alesito85 / tinman

Testnet management scripts

Overview

The tinman set of utilities is a set of scripts to create a testnet. A tinman testnet allows all, or some subset of, user accounts to easily be ported from the main network.

Tinman commands

This repository contains utilities to create a testnet.

tinman snapshot : Gets account data and other data from the blockchain as necessary to do offline initialization of testnet tinman txgen : Translate the output of snapshot.py to a set of actions to perform offline initialization of testnet tinman keysub : Substitute secret keys into a list of actions tinman submit : Submit the output of txgen.py to testnet node

Installation

Linux

$ sudo apt-get install virtualenv python3 libyajl-dev git

macOS

$ brew install python3 yajl
$ pip3 install virtualenv

Creating a virtualenv

In this step we create a virtualenv to isolate our project from the system-wide Python installation. The virtualenv is activated, modifying the PATH and the prompt of the current shell, by sourcing the activate script:

$ virtualenv -p $(which python3) ~/ve/tinman
$ source ~/ve/tinman/bin/activate

Dependency Notes

Tinman should work right out of the box, but on some more delicately configured machines, some users report ijson errors. Running pip install ijson or pip3 install ijson should take care of that.

The ijson requirement also uses yajl for performance improvements. But yajl is optional and if it cannot be installed, there will be a warning that can be ignored.

Using tinman

The tinman source can be checked out with git. This documentation assumes the source code lives in ~/src/tinman:

Note:tinman's default branch is develop. master is condsidered stablish.

$ mkdir -p ~/src
$ cd ~/src
$ git clone --branch master https://github.com/steemit/tinman.git
$ cd tinman
$ pip install pipenv
$ pipenv install
$ pip install .

If everything is set up correctly, you should be able to run commands such as tinman --help as follows:

# Execute inside tinman virtualenv
$ tinman --help

Note, the tinman script in ~/ve/tinman/bin/tinman may be symlinked elsewhere (for example, ln -s ~/ve/tinman/bin/tinman ~/bin/tinman) to allow tinman to run without the virtualenv being active.

Example Usage

# First, take a snapshot of all accounts on mainnet using your own local mainnet
# node on port 8090.
$ tinman snapshot -s http://127.0.0.1:8090 -o snapshot.json

Once the snapshot.json file has been created, copy txgen.conf.example to txgen.conf:

• snapshot_file - make sure this is the same name as your new snaptshot.json

# Next, create actions.
$ tinman txgen -c txgen.conf -o txgen.actions

Create a bash script, call it bootstrap.sh, make sure the --get-dev-key and --signer arguments point to the correct binaries:

# Port the actions over to your local bootstrap node on port 9990, with a secret
# set to "xyz-"
( \
  echo '["set_secret", {"secret":"xyz-"}]' ; \
  cat txgen.actions \
) | \
tinman keysub --get-dev-key /path/to/steem/programs/util/get_dev_key | \
tinman submit --realtime -t http://127.0.0.1:9990 \
  --signer /path/to/steem/programs/util/sign_transaction \
  -f fail.json \
  -t 600

After allowing this script to run, you have now bootstrapped your testnet and you can point your witnesses at this node to start seeding and signing blocks.

To check the number of accounts created on the bootstrap node:

$ curl -s --data '{"jsonrpc":"2.0", "method":"condenser_api.get_account_count", "params":[], "id":1}' http://localhost:9990

Detailed Usage

More detail about general usage:

• Mainnet steemd
• Taking a snapshot
• Generating actions
• Keys substitution
  • Deriving secret keys
  • Command-line key generator
• Running testnet fastgen node
  • Pipelining transactions to testnet
• Durables
• Warden
• Gatling transactions from mainnet
• Running testnet witness node(s)
• Tests

Mainnet steemd

First, we set up a steemd for the main network. This steemd must be the following characteristics:

• The steemd must be appbase version
• The chain, webserver and database_api plugins must be enabled
• The webserver-http-endpoint assumed by the following examples is 127.0.0.1:8090
• If a snapshot at a well-defined single point in time is desired, no seed nodes should be used, so it does not connect to the p2p network

Taking a snapshot

$ tinman snapshot -s http://127.0.0.1:8090 -o snapshot.json

As of this writing, the above command takes approximately 5 minutes, writing an approximately 2 GB JSON file with 1,000,000 lines. If you're running tinman snapshot interactively and you would like a visual progress indicator, you can install the pv program (apt-get install pv) and use it to display the output line count in real time:

$ tinman snapshot -s http://127.0.0.1:8090 | pv -l > snapshot.json

Generating actions

Now you can use tinman txgen to create a list of actions. Actions include transactions which create and fund the accounts, and wait-for-block instructions which control the rate at which transactions occur:

# As of this writing, this command takes ~10 minutes to start writing actions,
# consumes ~200MB of RAM, with all actions created in about two hours
$ tinman txgen -c txgen.conf -o tn.txlist

Some notes about tinman txgen:

• All accounts have porter as an additional authority, allowing the testnet creator to act as any account on the testnet
• The private keys for porter and other accounts are deterministically created based on the secret option in the config file
• Balances are created by dividing total_port_balance proportionally among the live STEEM and vesting, subject to min_vesting_per_account.
• Therefore, testnet balance is not equal to mainnet balance. Rather, it is proportional to mainnet balance.
• Accounts listed in txgen.conf are considered system accounts, any identically named account in the snapshot will not be ported

Keys substitution

To maintain separation of concerns in tinman tools, the tinman txgen tool does not directly generate transactions containing private keys (except the STEEM_INIT_PRIVATE_KEY WIF, 5JNHfZYKGaomSFvd4NUdQ9qMcEAC43kujbfjueTHpVapX1Kzq2n). Instead keystrings such as publickey:active-porter are outputted in place of the actual public key of the porter account.

