A test harness for functional integration testing of the Aion kernel(s).
This README documents details about the Test Harness library itself. For a guide to getting started with executing and authoring test cases, start with the Tests README.md.
We strive to make the master
branch a ready-to-ship product at every commit, so it is always safe to build the latest version of the project. Otherwise, you can also find jars built at various commits in the releases
tab.
git clone
the project to your machine.cd node_test_harness/TestHarness
ant
You can find the build in the newly created dist
directory.
- Configuring a local java kernel node
- Starting and stopping a local node
- Syncing to a network
- Using a remote node
- Creating transactions
- Sending transactions to a local node
- Waiting for transactions to process
- Collecting results in bulk
- Sending and waiting on transactions properly
- Getting transaction receipts
- Getting block numbers
- Getting blocks transactions are sealed into
- Making your own custom events
- How to benchmark events
- Getting basic statistics
You have a pre-built kernel. Not a tar file, but its extracted contents. This should be a directory named aion
(most likely) that contains the aion.sh
script. Now instead of building from source, you want the node to launch directly from this build.
DatabaseOption databaseOption = DatabaseOption.DO_NOT_PRESERVE_DATABASE;
NodeConfigurations configurations = NodeConfigurations.alwaysUseBuiltKernel(network, "/path/to/build", databaseOption);
LocalNode node = NodeFactory.getNewLocalNodeInstance(NodeFactory.NodeType.JAVA_NODE);
node.configure(configurations);
Result result = node.initialize();
Now each time we start up our node it will run directly from this build. Since we specified not to preserve the database, each time we initialize the database will be deleted.
Caution: if there is no subsequent call to initialize between start()
and stop()
the database will remain.
If we wanted to preserve the database we simply give the other database option: DatabaseOption.PRESERVE_DATABASE
.
Once you've configured and initialized the node it's time to start it up.
Result result = node.start();
if (!result.isSuccess()) {
System.out.println("Failed to start node: " + result.getError());
}
// do interesting stuff here.
node.stop();
You should always check the result of node.start()
. Don't assume the node was started correctly. Furthermore, you should always stop the node once you're done with it. The node runs as an external process and it is not guaranteed that it will be shutdown just because your program has exited. This is a common reason why node.start()
fails: a previous node was not shutdown and now the two are attempting to use the same database or listen on the same port, etc.
If you are connecting to a real network then it is always recommended that you wait until you have synced up with that network. However, if you are running a standalone node (you have no peers) then you definitely do not want to sync (this will be explained below), but the rpc server and the miner threads are not necessarily ready to be prodded yet - this is a known "bug" and is being tracked by issue #35. Until the bug is fixed, it is recommended that you sleep for 5-10 seconds just to be safe before proceeding.
Let's assume we've got some remote node running on some server. We don't actually have to do anything special to create our RemoteNode
representation of it. The "specialness" comes into play when we try to listen to that node with the NodeListener
or using RPC, as we see below. To actually get an instance of a RemoteNode
we just do the following:
RemoteNode node = NodeFactory.getNewRemoteNodeInstance(NodeType.JAVA);
This is a generic remote node, it is not actually "linked to" any existing remote node yet. We'll see how to do that below when NodeListener
comes into play.
To send events off to a remote node via the RPC
class we don't need to do anything special. When the RPC
instance is created, just give it the correct IP and port to listen to. That's it.
To listen to the log file of a remote node you just need some way of obtaining that log file, or its output stream in general (which you can then redirect to a log file). There are future plans to deal in terms of streams but for now the stdout stream is observed via a file.
File remoteOutput = // The stdout of the remote node, which is being written to this file.
RemoteNode node = NodeFactory.getNewRemoteNodeInstance(NodeType.JAVA);
node.connect(remoteOutput);
// Do some interesting stuff...
node.disconnect();
This is the only time we actually "link" a RemoteNode
to the actual remote node it represents, and all we are doing it linking it up to the output stream of that node. When we disconnect()
we have our generic RemoteNode
back and we can use it again to connect()
to a different log file representing a different remote node.
