In this commit, we increase the fwdpkg gc interval
to avoid having it conflict with switch tests that
inspect forwarding packages. The current timeout is
a little too short on travis, and sporadically fails
TestChannelLinkCleanupSpuriousResponses, which was
added recently.
This commit moves the logic handling responses to
locally-initiated payments to be asynchronous. The
reordering of operations into handleLocalDispatch
brings a serious performance burden to the switch's
main event loop. However, the at-most once semantics
of circuit map and idempotency of cleanup methods
allows concurrent operations to run in parallel.
Prior to this commit, the async_payments_benchmark
would timeout due to the forcibly serial nature of
the prior design. With this change, there is no
perceptible difference in the benchmark OMM, even
though we've added two extra db calls.
Composes the new payment status helper methods such that
we only require one db txn per state transition. This
also allows us to remove the exclusive lock from the
control tower, and enable more concurrent requests.
In this commit, we address an issue that could arise when using the
SendToRoute RPC. In this RPC, we specify the exact hops that a payment
should take. However, within the switch, we would set a constraint for
the first hop to be any hop as long as the first peer was at the end of
it. This would cause discrepancies when attempting to use the RPC as the
payment would actually go through another hop with the same peer. We fix
this by explicitly specifying the channel ID of the first hop.
Fixes#1500.
Fixes#1515.
In this commit, we modify the readHandler w/in
the mock peer to drop messages if it is unable
to find the target link. This has led to observed
race conditions related to removing a link and still
attempting to deliver messages. By removing this,
the readHandler shouldn't fail the test as a result.
This commit increases the fwdpkg garbage collection
interval to 15s, to mitigate the likelihood of it
interfering with our unit tests related to fwdpkgs.
This commit removes the concept of "circuit deletion
forgivness" from the link. This was originally
implemented due to the strict semantics of the original
DeleteCircuit implementation, which would fail if we tried
to delete unknown circuits. Forgivness is used on startup
to ignore this error in case the circuits had already been
deleted before shutting down.
Now that the circuit deletion has been relaxed, this
behavior is no longer necessary, as requests to delete
unknown (or previously deleted) circuits will be ignored.
This is necessary for future changes regarding switch
cleanup, which may attempt to cleanup already deleted
circuits.
Previously, we would only allow deletion of circuits if all circuit keys
were found in the pending map.
In this commit, we relax this to allow for deletion of any circuits
that are found pending, and ignore those that are not found. This
is a preliminary step to cleaning up duplicate forwards that get caught
by the switch. It also allows us to gracefully handle any nodes that
are still afflicted by the split mailbox issue.
Replaces the log statement in CommitCircuits so that
it prints the circuit key of the incoming channel. This way
we avoid spewing the secp curve stored in the ErrorEncrypter.
In this commit, we thread through a link's quit channel into
routeAsync, the primary helper method allowing links to send
htlcPackets through the switch. This is intended to remove
deadlocks from happening, where the link is synchronously
blocking on forwarding packets to the switch, but also
needs to shutdown.
This commit adds a test that verifies Stop does not block
if the link is concurrently forwarding incoming Adds to
the switch. This test fails prior to the commits that
thread through the link's quit channel.
This commit modifies the default BatchTicker
implementation such that it will generate a
new ticker with each call to Start(). This
allows us to create a new ticker after
releasing an old one due to the batch
being empty.
In this commit, we prevent the htlcManager from
being woken up by the batchTicker when there is no
work to be done. Profiling has shown a significant
portion of CPU time idling, since the batch ticker
endlessly demands resources. We resolve this by only
selecting on the batch ticker when we have a
non-empty batch of downstream packets from the
switch.
This commit corrects our exit hop logic to return
FailFinalExpiryTooSoon if the following check is true:
pd.Timeout-expiryGraceDelta <= heightNow
Previously we returned FailFinalIncorrectCltvExpiry, which
should only be returned if the packet was misconstructed.
This commit replaces the debug Config struct with an empty
one, so that the command line flags are hidden in production
builds.
