30 KiB
0.0.104 - 2021-12-17
API Updates
- A
PaymentFailed
event is now provided to indicate a payment has failed fully. This event is generated either afterChannelManager::abandon_payment
is called for a given payment, or the payment times out, and there are no further pending HTLCs for the payment. This event should be used to detect payment failure instead ofPaymentPathFailed::all_paths_failed
, unless no payment retries occur viaChannelManager::retry_payment
(#1202). - Payment secrets are now generated deterministically using material from
the new
KeysInterface::get_inbound_payment_key_material
(#1177). - A
PaymentPathSuccessful
event has been added to ease passing success info to a scorer, along with aScore::payment_path_successful
method to accept such info (#1178, #1197). Score::channel_penalty_msat
has additional arguments describing the channel's capacity and the HTLC amount being sent over the channel (#1166).- A new log level
Gossip
has been added, which is used for verbose information generated during network graph sync. Enabling themax_level_trace
feature or ignoringGossip
log entries reduces log growth during initial start up from many GiB to several MiB (#1145). - The
allow_wallclock_use
feature has been removed in favor of only using thestd
andno-std
features (#1212). NetworkGraph
can now remove channels that we haven't heard updates for in two weeks withNetworkGraph::remove_stale_channels{,with_time}
. The first is called automatically if aNetGraphMsgHandler
is passed toBackgroundProcessor::start
(#1212).InvoicePayer::pay_pubkey
was added to enable sending "keysend" payments to supported recipients, using theInvoicePayer
to handle retires (#1160).user_payment_id
has been removed fromPaymentPurpose
, andChannelManager::create_inbound_payment{,_for_hash}
(#1180).- Updated documentation for several
ChannelManager
functions to remove stale references to panics which no longer occur (#1201). - The
Score
andLockableScore
objects have moved into therouting::scoring
module instead of being in therouting
module (#1166). - The
Time
parameter toScorerWithTime
is no longer longer exposed, instead being fixed based on thestd
/no-std
feature (#1184). ChannelDetails::balance_msat
was added to fetch a channel's balance without subtracting the reserve values, lining up with on-chain claim amounts less on-chain fees (#1203).- An explicit
UserConfig::accept_inbound_channels
flag is now provided, removing the need to setmin_funding_satoshis
to > 21 million BTC (#1173). - Inbound channels that fail to see the funding transaction confirm within
2016 blocks are automatically force-closed with
ClosureReason::FundingTimedOut
(#1083). - We now accept a channel_reserve value of 0 from counterparties, as it is insecure for our counterparty but not us (#1163).
NetAddress::OnionV2
parsing was removed as version 2 onion services are no longer supported in modern Tor (#1204).- Generation and signing of anchor outputs is now supported in the
KeysInterface
, though no support for them exists in the channel itself (#1176)
Bug Fixes
- Fixed a race condition in
InvoicePayer
where paths may be retried after the retry count has been exceeded. In this case theEvent::PaymentPathFailed::all_paths_failed
field is not a reliable payment failure indicator. There was no acceptable alternative indicator,Event::PaymentFailed
as been added to provide one (#1202). - Reduced the blocks-before-timeout we expect of outgoing HTLCs before refusing to forward. This check was overly strict and resulted in refusing to forward som HTLCs to a next hop that had a lower security threshold than us (#1119).
- LDK no longer attempt to update the channel fee for outbound channels when we cannot afford the new fee. This could have caused force-closure by our channel counterparty (#1054).
- Fixed several bugs which may have prevented the reliable broadcast of our own channel announcements and updates (#1169).
- Fixed a rare bug which may have resulted in spurious route finding failures when using last-hop hints and MPP with large value payments (#1168).
KeysManager::spend_spendable_outputs
no longer adds a change output that is below the dust threshold for non-standard change scripts (#1131).- Fixed a minor memory leak when attempting to send a payment that fails due
to an error when updating the
ChannelMonitor
(#1143). - Fixed a bug where a
FeeEstimator
that returns values rounded to the next sat/vbyte may result in force-closures (#1208). - Handle MPP timeout HTLC error codes, instead of considering the recipient to have sent an invalid error, removing them from the network graph (#1148)
Serialization Compatibility
- All above new events/fields are ignored by prior clients. All above new events/fields are not present when reading objects serialized by prior versions of the library.