Since transactions can contain arbitrary user-specified data, a fixed escape sequence cannot be used to delimit keystrings, since it might appear in user-specified data. Instead, a variable escape sequence (a short value that does not appear in the data) is introduced and stored in the "esc" variable.

So a program that knows the keys must substitute them before the transactions can be submitted to the network. This is the role of the key substitution tool tinman keysub. The tinman keysub tool takes as input a list of actions, generates the specified keys, and substitutes them into each action.

Deriving secret keys

By default, the private keys generated by tinman keysub have known (i.e. insecure) seed strings. However, a secret may be added to each seed string by prepending a set_secret action to tinman keysub.

Command-line key generator

The get_dev_key program provided with steemd derives keys using the same algorithm as tinman keysub.

Running testnet fastgen node

Now that the transactions have been created, let's use them to initialize a testnet. Since many blocks worth of transactions are created, tinman submit will implement the block wait using debug_node_plugin to generate blocks in the past as rapidly as possible. So we will run a special node, let's call it the "fastgen node", with the debug plugin enabled. The fastgen node is only used to initialize the network. (It is called "fastgen" because it generates blocks as fast as possible, rather than waiting 3 seconds of real time between each block.) Later, one or more normal witness nodes will connect to the fastgen node over p2p, get blocks, and begin normal block production.

The fastgen node needs the following:

• The steemd must be appbase version
• The testnet blockchain directory should be empty (try rm -Rf testnet_datadir/blockchain)
• The following plugins should be enabled: chain p2p webserver debug_node database_api network_broadcast_api debug_node_api block_api
• The webserver-http-endpoint assumed by the following examples is 127.0.0.1:9990
• It must contain functionality from PR's #1722 #1723
• It should listen for p2p, the following examples assume it is listening on 0.0.0.0:12001

On the testnet, some serializations are different from the main network, and they are not handled properly by steem_python. Therefore, tinman submit outsources signing of those transactions to the sign_transaction binary included with steemd.

Pipelining transactions to testnet

( \
  echo '["set_secret", {"secret":"xyz-"}]' ; \
  tinman txgen -c txgen.conf \
) | \
tinman keysub | \
tinman submit -t http://127.0.0.1:9990 --signer steem/programs/util/sign_transaction -f fail.json

Other Modules

Once the testnet has been bootstrapped, other modules can be used to facilitate deeper orchestration.

Durables

For consistency across testnet deployments, fixture-like object that must exist for external testing are recreated by the durables module.

Copy durables.conf.example to durables.conf, add any desired objects, and run (typically after initial bootstrap and before gatling):

( \
  echo '["set_secret", {"secret":"xyz-"}]' ; \
  tinman durables -c durables.conf \
) | \
tinman keysub | \
tinman submit -t http://127.0.0.1:9990 --signer steem/programs/util/sign_transaction -f die

Warden

Use warden to check the current condition of a given chain. It does some basic checks to make sure the chain is up and running, then returns error codes.

Returning error code zero (0) means everything looks good. Non-zero means something is amiss.

$ tinman warden -s http://127.0.0.1:8090 && echo LGTM || echo Bummer.

As an example, you can add warden to your deployment script to delay the next step until your seed node has synchronized with the initial bootstrap node.

while [[ $all_clear -ne 0 ]]
do
    tinman warden -s http://my-seed-node:8080
    all_clear=$?
    echo Waiting for warden to sound the all-clear.
    sleep 60
done

echo Ready to proceed.

Gatling transactions from mainnet

Populating the test network with transactions from the main network.

To stream from genesis:

$ tinman gatling -f 1 -o -

To stream from block 25066272 to 25066292:

$ tinman gatling -f 25066272 -t 25066292 -o -

To stream starting from block 25066272:

$ tinman gatling -f 25066272 -o -

Running testnet witness node(s)

At the end of the transactions to be submitted, tinman txgen creates witnesses init-0 through init-20 and votes for them with large amount of TESTS. The keys of these witnesses are generated by a deterministic algorithm compatible with the get_dev_key utility program, so the keys of the witnesses may be obtained as follows:

$ programs/util/get_dev_key xxx- block-init-0:21

where xxx is the "secret" string in txgen.conf.

So in order to transition block production duties away from the initial node, all that is required is to connect witness nodes with the correct block production settings. Each witness node should specify the fastgen node using the p2p-seed-node option in the config file.

Therefore we may add the witness definitions and private keys to the witness config file:

i=0 ; while [ $i -lt 21 ] ; do echo witness = '"'init-$i'"' >> testnet_datadir/config.ini ; let i=i+1 ; done
steem/programs/util/get_dev_key xxx- block-init-0:21 | cut -d '"' -f 4 | sed 's/^/private-key = /' >> testnet_datadir/config.ini

Witness duties may of course be split among multiple nodes if desired, simply put the witness and private-key definitions in the datadir of each node, and have each witness node connect to all the others with the p2p-seed-node option.

Additionally, because there is a large gap in block timestamps between the end of the initialization blocks and the beginning of normal production, the witness will need to specify --enable-stale-production and --required-participation=0 flags. As long as a sufficient number of other witness nodes are timely producing blocks, it is not necessary to use these flags once 128 blocks have been produced after the transition.

Tests

To test tinman:

$ cd test
$ pip install .. && python -m unittest *_test.py