In this commit, we migrate the database away from a partially migrated
state. In a prior commit, we migrated the database in order to update
the Invoice struct with three new fields: add index, settle index, paid
amt. However, it was overlooked that the OutgoingPayment struct also
embedded an Invoice within it. As a result, nodes that upgraded to the
first migration found themselves unable to start up, or call
listpayments, as the internal invoice within the OutgoignPayment hadn't
yet been updated. This would result in an OOM typically as we went to
allocate a slice with a integer that should have been small, but may
have ended up actually being a set of random bytes, so a very large
number.
In this commit, we finish the DB migration by also migrating the
internal invoice within each OutgoingPayment.
Fixes#1538.
Fixes#1546.
In this commit, we modify the graph topology notifications test to wait
until a graph update notification has been received for a closed
channel, rather than assume the next graph update will be for the closed
channel itself.
In this commit, we fix an existing bug that could at times lead to a
panic if a user manually crafts a route via SendToRoute, and that route
results in a payment error. The fix is simple: create the map even
though it won't be used in the sessions since the user is feeding the
router manual routes.
In this commit, we modify the granularity of the locking
around the filterMtx in the bitcoind chainview, such that
we only lock once per block connected or filter update.
Currently, we acquire and release the lock for every
update to the map.
We also fix a bug that would cause us to not fully remove
all previous outpoints spent by a txn when doing manual
filter, as we previously would only remove the first output
detected.
In this commit, we modify the granularity of the locking
around the filterMtx in the btcd chainview, such that we
only lock once per block connected or filter update.
Currently, we acquire and release the lock for every
update to the map.
We also fix a bug that would cause us to not fully remove
all previous outpoints spent by a txn when doing manual
filter, as we previously would only remove the first output
detected.
In this commit, we fix an existing bu gin the invoice time series
migration code. Before this commit, the migration would fail as we would
try to migrate an empty invoice. We now detect this case and skip all
empty invoices.
We also add a bit more logging on both the info and trace logging level.
In this commit, we fix an issue where we would always assume the dust
limit was Bitcoin's dust limit, rather than the active chain. This would
lead to issues when attempting to open channels on the Litecon chain.
In this commit, we add additional logic to the primary notification
dispatch loop to ensure that we'll never send the same add/settle event
to the same client twice.
Consider the case where as we're sending a client its notification
backlog, a new invoice is settled (index=9). In this case, the database
will be reflected immediately, and the event will also be queued for
notifying later. We'll send out this latest event during the backlog
clear, but then will send it again once we return back to the main loop.
To ensure that this never happens, we'll now record the last index that
we’ve sent to a client to ensure that we no longer are able to send
duplicate notification events.
In this commit, we now ensure that all modifications to the invoice DB
are properly serialized. This ensures that our time series within the
database will be properly coherent. Additionally, within SettleInvoice,
we remove an extra DB call by taking advantage of the new SettleInvoice
method which will return the invoice being settled as well.
In this commit, we re-work the existing invoiceRegistry struct to
support delivering backlog notifications to subscription clients if
needed. Rather than using 1 goroutine per-client per-event, each client
now gains a concurrent notification queue. This queue will then be used
to ensure in-order delivery of all notifications from the
invoiceEventNotifier.
The SubscribeNotifications method now takes two params: addIndex, and
settleIndex. These should be the values of the last add index and settle
index the caller knows of. If specified (not zero), then we'll look up
all the notifications that the caller has missed, and then deliver those
before sending out any new notifications. In order to do this without
losing ordering of events, we've added a new central goroutine which
will ensure that all events are properly serialized.
In this commit, we add two new methods: InvoicesAddedSince and
InvoicesSettledSince. These methods will be used by higher level
sub-systems that implement notifications to deliver any notifications
backlog based on the last add index, and last settle index that the
client knows of.
It's important to note that care has been taken to ensure that this new
API can be used in a backwards compatible manner. If a client specifies
and index of 0 for either of the methods, then no backlog will be sent.
This is due to the fact that current users of the API don't expect any
backlog notifications to be sent. Additionally, the index actually
starts at 1, instead of 0.
In this commit, we add two new indexes to the invoice database: the add
index, and the settle index. These to indexes essentially form a time
series index on top of the existing primary index bucket. Each time an
invoice is added, we'll advance the addIndex seqno, and then create a
mapping from seqNo -> invoiceNum. Each time an invoice is settled, we'll
do the same, but within the settle index.
This change is required in order to allow callers to effectively seek
into the current invoice database in order to obtain notifications for
any invoices they may have missed out on while they were disconnected.
This will allow us to implement robust streaming invoice notifications
within lnd to ensure that clients never miss an event.