Production help before commit:
Tor:
--tor.active
--tor.socks=
--tor.dns=
--tor.streamisolation
--tor.control=
--tor.v2
--tor.v2privatekeypath=
--tor.v3
hodl:
--hodl.exit-settle
--hodl.add-incoming
--hodl.settle-incoming
--hodl.fail-incoming
--hodl.add-outgoing
--hodl.settle-outgoing
--hodl.fail-outgoing
--hodl.commit
--hodl.bogus-settle
Help Options:
-h, --help
Production help after commit:
Tor:
--tor.active
--tor.socks=
--tor.dns=
--tor.streamisolation
--tor.control=
--tor.v2
--tor.v2privatekeypath=
--tor.v3
Help Options:
-h, --help
In this commit, we modify the InvoiceDatabase slightly to allow the link
to record what the final payment about for an invoice was. It may be the
case that the invoice actually had no specified value, or that the payer
paid more than necessary. As a result, it's important that our on-disk
records properly reflect this.
To fix this issue, the SettleInvoice method now also accepts the final
amount paid.
Fixes#856.
This commit corrects the distribution used to
schedule a link's randomized backoff for fee
updates. Currently, our algorithm biases the
lowest value in the range, with probability
equal to lower/upper, or the ratio of the lower
bound to the upper. This distribution is skewed
more heavily as lower approaches upper.
The solution is to sample a random value in the
range upper-lower, then add this to our lower
bound. The effect is a uniformly distributed
timeout in [lower, upper).
In this commit, we modify the existing logic that would attempt to read
the min CLTV information from the invoice directly. With this route, we
avoid any sort of DB index modifications, as this information is already
stored within the payment request, which is already available to the
outside callers. By modifying the InvoiceDatabase interface, we avoid
having to make the switch aware of what the "primary" chain is.
This commit fixes a bug in the
TestDecayedLogPersistentGarbageCollector unit test.
The test generates a second hash prefix, which is never
added to the log, and used to query for the final
existence check. This commit reverts the behavior so
that the same hash prefix is used throughout the test.
In this commit, we fix a lingering bug within the link when we're the
exit node for a particular payment. Before this commit, we would assert
that the invoice gives us enough of a delta based on our current routing
policy. However, if the invoice was generated with a lower delta, or
we've changed from the default routing policy, then this would case us
to fail back any payments sent to us.
We fix this by instead using the newly available final CLTV delta
information within the extracted invoice.
Fixes#1431.
In this commit, we extract the time lock policy verification logic from
the processRemoteAdds method to the HtlcSatifiesPolicy method. With this
change, we fix a lingering bug within the link: we'll no longer verify
time lock polices within the incoming link, instead we'll verify it at
forwarding time like we should. This is a bug left over from the switch
of what the CLTV delta denotes in the channel update message we made
within the spec sometime last year.
In this commit, we extend the existing HtlcSatifiesPolicy method to also
accept timelock and height information. This is required as an upcoming
commit will fix an existing bug in the forwarding logic wherein we use
the time lock policies of the incoming node rather than that of the
outgoing node.
In this commit, we fix a bug in the generateHops helper function. Before
this commit, it erroneously used the CLTV delta of the current hop,
rather than that of the prior hop when computing the payload. This was
incorrect, as when computing the timelock for the incoming hop, we need
to factor in the CTLV delta of the outgoing lock, not the incoming lock.
In this commit, we add a new test to the switch:
TestForwardingAsymmetricTimeLockPolicies. This test ensures that a link
has two channels, one of which has a greater CLTV delta than the latter,
that a payment will successfully be routed across the channels. Atm, the
test fails (including the fix to hop payload generation included in the
next commit).
Atm, due to the way that we check forwarding policies, we'll reject this
payment as we're attempting to enforce the policy of the incoming link
(cltv delta of 7), instead of that of the outgoing link (cltv delta of
6). As a result, atm, the incoming link checks if (incoming_timeout -
delta < outgoing_timeout). For the values in the test case: 112 - 7 <
106 -> 105 < 106, this check fails. The payload is proper, but the check
itself should be applied at the outgoing hop.
In this commit, we modify the removeLink method to be more asynchronous.
Before this commit, we would attempt to block until the peer exits.
However, it may be the case that at times time, then target link is
attempting to forward a batch of packets to the switch (forwardBatch).
Atm, this method doesn't pass in an external context/quit, so we can't
cancel this message/request. As a result, we'll now ensure that
`removeLink` doesn't block, so we can resume the switch's main loop as
soon as possible.
In this commit, we move the block height dependency from the links in
the switch to the switch itself. This is possible due to a recent change
on the links no longer depending on the block height to update their
commitment fees.
We'll now only have the switch be alerted of new blocks coming in and
links will retrieve the height from it atomically.