Most interesting interactions with a node are done via sending it a RawTransaction
. You can construct a RawTransaction
as follows:
PrivateKey key = PrivateKey.random(); // The private key of the sender
BigInteger nonce = BigInteger.ZERO;
Address dest = null; // The destination/recipient address
byte[] data = // your data
long limit = 2_000_000; // The energy limit
long price = 10_000_000_000L; // The energy price
BigInteger value = BigInteger.ZERO;
TransactionResult result;
if (transactionIsForFvm) {
// A transaction that will be run on the FVM
result = Transaction.buildAndSignFvmTransaction(key, nonce, dest, data, limit, price, value);
} else {
// A transaction that will be run on the AVM
result = Transaction.buildAndSignAvmTransaction(key, nonce, dest, data, limit, price, value);
}
if (!result.isSuccess()) {
System.out.println("Failed to construct the transaction: " + result.getError());
}
RawTransaction transaction = result.getTransaction();
The test harness is programmed in a defensive style: if it knows you are making a mistake then it tells you. A transaction will fail to be built if it is supplied nonsensical data (a null
private key, a negative energy limit, etc.), so it's always worth checking the result.
Often we aren't too interesting in only sending a single transaction, we want to send a whole bunch. The test harness has a very flexible builder class that can be used to construct transactions in bulk, and this is definitely the recommended way of doing so.
Every field that can be specified in a transaction can either be reused by all transactions or can be unique for them all. Let's say we want to send 10 transactions from the same sender address but to 10 different recipient addresses, and we want to transfer a different amount of funds to each of these recipients. Here's how we can easily construct these transactions:
PrivateKey senderKey = PrivateKey.random();
BigInteger senderInitialNonce = BigInteger.ZERO;
List<Address> beneficiaries = // Produce all of the recipient addresses
List<BigInteger> amounts = // Produce all of the amounts to transfer
BulkResult<Transaction> result = new BulkRawTransactionBuilder(10)
.useSameSender(senderKey, senderInitialNonce)
.useMultipleDestinations(beneficiaries)
.useMultipleTransferValues(amounts)
.useSameTransactionData(new byte[0])
.useSameEnergyLimit(21_000)
.useSameEnergyPrice(10_000_000_000L)
.useSameTransactionType(TransactionType.FVM)
.build();
if (!result.isSuccess()) {
System.out.println("Failed to construct the bulk transactions: " + result.getError());
}
List<RawTransaction> transactions = result.getResults();
This will create 10 transactions all from the same sender address. The first transaction will have a nonce equal to senderInitialNonce
, and each subsequent transaction will have a nonce that is one larger than the previous. The Address
at index i
in beneficiaries
will be the recipient of the i
'th transaction and will be transfered the amount at index i
in amounts
. All transactions will have the same specified energy limit and price and all will be executed by the FVM.
Now that we've built our transaction or transactions, it's time to send them off to the node.
RPC rpc = new RPC("127.0.0.1", "8545");
// Sending a single transaction.
RawTransaction transaction = // The transaction built from the step above.
RpcResult<ReceiptHash> singleResult = rpc.sendTransaction(transaction);
if (!singleResult.isSuccess()) {
System.out.println("Failed to send the transaction: " + singleResult.getError());
}
// Sending transactions in bulk.
List<RawTransaction> transactions = // The transactions built from the step above.
List<RpcResult<ReceiptHash> bulkResult = rpc.sendTransactions(transactions);
if (!bulkResult.isSuccess()) {
System.out.println("Failed to send the transactions: " + bulkResult.getError());
}
This is typically not the full pattern you would use when sending a transaction. Usually you would want to listen for an event, such as waiting for the transaction to be processed. We will put that all together once we show you how to listen for a transaction to be processed.
