* Improve logging for initial peer connections
* Decrease the initial peer crawl timeout to make tests more reliable
Co-authored-by: Conrado Gouvea <conrado@zfnd.org>
* Implement initial service mocking helpers
Adds a [`MockService`] type, which can be configured and built for usage
in unit tests or proptests. The mocked service can then be used to
intercept requests and respond indivdiually to them.
* Use `MockService in the `mempool::Crawler` test
Refactor it to remove the helper mock function, and use the new
`MockService` helper type.
* Use `MockService` in `CandidateSet` test vectors
Refactor to remove the manual mocking of the peer set service.
* Panic if a response is not sent by `MockService`
Change the current semantics to require all `MockService` usages to
respond to every intercepted request.
A `must_use` attribute was added to the `ResponseSender` so that the
compiler can warn when this doesn't happen.
* Allow generic error types in `MockService`
Replace the hard-coded `BoxError` as the `Service`'s error type with a
generic type parameter. This allows mocking services in locations that
require specific error types.
* Add a `ResponseSender::request` getter
Allow inspecting the request again before responding, and using
information from the request in the response.
Co-authored-by: Conrado Gouvea <conrado@zfnd.org>
* Rename ChainTipReceiver to CurrentChainTip
`fastmod ChainTipReceiver CurrentChainTip zebra*`
* Update chain tip documentation and variable names
* Basic chain tip change implementation, without resets
Also includes the following name changes:
```
fastmod CurrentChainTip LatestChainTip zebra*
fastmod chain_tip_receiver latest_chain_tip zebra*
```
* Clarify the difference between `LatestChainTip` and `ChainTipChange`
* Rename BestTipHeight so it can be generalised to ChainTipSender
`fastmod BestTipHeight ChainTipSender zebra*`
For senders:
`fastmod best_tip_height chain_tip_sender zebra*`
For receivers:
`fastmod best_tip_height chain_tip_receiver zebra*`
* Rename best_tip_height module to chain_tip
* Wrap the chain tip watch channel in a ChainTipReceiver type
* Create a ChainTip trait to avoid tricky crate dependencies
And add convenience impls for optional and empty chain tips.
* Use the ChainTip trait in zebra-network
* Replace `Option<ChainTip>` with `NoChainTip`
Co-authored-by: Janito Vaqueiro Ferreira Filho <janito.vff@gmail.com>
Co-authored-by: Janito Vaqueiro Ferreira Filho <janito.vff@gmail.com>
* Rename internal network requests for wide transaction IDs
fastmod TransactionsByHash TransactionsById zebra*
fastmod AdvertiseTransactions AdvertiseTransactionIds zebra*
fastmod MempoolTransactions MempoolTransactionIds zebra*
fastmod TransactionHashes TransactionIds zebra*
* Update network transaction request/response comments
* Rename a transaction hash method for wide transaction IDs
fastmod transaction_hashes transaction_ids zebra-network
* Add UnminedTxId methods and conversions for InventoryHash
* Map WtxIds to unmined transaction network messages
Also, use UnminedTxId and UnminedTx in:
* Zebra's internal request and response format, and
* external Zcash network protocol messages.
* Enable WtxId mempool inventory tracking for peers
* Further clarify transaction IDs
* Use Witnessed rather than Wide for transaction IDs
And rename narrow to legacy when it only applies to v1-v4 transactions.
Otherwise, rename it to mined ID.
* Rename a missed binding
* Remove an incorrectly named binding
Co-authored-by: Janito Vaqueiro Ferreira Filho <janito.vff@gmail.com>
* Simplify state service initialization in test
Use the test helper function to remove redundant code.
* Create `BestTipHeight` helper type
This type abstracts away the calculation of the best tip height based on
the finalized block height and the best non-finalized chain's tip.
* Add `best_tip_height` field to `StateService`
The receiver endpoint is currently ignored.
* Return receiver endpoint from service constructor
Make it available so that the best tip height can be watched.
* Update finalized height after finalizing blocks
After blocks from the queue are finalized and committed to disk, update
the finalized block height.
* Update best non-finalized height after validation
Update the value of the best non-finalized chain tip block height after
a new block is committed to the non-finalized state.
* Update finalized height after loading from disk
When `FinalizedState` is first created, it loads the state from
persistent storage, and the finalized tip height is updated. Therefore,
the `best_tip_height` must be notified of the initial value.
