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.
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
$ sudo apt-get install virtualenv python3 libyajl-dev git
$ brew install python3 yajl
$ pip3 install 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
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.
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.
# 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 newsnaptshot.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
More detail about general usage:
- Mainnet steemd
- Taking a snapshot
- Generating actions
- Keys substitution
- Running testnet fastgen node
- Durables
- Warden
- Gatling transactions from mainnet
- Running testnet witness node(s)
- Tests
First, we set up a steemd
for the main network. This steemd
must be the following characteristics:
- The
steemd
must beappbase
version - The
chain
,webserver
anddatabase_api
plugins must be enabled - The
webserver-http-endpoint
assumed by the following examples is127.0.0.1:8090
- 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
$ 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
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 thesecret
option in the config file - Balances are created by dividing
total_port_balance
proportionally among the live STEEM and vesting, subject tomin_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
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.
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
.
The get_dev_key
program provided with steemd
derives
keys using the same algorithm as tinman keysub
.
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 beappbase
version - The testnet
blockchain
directory should be empty (tryrm -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 is127.0.0.1:9990
- 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
.
( \
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
Once the testnet has been bootstrapped, other modules can be used to facilitate deeper orchestration.
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
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.
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 -
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.
To test tinman
:
$ cd test
$ pip install .. && python -m unittest *_test.py