Sending a transaction is an asynchronous event. The transaction hash that it returns is meaningless until the transaction has been processed. Here's how to wait until that happens:
LocalNode node = // A local node obtained from one of the steps above.
NodeListener listener = NodeListener.listenTo(node);
// Listen for a single transaction to be processed.
RawTransaction transaction = // The transaction built from the step above.
FutureResult<LogEventResult> future = listener.listenForTransactionToBeProcessed(transaction, 1, TimeUnit.MINUTES);
LogEventResult result = future.get();
if (!result.eventWasObserved()) {
System.out.println("Failed to observe the transaction being processed: " + result.getError());
}
// Listen for multiple transactions to be processed.
List<RawTransaction> transactions = // The transactions built from the step above.
List<FutureResult<LogEventResult>> futures = listener.listenForTransactionsToBeProcessed(transactions, 1, TimeUnit.MINUTES);
TestHarnessHelper.waitOnFutures(futures);
// This is NOT the best way to do this, below we show you how to collect the results in a single step
for (FutureResult<LogEventResult> futureResult : futures) {
LogEventResult eventResult = futureResult.get(); // This method will no longer block, we've already waited.
if (!eventResult.eventWasObserved()) {
System.out.println("Failed to observe the transaction being processed: " + eventResult.getError());
}
}
We create a new NodeListener
so that we can listen to the node's log file. Then we tell the listener to listen for the transaction to be processed (either it is sealed into a block or it is rejected) and we will timeout after 1 minute of waiting for this event. The listener returns a Future
which we can then block on when we call future.get()
to wait for its result.
In the case where we submitted multiple transactions to be listened to, we can use the TestHarnessHelper
class to wait on all of the futures to complete. This method blocks until all of the futures are finished. After this we can collect their results.
Note that when we wait on multiple events, the timeout value we provide is applied to each of the events. The timeout value should be large enough, then, to wait for all of the events to finish.
So you've called a method that has returned a list of results of some sort and you want to do two things: 1) check that all of the results were successful, 2) extract the meaningful part of the results into a list. We've got an easy way to do that for you.
List<FutureResult<LogEventResult>> futures = // The futures we obtained in the step above.
BulkResult<LogEventResult> bulkResult = TestHarnessHelper.extractResults(futures);
if (!bulkResult.isSuccess()) {
System.out.println("At least one of the results failed: " + bulkResult.getError());
}
List<LogEventResult> results = bulkResult.getResults();
The TestHarnessHelper.extractResults
method will pull all of the LogEventResult
objects into a single list, which can be obtained via the bulkResult.getResults()
call, but this method will also return an unsuccessful result if any one of the results being extracted was unsuccessful. In other words, if bulkResult.isSuccess()
is true
, then all of the results themselves were successful.
Let's put all of this together now. First we'll walk through how to send a single transaction and wait for it to be processed.
LocalNode node = // A local node obtained from one of the steps above.
RPC rpc = new RPC("127.0.0.1", "8545");
NodeListener listener = NodeListener.listenTo(node);
RawTransaction transaction = // The transaction built from the step above.
// Start listening for the transaction before we send it so we know we don't miss it.
FutureResult<LogEventResult> future = listener.listenForTransactionToBeProcessed(transaction, 1, TimeUnit.MINUTES);
// Send the transaction off.
RpcResult<ReceiptHash> sendResult = rpc.sendTransaction(transaction);
if (!sendResult.isSuccess()) {
System.out.println("Failed to send the transaction: " + sendResult.getError());
// You probably do not want to continue now.
}
// Wait for the transaction to be processed.
LogEventResult futureResult = future.get();
if (!futureResult.eventWasObserved()) {
System.out.println("Failed to observe the transaction being processed: " + futureResult.getError());
}
It is vital to start listening for the event before the event could even happen, that way we know that we will observe it. Once all of this code executes and if none of the results were unsuccessful, we can be sure that we sent the transaction and that the transaction was sealed into a block.
Let's apply the same idea now to transactions in bulk.