- Payment secrets are now generated deterministically. This reduces the memory
footprint for inbound payments, however, newly-generated inbound payments
using
ChannelManager::create_inbound_payment{,_for_hash}
will not be receivable using versions prior to 0.0.104.ChannelManager::create_inbound_payment{,_for_hash}_legacy
are provided for backwards compatibility (#1177). PaymentPurpose::InvoicePayment::user_payment_id
will be 0 when reading objects written with 0.0.104 when read by 0.0.103 and previous (#1180).
In total, this release features 51 files changed, 5356 insertions, 2238 deletions in 107 commits from 9 authors, in alphabetical order:
- Antoine Riard
- Conor Okus
- Devrandom
- Duncan Dean
- Elias Rohrer
- Jeffrey Czyz
- Ken Sedgwick
- Matt Corallo
- Valentine Wallace
0.0.103 - 2021-11-02
API Updates
- This release is almost entirely focused on a new API in the
lightning-invoice
crate - theInvoicePayer
.InvoicePayer
is a struct which takes a reference to aChannelManager
and aRouter
and retries payments as paths fail. It limits retries to a configurable number, but is not serialized to disk and may retry additional times across a serialization/load. In order to learn about failed payments, it must receiveEvent
s directly from theChannelManager
, wrapping a user-providedEventHandler
which it provides all unhandled events to (#1059). get_route
has been renamedfind_route
(#1059) and now takes aRouteParameters
struct in replacement of a number of its long list of arguments (#1134). ThePayee
in theRouteParameters
is stored in theRoute
object returned and provided in theRouteParameters
contained inEvent::PaymentPathFailed
(#1059).ChannelMonitor
s must now be persisted after calls that provide new block data, prior toMonitorEvent
s being passed back toChannelManager
for processing. If you are using aChainMonitor
this is handled for you. ThePersist
API has been updated toOption
ally take theChannelMonitorUpdate
as persistence events that result from chain data no longer have a corresponding update (#1108).routing::Score
now has apayment_path_failed
method which it can use to learn which channels often fail payments. It is automatically called byInvoicePayer
for failed payment paths (#1144).- The default
Scorer
implementation is now a type alias to a type generic across different clocks and supports serialization to persist scoring data across restarts (#1146). Event::PaymentSent
now includes the full fee which was spent across all payment paths which were fulfilled or pending when the payment was fulfilled (#1142).Event::PaymentSent
andEvent::PaymentPathFailed
now include thePaymentId
which matches thePaymentId
returned fromChannelManager::send_payment
orInvoicePayer::pay_invoice
(#1059).NetGraphMsgHandler
now takes aDeref
to theNetworkGraph
, allowing for shared references to the graph data to make serialization and references to the graph data in theInvoicePayer
'sRouter
simpler (#1149).routing::Score::channel_penalty_msat
has been updated to provide theNodeId
of both the source and destination nodes of a channel (#1133).
Bug Fixes
- Previous versions would often disconnect peers during initial graph sync due to ping timeouts while processing large numbers of gossip messages. We now delay disconnecting peers if we receive messages from them even if it takes a while to receive a pong from them. Further, we avoid sending too many gossip messages between pings to ensure we should always receive pongs in a timely manner (#1137).
- If a payment was sent, creating an outbound HTLC and sending it to our
counterparty (implying the
ChannelMonitor
was persisted on disk), but theChannelManager
was not persisted prior to shutdown/crash, noEvent::PaymentPathFailed
event was generated if the HTLC was eventually failed on chain. Events are now consistent irrespective ofChannelManager
persistence or non-persistence (#1104).
Serialization Compatibility
- All above new Events/fields are ignored by prior clients. All above new Events/fields are not present when reading objects serialized by prior versions of the library.
- Payments for which a
Route
was generated using a previous version or for which the payment was originally sent by a previous version of the library will not be retried by anInvoicePayer
.
