This adds the scenario where a channel is closed while the node is
offline, the node loses state and comes back online. In this case the
node should attempt to resync the channel, and the peer should resend a
channel sync message for the closed channel, such that the node can
retrieve its funds.
This extracts part of the test into a new helper method timeTravel,
which can be used to easily reset a node back to a state where channel
state is lost.
This commit introduces a new utility method
waitForChannelPendingForceClose, that is used to ensure a force closed
channel has been recognized by the UTXO nursery, and is ready to be
swept as soon as it matures.
The commit also utilizes this method to properly wait before mining
blocks in certain tests, as it makes sure that the UTXO nursery will
react properly to the new blocks.
In preparation for the added propagation delay by separating the miner
and the chain backend, we increase several timeouts throughout the test,
and extract them into constants that can easily be altered.
This commit modifies the graph topology test to
properly count channel updates and node
announcments in the event that they are batched
into a single topology update. The prior logic
made the assumption that they were always in
distinct topology updates, so this method should
be more general and robust.
In this commit, we open an additional channel between Bob and Carol to
ensure that Bob gets selected as the only routing hint. Previously Bob
would get selected, but with the recent changes, it would no longer
happen due to him not having any advertised edges.
In this commit, we fix a flake in the link node garbage collection test
by ensuring the channels have been fully closed on both sides before we
attempt to restart and ensure that they don't actually establish
connections. Without this check, it's possible that either side hasn't
yet processed all the blocks, so they'll still reconnect to each other on
start up.
This commit prevents an error that I've seen on travis,
wherein the test fails because a call to Fatal happens
after the test finishes. The root cause is that we call
Fatal in a goroutine that is reading from the subscribe
graph rpc call.
To fix this, we now pass an err chan back into the main
test context, where we can receive any errors and fail
the test if one comes through.
This commit fixes a bug in the integration test, that
reliably fails after disabling the height hint cache.
The test originally asserted that the htlc funds were
in limbo, but was reading a stale copy of the force
close information. Recently, the test was amended to
provided a valid read of the force close in
96a079873a6da2f0a93adc30945971fc07c5e610. However,
the issue was not apparent until build against the
disabled height hint cache.
The test is now correct to assert that there are no
funds in limbo, as the commitment output has been
swept, but the htlcs are still in the contract
court, so the nursery is unaware of them. We also
add another sanity check to validate that there are
no pending htlcs on the force close at that point
in time.
In this commit, we remove the sleep timeouts used within the channel
force closure integration test. This is needed because the recent changes
within the ChainNotifier require a longer timeout due to making a
database transaction on every new block to update the confirm/spend
hints of transactions. Rather than increasing the timeouts, we simply
remove them to ensure this isn't an issue down the road.
This commit changes the test testSendUpdateDisableChannel to use Dave to
listen for channel updates, as was originally intended.
Since a ChannelUpdate won't propagate in the network if the channel is
already closed (if the closing transaction has been mined), we change
the closing procedure to initiate the closing process, wait for the
ChannelUpdates, then finally mining the closing transaction. This avoids
a propagation race between the mined block and the ChannelUpdate.
This commit adds the integration test testDataLossProtection, that
ensures that when a node loses state, the channel counterparty will
force close the channel, and they both can recover their funds.
In this commit we modify the integration tests slightly, by setting the
parties that gets breached during the breach tests to --nolisten. We do
this to ensure that once the data protection logic is in place, they
nodes won't automatically connect, detect the state desync and recover
before we are able to trigger the breach.
This commit makes sure the channels that are force closed also are put
into the state "waiting close" before the commitment transaction is
confirmed, and exits this state when it confirms.
This was previously not checked, as this check was added before the
"waiting close" state was introduced.
This commit fixes a flake within the integration tests, where we would
mine a set of blocks before checking if Bob's sweep tx was in the
mempool. Usually this would pass since the blocks were generated before
the tx hit the miner's mempool, but sometimes it was mined and then we
would check the mempool.
This commit fixes this by correctly waiting immediately for Bob to sweep
his funds, as they are not time locked.