In this commit, we modify the behavior of links updating their
commitment fees. Rather than attempting to update the commitment fee for
each link every time a new block comes in, we'll use a timer with a
random interval between 10 and 60 minutes for each link to determine
when to update their corresponding commitment fee. This prevents us from
oscillating the fee rate for our various commitment transactions.
In this commit, we fix a bug in the way we handle removing items from
the interfaceIndex. Before this commit, we would delete all items items
with the target public key that of the peer that owns the link being
removed. However, this is incorrect as the peer may have other links
sill active.
In this commit, we fix this by first only deleting the link from the
peer's index, and then checking to see if the index is empty after this
deletion. Only if so do we delete the index for the peer all together.
In this commit, we modify the interfaceIndex to no longer key the second
level of the index by the ChannelLink. Instead, we'll use the chan ID as
it's a stable identifier, unlike a reference to an interface.
In this commit, we ensure that the packet queue will always exit, by
continually signalling the main goroutine until it atomically sets a
bool that indicates its has been fully shutdown. It has been observed
that at times the main goroutine will wake up (due to the signal), but
then bypass the select and actually miss the quit signal, as a result
another signal is required. We'll continue to signals in a lazy loop
until the goroutine has fully exited.
This commit removes a possible deadlock in the switch,
which can be triggered under certain failure conditions.
Previously, we would acquire the link's read lock for
the duration of HtlcSatisfiesPolicy, though we only
need to use it grab the current policy. The deadlock could
be caused in the cases where we attempt to log the failure,
which access the read-protected ShortChanID method.
In this commit, we add and enforce a min fee rate for commitment
transactions created, and also any updates we propose to the remote
party. It's important to note that this is only a temporary patch, as
nodes can dynamically raise their min fee rate whenever their mempool is
saturated.
Fixes#1330.
This commit changes the decayed log tests to create
a new temporary database for each test. Previously, all
instances referenced the same db path. Since the tests
are run in parallel, the tests would create/delete the
shared db out from under each other, causing flakes in
the unit tests.
Adds a new closure OnChannelFailure to the link config, which is called
when the link fails. This function closure should use the given
LinkFailureError to properly force close the channel, send an error to
the peer, and disconnect the peer.
This commit introduces a new error type LinkFailureError which is used
to distinguish the different kinds of errors that we can encounter
during link operation. It encapsulates the information necessary to
decide how we should handle the error.
In this commit, we fix an existing source of a panic, that could at
times lead to a deadlock. If the circuit returned from closeCircuit
didn't have an outgoing key (as it was an incomplete forward), then we
would attempt to de-ref a nil pointer. This would trigger a panic, and
the runtime would start to unwind the stack, and execute each defer in
line. A deadlock can arise here, as in the defer at the root goroutine,
we need to grab the fwdingEventMtx. However, we already have it at the
panic site.
We fix this issue by ensuring we only attempt to add the event if it's a
_settle_ and also actually has an outgoing circuit (which it should
already, just a defensive check).
This commit adds a test where we trigger a situation which would
previously make the link think it was never in sync, and potentially
create a lot of empty state updates. This would happen if we were
waiting for a revocation, while still receiving updates from the remote.
Since in this case we could not ACK the updates because of the exhausted
revocation window, our local commitchain would extend, while the remote
chain would stall. When we finally got the revocation the local
commitment height would be far larger than the remote, and FullySynced
would return false from that point on.
This commit adds a new test that makes sure we don't try to send
commitments for states where there are now new updates. Before the
recent change to FullySynced we would do this in this test scenario, as
the local an remote commitment heights would differ.
The test makes the local commitment chain extend by 1 vs the remote,
which would earlier trigger another state update, and checks taht no
such state update is made.
This commit makes the call to forwardBatch after locking
in Adds synchronous. This ensures that circuits for any Add
packets are added to the switch in the same order that they
are prescribed in the channel state. Though it is very unlikely
this case would arise, it may happen under more greater loads.
In addition, this also makes some trivial optimizations wrt. to
not spawning unnecessary goroutines if no settle/fail packets
are locked in.
In this commit, we ensure that any time we send a TempChannelFailure
that's destined for a multi-hop source sender, then we'll always package
the latest channel update along with it.
In this commit, we fix a race in the set of TestChannelLinkTrimCircuits*
tests. Before this commit, we would trim the circuits in the htlcManager
goroutine. However, this was problematic as the scheduling order of
goroutines isn't predictable. Instead, we'll now trim the circuits in
the Start method.
Additionally, we fix a series of off-by-2 bugs in the tests themselves.