This release was singularly focused and some contributions by third parties were delayed. In total, this release features 38 files changed, 4414 insertions, and 969 deletions in 71 commits from 2 authors, in alphabetical order:
- Jeffrey Czyz
- Matt Corallo
0.0.102 - 2021-10-18
API Updates
get_route
now takes aScore
as an argument.Score
is queried during the route-finding process, returning the absolute amounts which you are willing to pay to avoid routing over a given channel. As a default, aScorer
is provided which returns a constant amount, with a suggested default of 500 msat. This translates to a willingness to pay up to 500 msat in additional fees per hop in order to avoid additional hops (#1124).Event::PaymentPathFailed
now contains ashort_channel_id
field which may be filled in with a channel that can be "blamed" for the payment failure. Payment retries should likely avoid the given channel for some time (#1077).PublicKey
s inNetworkGraph
have been replaced with aNodeId
struct which contains only a simple[u8; 33]
, substantially improvingNetworkGraph
deserialization performance (#1107).ChainMonitor
'sHashMap
ofChannelMonitor
s is now private, exposed viaChainmonitor::get_monitor
andChainMonitor::list_monitors
instead (#1112).- When an outbound channel is closed prior to the broadcasting of its funding
transaction, but after you call
ChannelManager::funding_transaction_generated
, a new event type,Event::DiscardFunding
, is generated, informing you the transaction was not broadcasted and that you can spend the same inputs again elsewhere (#1098). ChannelManager::create_channel
now returns the temporary channel ID which may later appear inEvent::ChannelClosed
orChannelDetails
prior to the channel being funded (#1121).Event::PaymentSent
now contains the payment hash as well as the payment preimage (#1062).ReadOnlyNetworkGraph::get_addresses
now returns ownedNetAddress
rather than references. As a side-effect this method is now exposed in foreign language bindings (#1115).- The
Persist
andChannelMonitorUpdateErr
types have moved to thelightning::chain::chainmonitor
andlightning::chain
modules, respectively (#1112). ChannelManager::send_payment
now returns aPaymentId
which identifies a payment (whether MPP or not) and can be used to retry the full payment or MPP parts throughretry_payment
(#1096). Note that doing so is currently not crash safe, and you may find yourself sending twice. It is recommended that you not use theretry_payment
API until the next release.
Bug Fixes
- Due to an earlier fix for the Lightning dust inflation vulnerability tracked in CVE-2021-41591/CVE-2021-41592/CVE-2021-41593 in 0.0.100, we required counterparties to accept a dust limit slightly lower than the dust limit now required by other implementations. This appeared as, at least, latest lnd always refusing to accept channels opened by LDK clients (#1065).
- If there are multiple channels available to the same counterparty,
get_route
would only consider the channel listed last as available for sending (#1100). Persist
implementations returningChannelMonitorUpdateErr::TemporaryFailure
fromwatch_channel
previously resulted in theChannelMonitor
not being stored at all, resulting in a panic after monitor updating is complete (#1112).- If payments are pending awaiting forwarding at startup, an
Event::PendingHTLCsForwardable
event will always be provided. This ensures user code callsChannelManager::process_pending_htlc_fowards
even if it shut down while awaiting the batching timer during the previous run (#1076). - If a call to
ChannelManager::send_payment
failed due to lack of availability of funds locally, LDK would store the payment as pending forever, with no ability to retry or fail it, leaking memory (#1109).
Serialization Compatibility
- All above new Events/fields are ignored by prior clients. All above new
Events/fields, except for
Event::PaymentSent::payment_hash
are not present when reading objects serialized by prior versions of the library.