* Update the finalized height on checkpoint commit
When a checkpointed block is commited, it bypasses the non-finalized
state, so there's an extra place where the finalized height has to be
updated.
* Add `best_tip_height` to `Handshake` service
It can be configured using the `Builder::with_best_tip_height`. It's
currently not used, but it will be used to determine if a connection to
a remote peer should be rejected or not based on that peer's protocol
version.
* Require best tip height to init. `zebra_network`
Without it the handshake service can't properly enforce the minimum
network protocol version from peers. Zebrad obtains the best tip height
endpoint from `zebra_state`, and the test vectors simply use a dummy
endpoint that's fixed at the genesis height.
* Pass `best_tip_height` to proto. ver. negotiation
The protocol version negotiation code will reject connections to peers
if they are using an old protocol version. An old version is determined
based on the current known best chain tip height.
* Handle an optional height in `Version`
Fallback to the genesis height in `None` is specified.
* Reject connections to peers on old proto. versions
Avoid connecting to peers that are on protocol versions that don't
recognize a network update.
* Document why peers on old versions are rejected
Describe why it's a security issue above the check.
* Test if `BestTipHeight` starts with `None`
Check if initially there is no best tip height.
* Test if best tip height is max. of latest values
After applying a list of random updates where each one either sets the
finalized height or the non-finalized height, check that the best tip
height is the maximum of the most recently set finalized height and the
most recently set non-finalized height.
* Add `queue_and_commit_finalized` method
A small refactor to make testing easier. The handling of requests for
committing non-finalized and finalized blocks is now more consistent.
* Add `assert_block_can_be_validated` helper
Refactor to move into a separate method some assertions that are done
before a block is validated. This is to allow moving these assertions
more easily to simplify testing.
* Remove redundant PoW block assertion
It's also checked in
`zebra_state::service::check::block_is_contextually_valid`, and it was
getting in the way of tests that received a gossiped block before
finalizing enough blocks.
* Create a test strategy for test vector chain
Splits a chain loaded from the test vectors in two parts, containing the
blocks to finalize and the blocks to keep in the non-finalized state.
* Test committing blocks update best tip height
Create a mock blockchain state, with a chain of finalized blocks and a
chain of non-finalized blocks. Commit all the blocks appropriately, and
verify that the best tip height is updated.
Co-authored-by: teor <teor@riseup.net>
* Rename some methods and constants for clarity
Using the following commands:
```
fastmod '\bis_ready_for_attempt\b' is_ready_for_connection_attempt
# One instance required a tweak, because of the ASCII diagram.
fastmod '\bwas_recently_live\b' has_connection_recently_responded
fastmod '\bwas_recently_attempted\b' was_connection_recently_attempted
fastmod '\bwas_recently_failed\b' has_connection_recently_failed
fastmod '\bLIVE_PEER_DURATION\b' MIN_PEER_RECONNECTION_DELAY
```
* Use `Instant::elapsed` for conciseness
Instead of `Instant::now().saturating_duration_since`. They're both
equivalent, and `elapsed` only panics if the `Instant` is somehow
synthetically generated.
* Allow `Duration32` to be created in other crates
Export the `Duration32` from the `zebra_chain::serialization` module.
* Add some new `Duration32` constructors
Create some helper `const` constructors to make it easy to create
constant durations. Add methods to create a `Duration32` from seconds,
minutes and hours.
* Avoid gossiping unreachable peers
When sanitizing the list of peers to gossip, remove those that we
haven't seen in more than three hours.
* Test if unreachable addresses aren't gossiped
Create a property test with random addreses inserted into an
`AddressBook`, and verify that the sanitized list of addresses does not
contain any addresses considered unreachable.
* Test if new alternate address isn't gossipable
Create a new alternate peer, because that type of `MetaAddr` does not
have `last_response` or `untrusted_last_seen` times. Verify that the
peer is not considered gossipable.
* Test if local listener is gossipable
The `MetaAddr` representing the local peer's listening address should
always be considered gossipable.
* Test if gossiped peer recently seen is gossipable
Create a `MetaAddr` representing a gossiped peer that was reported to be
seen recently. Check that the peer is considered gossipable.
* Test peer reportedly last seen in the future
Create a `MetaAddr` representing a peer gossiped and reported to have
been last seen in a time that's in the future. Check that the peer is
considered gossipable, to check that the fallback calculation is working
as intended.
* