_DOWNLOAD_BINARIES=on, and we can access a canonical url for the new binary, then the upgrade_manager will download it with go-getter and unpack it into the upgrades/<name> folder to be run as if we installed it manually

Note that for this mechanism to provide strong security guarantees, all URLS should include a sha{256,512} checksum. This ensures that no false binary is run, even if someone hacks the server or hijacks the dns. go-getter will always ensure the downloaded file matches the checksum if it is provided. And also handles unpacking archives into directories (so these download links should be a zip of all data in the bin directory).

To properly create a checksum on linux, you can use the sha256sum utility. eg. sha256sum ./testdata/repo/zip_directory/autod.zip which should return 29139e1381b8177aec909fab9a75d11381cab5adf7d3af0c05ff1c9c117743a7. You can also use sha512sum if you like longer hashes, or md5sum if you like to use broken hashes. Make sure to set the hash algorithm properly in the checksum argument to the url.