In total, this release features 32 files changed, 2248 insertions, and 1483 deletions in 51 commits from 7 authors, in alphabetical order:
- 1nF0rmed
- Duncan Dean
- Elias Rohrer
- Galder Zamarreño
- Jeffrey Czyz
- Matt Corallo
- Valentine Wallace
0.0.101 - 2021-09-23
API Updates
- Custom message types are now supported directly in the
PeerManager
, allowing you to send and receive messages of any type that is not natively understood by LDK. This requires a new type bound onPeerManager
, aCustomMessageHandler
.IgnoringMessageHandler
provides a simple default for this new bound for ignoring unknown messages (#1031, #1074). - Route graph updates as a result of failed payments are no longer provided as
MessageSendEvent::PaymentFailureNetworkUpdate
but instead included in a new field in theEvent::PaymentFailed
events. Generally, this means route graph updates are no longer handled as a part of thePeerManager
but instead through the newEventHandler
implementation forNetGraphMsgHandler
. To make this easy, a new parameter tolightning-background-processor::BackgroundProcessor::start
is added, which contains anOption
alNetGraphmsgHandler
. If provided asSome
, relevant events will be processed by theNetGraphMsgHandler
prior to normal event handling (#1043). NetworkGraph
is now, itself, thread-safe. Accordingly, most functions now take&self
instead of&mut self
and the graph data can be accessed throughNetworkGraph.read_only
(#1043).- The balances available on-chain to claim after a channel has been closed are
now exposed via
ChannelMonitor::get_claimable_balances
andChainMonitor::get_claimable_balances
. The second can be used to get information about all closed channels which still have on-chain balances associated with them. See enum variants ofln::channelmonitor::Balance
and method documentation for the above methods for more information on the types of balances exposed (#1034). - When one HTLC of a multi-path payment fails, the new field
all_paths_failed
inEvent::PaymentFailed
is set tofalse
. This implies that the payment has not failed, but only one part. Payment resolution is only indicated by anEvent::PaymentSent
event or anEvent::PaymentFailed
withall_paths_failed
set totrue
, which is also set for the last remaining part of a multi-path payment (#1053). - To better capture the context described above,
Event::PaymentFailed
has been renamed toEvent::PaymentPathFailed
(#1084). - A new event,
ChannelClosed
, is provided byChannelManager
when a channel is closed, including a reason and error message (if relevant, #997). lightning-invoice
now considers invoices with sub-millisatoshi precision to be invalid, and requires millisatoshi values during construction (thus you must callamount_milli_satoshis
instead ofamount_pico_btc
, #1057).- The
BaseSign
interface now includes two new hooks which provide additional information about commitment transaction signatures and revocation secrets provided by our counterparty, allowing additional verification (#1039). - The
BaseSign
interface now includes additional information for cooperative close transactions, making it easier for a signer to verify requests (#1064). Route
has two additional helper methods to get fees and amounts (#1063).Txid
andTransaction
objects can now be deserialized from responses when using the HTTP client in thelightning-block-sync
crate (#1037, #1061).
Bug Fixes
- Fix a panic when reading a lightning invoice with a non-recoverable signature. Further, restrict lightning invoice parsing to require payment secrets and better handle a few edge cases as required by BOLT 11 (#1057).
- Fix a panic when receiving multiple messages (such as HTLC fulfill messages)
after a call to
chain::Watch::update_channel
returnedErr(ChannelMonitorUpdateErr::TemporaryFailure)
with noChannelManager::channel_monitor_updated
call in between (#1066). - For multi-path payments,
Event::PaymentSent
is no longer generated multiple times, once for each independent part (#1053). - Multi-hop route hints in invoices are now considered in the default router
provided via
get_route
(#1040). - The time peers have to respond to pings has been increased when building with debug assertions enabled. This avoids peer disconnections on slow hosts when running in debug mode (#1051).
- The timeout for the first byte of a response for requests from the
lightning-block-sync
crate has been increased to 300 seconds to better handle the long hangs in Bitcoin Core when it syncs to disk (#1090).
Serialization Compatibility
- Due to a bug in 0.0.100,
Event
s written by 0.0.101 which are of a type not understood by 0.0.100 may lead toErr(DecodeError::InvalidValue)
or corrupt deserialized objects in 0.100. SuchEvent
s will lead to anErr(DecodeError::InvalidValue)
in versions prior to 0.0.100. The only such new event written by 0.0.101 isEvent::ChannelClosed
(#1087). - Payments that were initiated in versions prior to 0.0.101 may still
generate duplicate
PaymentSent
Event
s or may have spurious values forEvent::PaymentPathFailed::all_paths_failed
(#1053). - The return values of
ChannelMonitor::get_claimable_balances
(and, thus,ChainMonitor::get_claimable_balances
) may be spurious for channels where the spend of the funding transaction appeared on chain while running a version prior to 0.0.101.Balance
information should only be relied upon for channels that were closed while running 0.0.101+ (#1034). - Payments failed while running versions prior to 0.0.101 will never have a
Some
for thenetwork_update
field (#1043).