LocalNode node = // A local node obtained from one of the steps above.
RPC rpc = new RPC("127.0.0.1", "8545");
NodeListener listener = NodeListener.listenTo(node);
List<RawTransaction> transactions = // The transactions built from the step above.
// Start listening for the transactions before we send them so we know we don't miss them.
List<FutureResult<LogEventResult>> futures = listener.listenForTransactionsToBeProcessed(transactions, 1, TimeUnit.MINUTES);
// Send the transactions off.
List<RpcResult<ReceiptHash>> sendResults = rpc.sendTransactions(transaction);
BulkResult<ReceiptHash>> bulkResults = TestHarnessHelper.extractResults(sendResults);
if (!bulkResults.isSuccess()) {
System.out.println("Failed to send all of the transactions: " + bulkResults.getError());
// You probably do not want to continue now.
}
// Wait for the transactions to be processed.
TestHarnessHelper.waitOnFutures(futures);
BulkResult<LogEventResult> bulkEventResults = TestHarnessHelper.extractFutures(futures);
if (!bulkEventResults.isSuccess()) {
System.out.println("Failed to observe the transaction being processed: " + bulkEventResults.getError());
}
Okay, so we've sent off our transactions to the node and we've observed them being processed. But now we've got these ReceiptHash
objects hanging around, and what do we do with them? You can use these ReceiptHash
es to get their corresponding TransactionReceipt
s.
RPC rpc = new RPC("127.0.0.1", "8545");
ReceiptHash hash = // The receipt hash we obtained in the step above.
RpcResult<TransactionReceipt> receiptResult = rpc.getTransactionReceipt(hash);
if (!receiptResult.isSuccess()) {
System.out.println("Failed to get the transaction receipt: " + receiptResult.getError());
}
TransactionReceipt receipt = receiptResult.getResult();
It's just one more RPC call. Since we know the transaction has been processed, we can safely get the transaction receipt. If we are not yet sure that the transaction has been processed, then the receipt hash may not yet correspond to any transaction receipt and our attempt will be unsuccessful.
And here's how we do it in the bulk case:
RPC java = new RPC("127.0.0.1", "8545");
List<ReceiptHash> hashes = // The receipt hashes we obtained in the step above.
List<RpcResult<TransactionReceipt>> receiptResults = rpc.getTransactionReceipts(hashes);
BulkResult<TransactionReceipt> bulkResults = TestHarnessHelper.extractResults(receiptResults);
if (!bulkResults.isSuccess()) {
System.out.println("Failed to get all of the transaction receipts: " + bulkResults.getError());
}
List<TransactionReceipt> receipts = bulkResults.getResults();
Like above, if we want this to work, we have to have observed all of the transactions being processed first.
We've sent our transactions off and we've got the transaction receipts back. Our transactions have been sealed into one or more blocks on the blockchain, but we want to know which ones. And from here, we then want to get the blocks themselves.
// Getting the block number of a single transaction.
TransactionReceipt receipt = // The receipt we obtained from the step above.
BigInteger blockNumber = receipt.getBlockNumber();
// Getting the block numbers of multiple transactions.
List<TransactionReceipt> receipts = // The receipts we obtained from the step above.
List<BigInteger> blockNumbers = TestHarnessHelper.extractBlockNumbers(receipts);
We now have the block number or numbers of all the blocks that our transactions were sealed into. Let's get the Block
object itself.
RPC rpc = new RPC("127.0.0.1", "8545");
// Getting a single block.
BigInteger blockNumber = // The number we obtained from the step above.
RpcResult<Block> blockResult = rpc.getBlockByNumber(blockNumber);
if (!blockResult.isSuccess()) {
System.out.println("Failed to get the block: " + blockResult.getError());
}
Block block = blockResult.getResult();
// Getting multiple blocks.
List<BigInteger> blockNumbers = // The numbers we obtained from the step above.
