In this commit, we make a series of changes to ensure that we'll be able
to properly survive restarts if we crash right after we call
MarkChannelClosed. In order to ensure we can survive restarts, we'll now
long the confirmed CommitSet to disk right before we close the channel.
Upon restart, we'll read these from disk so we can pick up where we left
over.
Additionally, we also will now consult the legacy chain actions if it
turns out that the channel has been closed, but we don't have a
confCommitSet written to disk. This will only be the case for nodes that
had pending close channels before this commitment.
In this commit, we add storage to the Briefcase for reading/writing a
confirmed CommitSet. This will be used in follow up commits to ensure
that we're able to survive restarts after we mark a channel as pending
closed. Along the way, we also re-add the FetchChainActions struct as
legacy nodes will need this storage.
Since we no longer have up to date chain actions on disk, we'll use the
HTLC sets in memory which contain the necessary information we need to
in order to obtain the HTLC amounts.
In this commit, we change the behavior of the channel arb to no longer
write chain actions to disk. Instead, using the new CommitSet struct,
we'll replay our set of prior actions based on what actually got into
the chain. As a result, we no longer need to write the chain actions at
all, instead they're reconstructed at run time to determine decisions,
and before any commitments are broadcast in order to determine if we
need to go to chain at all.
In this commit, we add a new `checkLocalChainActions` method. This
method differs from the existing `checkChainActions` method in that it's
only concerned with actions we should take on chain for our local state
based on the local _and_ remote state. This change ensures that we'll
now to go to chain order to cancel an HTLC that was on the remote
party's commitment transaction, but not our own.
In this commit, we fix a lingering TOOD statement in the channel arb.
Before this commitment, we would simply wipe our our local HTLC set of
the HTLC set that was on the remote commitment transaction on force
close. This was incorrect as if our commitment transaction had an HTLC
that the remote commitment didn't, then we would fail to cancel that
back, and cause both channels to time out on chain.
In order to remedy this, we introduce a new `HtlcSetKey` struct to track
all 3 possible in-flight set of HTLCs: ours, theirs, and their pending.
We also we start to tack on additional data to all the unilateral close
messages we send to subscribers. This new data is the CommitSet, or the
set of valid commitments at channel closure time. This new information
will be used by the channel arb in an upcoming commit to ensure it will
cancel back HTLCs in the case of split commitment state.
Finally, we start to thread through an optional *CommitSet to the
advanceState method. This additional information will give the channel
arb addition information it needs to ensure it properly cancels back
HTLCs that are about to time out or may time out depending on which
commitment is played.
Within the htlcswitch pakage, we modify the `SignNextCommitment` method
to return the new set of pending HTLCs for the remote party's commitment
transaction and `ReceiveRevocation` to return the latest set of
commitment transactions on the remote party's commitment as well. This
is a preparatory change which is part of a larger change to address a
lingering TODO in the cnct.
Additionally, rather than just send of the set of HTLCs after the we
revoke, we'll also send of the set of HTLCs after the remote party
revokes, and we create a pending commitment state for it.
This commit adds persisted status bit-field to ClientSessions, that can
be used to modify behavior of their handling in the client. Currently,
only a default CSessionActive status is defined. However, the intention
is that this could later be used to signal that a session is abandoned
without needing to perform a db migration to add the field. As we move
forward with testing, this will likely be useful if a session gets
borked and we need a simple method of the client to temporarily ignore
certain sessions.
The field may be useful in signaling other types of status changes,
though this was the primary motivation that warranted the addition.
Now that the committed and acked updates are persisted across restarts,
we will use them to filter out duplicate commit heights presented by the
client.
This commit adds the full bbolt-backed client database as well as a set
of unit tests to assert that it exactly implements the same behavior as
the mock ClientDB.
In this commit, we address another issue that arose with the
introduction of the fee rate buckets. We'll use an example to explain
the problem space:
Let's say we have inputs A, B, and C within the same fee rate bucket. If
A's fee rate is bumped to a higher bucket, then it's currently possible
for the lower fee rate bucket to be swept first, which would produce an
invalid RBF transaction since we're removing an input from the original
without providing a higher fee. By the time we get to the higher fee
rate bucket, we broadcast a valid RBF transaction _only_ sweeping input
A, which would evict the transaction sweeping inputs B and C from the
mempool.
To prevent this eviction, we can simply broadcast the higher fee rate
sweep transactions first, to ensure we have valid RBF transactions.
In this commit, we introduce support for arbitrary client fee
preferences when accepting input sweep requests. This is possible with
the addition of fee rate buckets. Fee rate buckets are buckets that
contain inputs with similar fee rates within a specific range, e.g.,
1-10 sat/vbyte, 11-20 sat/vbyte, etc. Having these buckets allows us to
batch and sweep inputs from different clients with similar fee rates
within a single transaction, allowing us to save on chain fees.
With this addition, we can now get rid of the UtxoSweeper's default fee
preference. As of this commit, any clients using the it to sweep inputs
specify the same fee preference to not change their behavior. Each of
these can be fine-tuned later on given their use cases.
Previous to this commit, running `make unit-cover pkg=xx`
would ignore the selected package and run unit tests and
coverage for all packages.
After this commit, the package selected with pkg= is the
only one that is tested and coverage output generated for.
If no pkg is selected, the default is as before, all pkgs.
TestSyncManagerHistoricalSyncOnReconnect tests that the sync manager will
re-trigger a historical sync when a new peer connects after a historical
sync has completed, but we have lost all peers.
To handle the case where we have been without peers, and get a new
connection, we reset the historical scan booleans when the first active
syncer is connected to trigger another historical sync.
A ClientChanSummary will be inserted for each channel registered with
the client, which for now will just track the sweep pkscript to use. In
the future, this will be extended with additional information to enable
the client to efficiently compute which historical states need to be
backed up under a given policy.
In advance of the upcoming wtdb.ClientDB, we'll modify the behavior
of the mockdb to be more like the final bbolt backed one, and assert
that all or our tests are still passing.
This commit replaces the map-based CommittedUpdates field with a slice.
When reading from disk, these will already be sorted by bbolt, so the
client restore the updates as presented without needing to sort them
first.
Since the key in the map variant was the sequence number, we refactor
the CommittedUpdate struct to have a sequence number and an embedded
CommittedUpdateBody (which is equivalent to the old CommittedUpdate).
The database is then expected to populate the sequence number from the
key on disk.
Since the sequence number is now directly integrated in the
CommittedUpdate struct, this allow allows us to remove the now redundant
seqNum argument from CommitUpdate.
This commit renames the variables dbName to towerDBName and dbVersions
to towerDBVersions, to distinguish between the upcoming clientDBName
clientDBVersions. We also move resusable portions of the database
initialization and default endianness to its own file so that it can be
shared between both tower and client databases.