In total, this release features 67 files changed, 4980 insertions, 1888 deletions in 89 commits from 12 authors, in alphabetical order:
- Antoine Riard
- Devrandom
- Galder Zamarreño
- Giles Cope
- Jeffrey Czyz
- Joseph Goulden
- Matt Corallo
- Sergi Delgado Segura
- Tibo-lg
- Valentine Wallace
- abhik-99
- vss96
0.0.100 - 2021-08-17
API Updates
- The
lightning
crate can now be built in no_std mode, making it easy to target embedded hardware for rust users. Note that mutexes are replaced with no-ops for such builds (#1008, #1028). - LDK now supports sending and receiving "keysend" payments. This includes
modifications to
lightning::util::events::Event::PaymentReceived
to indicate the type of payment (#967). - A new variant,
lightning::util::events::Event::PaymentForwarded
has been added which indicates a forwarded payment has been successfully claimed and we've received a forwarding fee (#1004). lightning::chain::keysinterface::KeysInterface::get_shutdown_pubkey
has been renamed toget_shutdown_scriptpubkey
, returns a script, and is now called on channel open only iflightning::util::config::ChannelConfig::commit_upfront_shutdown_pubkey
is set (#1019).- Closing-signed negotiation is now more configurable, with an explicit
lightning::util::config::ChannelConfig::force_close_avoidance_max_fee_satoshis
field allowing you to select the maximum amount you are willing to pay to avoid a force-closure. Further, we are now less restrictive on the fee placed on the closing transaction when we are not the party paying it. To control the feerate paid on a channel at close-time, useChannelManager::close_channel_with_target_feerate
instead ofclose_channel
(#1011). lightning_background_processor::BackgroundProcessor
now stops the background thread when dropped (#1007). It is marked#[must_use]
so that Rust users will receive a compile-time warning when it is immediately dropped after construction (#1029).- Total potential funds burn on force-close due to dust outputs is now limited
to
lightning::util::config::ChannelConfig::max_dust_htlc_exposure_msat
per channel (#1009). - The interval on which
lightning::ln::peer_handler::PeerManager::timer_tick_occurred
should be called has been reduced to once every five seconds (#1035) andlightning::ln::channelmanager::ChannelManager::timer_tick_occurred
should now be called on startup in addition to once per minute (#985). - The rust-bitcoin and bech32 dependencies have been updated to their respective latest versions (0.27 and 0.8, #1012).
Bug Fixes
- Fix panic when reading invoices generated by some versions of c-lightning (#1002 and #1003).
- Fix panic when attempting to validate a signed message of incorrect length (#1010).
- Do not ignore the route hints in invoices when the invoice is over 250k sats (#986).
- Fees are automatically updated on outbound channels to ensure commitment transactions are always broadcastable (#985).
- Fixes a rare case where a
lightning::util::events::Event::SpendableOutputs
event is not generated after a counterparty commitment transaction is confirmed in a reorg when a conflicting local commitment transaction is removed in the same reorg (#1022). - Fixes a remotely-triggerable force-closure of an origin channel after an HTLC was forwarded over a next-hop channel and the next-hop channel was force-closed by our counterparty (#1025).
- Fixes a rare force-closure case when sending a payment as a channel fundee when overdrawing our remaining balance. Instead the send will fail (#998).
- Fixes a rare force-closure case when a payment was claimed prior to a peer disconnection or restart, and later failed (#977).
Serialization Compatibility
- Pending inbound keysend payments which have neither been failed nor claimed
when serialized will result in a
ChannelManager
which is not readable on pre-0.0.100 clients (#967). - Because
lightning::chain::keysinterface::KeysInterface::get_shutdown_scriptpubkey
has been updated to return a script instead of only aPublicKey
,ChannelManager
s constructed with customKeysInterface
implementations on 0.0.100 and later versions will not be readable on previous versions.ChannelManager
s created with 0.0.99 and prior versions will remain readable even after the a serialization roundtrip on 0.0.100, as long as no new channels are opened. Further, users using alightning::chain::keysinterface::KeysManager
as theirKeysInterface
will haveChannelManager
s which are readable on prior versions as well (#1019). ChannelMonitorUpdate
s created by 0.0.100 and later for channels whenlightning::util::config::ChannelConfig::commit_upfront_shutdown_pubkey
is not set may not be readable by versions prior to 0.0.100 (#1019).- HTLCs which were in the process of being claimed on-chain when a pre-0.0.100
ChannelMonitor
was serialized may generatePaymentForwarded
events with spuriousfee_earned_msat
values. This only applies to payments which were unresolved at the time of the upgrade (#1004). - 0.0.100 clients with pending
Event::PaymentForwarded
events at serialization-time will generate serializedChannelManager
objects which 0.0.99 and earlier clients cannot read. The likelihood of this can be reduced by ensuring you process all pending events immediately before serialization (as is done by thelightning-background-processor
crate, #1004).