List<RpcResult<Block>> blockResults = rpc.getBlocksByNumber(blockNumbers);
BulkResult<Block> bulkResult = TestHarnessHelper.extractResults(blockResults);
if (!bulkResult.isSuccess()) {
System.out.println("Failed to get the blocks: " + bulkResult.getError());
}
List<Block> blocks = bulkResult.getResults();
The NodeListener
class has some basic events it will listen for, but it also has a general method for listening for any event you want to give it. These are strictly log events: you use the NodeListener
to listen for specific strings in the output log. The log file is read line by line, so your string has to be something that is actually logged on the same line. If your string actually occurs over multiple lines then you want to make a complex event (we cover that below).
Here's how you make your own simple event and listen for it.
LocalNode node = // A local node obtained from one of the steps above.
NodeListener listener = NodeListener.listenTo(node);
IEvent event = new Event("your string to listen for");
FutureResult<LogEventResult> future = listener.listenForEvent(event, 1, TimeUnit.MINUTES);
LogEventResult result = future.get();
A simple event may not be sufficient for you. Maybe you've got the following scenario: if one part of your contract logic is triggered then you expect the following 3 separate lines to be printed to screen: "apple", "banana", "peach". But if another part is triggered, you expect to see the following 2 separate lines printed: "big", "small". However, all you actually care about at the moment is whether or not either of these two events were triggered. Let's see how it's done.
LocalNode node = // A local node obtained from one of the steps above.
NodeListener listener = NodeListener.listenTo(node);
IEvent fruits = Event.and(new Event("apple"), Event.and("banana", "peach"));
IEvent sizes = Event.and("big", "small");
IEvent event = Event.or(fruits, sizes);
FutureResult<LogEventResult> future = listener.listenForEvent(event, 1, TimeUnit.MINUTES);
LogEventResult result = future.get();
This event will be observed only if the fruits
event is observed or the sizes
event is. The fruits
event is observed only if "apple" and "banana" and "peach" are all observed. The sizes
event is observed only if "big" and "small" are both observed.
You can use these and-or operators to create arbitrarily complex events to be listened for.
The events that the NodeListener
allows you to listen for by default can be found in the subclasses of the PrepackagedLogEvents
interface if you want to handle them directly. This may be desirable if you want to do something like: listen for a transaction to be processed and listen for specific output it will generate.
A quick example of this:
PrepackagedLogEventsFactory prepackagedLogEventsFactory = new PrepackagedLogEventsFactory();
prepackagedLogEventsFactory.setKernel(JAVA_NODE); // or RUST_NODE
LocalNode node = // A local node obtained from one of the steps above.
NodeListener listener = NodeListener.listenTo(node);
Transaction transaction = // A transaction that was built from one of the above steps.
IEvent output = new Event("contract output");
IEvent transactionProcessed = prepackagedLogEventsFactory.build().getTransactionProcessedEvent(transaction);
IEvent event = Event.and(output, transactionProcessed);
FutureResult<LogEventResult> future = listener.listenForEvent(event, 1, TimeUnit.MINUTES);
LogEventResult result = future.get();
Let's take the example above, we have a complex event that we have defined as follows:
IEvent apple = new Event("apple");
IEvent banana = new Event("banana");
IEvent peach = new Event("peach");
IEvent bananaAndPeach = Event.and(banana, peach);
IEvent fruits = Event.and(apple, bananaAndPeach);
IEvent big = new Event("big");
IEvent small = new Event("small");
IEvent sizes = Event.and(big, small);
IEvent event = Event.or(fruits, sizes);
The NodeListener
has confirmed that event
was observed. But event
is a complex event constructed out of multiple simple events, and we want to know which of these simple events actually occurred? Or perhaps we want to know whether it was fruits
or sizes
that was observed.
This is all we have to do:
if (fruits.hasBeenObserved()) {
System.out.println("Observed the event: fruits");
} else {
System.out.pritnln("Observed the event: sizes");
}
System.out.println("All events that were observed: " + event.getAllObservedEvents());
We just use the hasBeenObserved()
method.
If we want to get a list of all the String
s that have been seen then we can call getAllObservedEvents()
.
Again, let's say we have the same complex event as defined in the step above. Now we want to know what lines in the log file actually caused these events to be observed. Perhaps we are doing some log monitoring and the line contains variable information but we know these static keywords will capture that line, and we can then parse it once we have it.
Here's a sample of the log file that caused our event to be observed:
Count for apple: 17.
Count for banana: 6.
Count for peach: 111.
These lines caused the fruits
event defined above to be observed. But now we want to retrieve these log lines back. Here's how:
List<String> logLines = fruits.getAllObservedLogs();
Now we can parse out the 17
, 6
, and 111
from the lines that we were after.
We have some events we want to listen for, but we're actually most interesting in how long it takes for these events to occur. Let's take the simple example of a transaction being processed. What we want to measure is the amount of time it takes from sending the transaction over RPC until it gets sealed into a block. This information is already available to you.
LocalNode node = // A local node obtained from one of the steps above.
RPC rpc = new RPC("127.0.0.1", "8545");
NodeListener listener = NodeListener.listenTo(node);
RawTransaction transaction = // A transaction built from one of the steps above.
FutureResult<LogEventResult> future = listener.listenForTransactionToBeProcessed(transaction);
RpcResult<ReceiptHash> result = rpc.sendTransaction(transaction);
if (!result.isSuccess()) {
System.out.println("Failed to send the transaction: " + result.getError());
}
LogEventResult logResult = future.get();
if (!logResult.eventWasObserved()) {
System.out.println("Failed to observe the transaction being processed: " + logResult.getError());
}
long sendTime = result.getTimeOfCall(TimeUnit.NANOSECONDS);
long observedTime = logResult.timeOfObservation(TimeUnit.NANOSECONDS);
long duration = observedTime - sendTime;
This gives us the time, in nanoseconds, at which the RPC call was made to send the transaction, and the time at which the NodeListener
observed the transaction being processed. Internally, these measurements are made using System.nanoTime()
so no information is lost when extracting these times in nanoseconds.
Let's say we've sent off a batch of transactions to be processed and we listened for them to be observed. Even more, we obtained all of the Block
s that these transactions were sealed into (see the above steps). Now let's get some interesting information out of all of this.
Above we saw how to get the duration that it took for an event to be sent over the RPC and then observed in the log file. Let's see how to do this for multiple events.
List<RpcResult<?>> rpcResults = // rpc results obtained somehow
List<FutureResult<LogEventResult>> futureResults = // futures obtained somehow
long[] sendTimes = TestHarnessHelper.extractResultTimestamps(rpcResults, TimeUnit.NANOSECONDS);
long[] observedTimes = TestHarnessHelper.extractFutureTimestamps(futureResults, TimeUnit.NANOSECONDS);
DurationStatistics durationStats = DurationStatistics.from(sendTimes, observedTimes);
durationStats.printStatistics();
Now let's assume we've sent a bunch of transactions off and we want to see how many transactions made it into each block. For example, 100 transactions were sent and they were sealed into 8 blocks, but what we want to know is the distribution.
List<TransactionReceipt> receipts = // transaction receipts obtained somehow
TransactionStatistics transactionStats = TransactionStatistics.from(receipts);
transactionStats.printStatistics();
Finally, we want some better statistics on block saturation in terms of energy. Again, let's assume we have 100 transactions that were sealed into 8 blocks. We now know how many transactions were put in each block, and that's useful, but now we want some hard numbers representing the energy limits of these blocks and how much of this limit was used by by the transactions sealed into it. In other words, we want more information about the energy distribution over these blocks.
int numberOfTransactionsSent = 100;
List<Block> blocks = // blocks obtained somehow
BlockStatistics blockStats = BlockStatistics.from(numberOfTransactionsSent, blocks);
blockStats.printStatistics();