In total, this release features 59 files changed, 5861 insertions, and 2082 deletions in 95 commits from 6 authors.
0.0.99 - 2021-07-09
API Updates
lightning_block_sync::poll::Validate
is now public, allowing you to implement thelightning_block_sync::poll::Poll
trait withoutlightning_block_sync::poll::ChainPoller
(#956).lightning::ln::peer_handler::PeerManager
no longer requires that no calls are made to referencing the sameSocketDescriptor
afterdisconnect_socket
returns. This makes the API significantly less deadlock-prone and simplifiesSocketDescriptor
implementations significantly. The relevant changes have been made tolightning_net_tokio
andPeerManager
documentation has been substantially rewritten (#957).lightning::util::message_signing
'ssign
andverify
methods now take secret and public keys by reference instead of value (#974).- Substantially more information is now exposed about channels in
ChannelDetails
. See documentation for more info (#984 and #988). - The latest best block seen is now exposed in
ChannelManager::current_best_block
andChannelMonitor::current_best_block
(#984). - Feerates charged when forwarding payments over channels is now set in
ChannelConfig::fee_base_msat
when the channel is opened. For existing channels, the value is set to the value provided inChannelManagerReadArgs::default_config::channel_options
the first time theChannelManager
is loaded in 0.0.99 (#975). - We now reject HTLCs which are received to be forwarded over private channels
unless
UserConfig::accept_forwards_to_priv_channels
is set. Note thatUserConfig
is never serialized and must be provided viaChannelManagerReadArgs::default_config
at each start (#975).
Bug Fixes
- We now forward gossip messages to peers instead of only relaying locally-generated gossip or sending gossip messages during initial sync (#948).
- Correctly send
channel_update
messages to direct peers on private channels (#949). Without this, a private node connected to an LDK node over a private channel cannot receive funds as it does not know which fees the LDK node will charge. lightning::ln::channelmanager::ChannelManager
no longer expects to be persisted spuriously after we receive achannel_update
message about any channel in the routing gossip (#972).- Asynchronous
ChannelMonitor
updates (using theChannelMonitorUpdateErr::TemporaryFailure
return variant) no longer cause spurious HTLC forwarding failures (#954). - Transaction provided via
ChannelMonitor::transactions_confirmed
afterChannelMonitor::best_block_updated
was called for a much later block now trigger all relevant actions as of the later block. Previously some transaction broadcasts or other responses required an additional block be provided viaChannelMonitor::best_block_updated
(#970). - We no longer panic in rare cases when an invoice contained last-hop route hints which were unusable (#958).
Node Compatibility
- We now accept spurious
funding_locked
messages sent prior tochannel_reestablish
messages after reconnect. This is a known, long-standing bug in lnd (#966). - We now set the
first_blocknum
andnumber_of_blocks
fields inreply_channel_range
messages to values which c-lightning versions prior to 0.10 accepted. This avoids spurious force-closes from such nodes (#961).
Serialization Compatibility
- Due to a bug discovered in 0.0.98, if a
ChannelManager
is serialized on version 0.0.98 while anEvent::PaymentSent
is pending processing, theChannelManager
will fail to deserialize both on version 0.0.98 and later versions. If you have such aChannelManager
available, a simple patch will allow it to deserialize. Please file an issue if you need assistance (#973).
0.0.98 - 2021-06-11
0.0.98 should be considered a release candidate to the first alpha release of Rust-Lightning and the broader LDK. It represents several years of work designing and fine-tuning a flexible API for integrating lightning into any application. LDK should make it easy to build a lightning node or client which meets specific requirements that other lightning node software cannot. As lightning continues to evolve, and new use-cases for lightning develop, the API of LDK will continue to change and expand. However, starting with version 0.1, objects serialized with prior versions will be readable with the latest LDK. While Rust-Lightning is approaching the 0.1 milestone, language bindings components of LDK available at https://github.com/lightningdevkit are still of varying quality. Some are also approaching an 0.1 release, while others are still much more experimental. Please note that, at 0.0.98, using Rust-Lightning on mainnet is strongly discouraged.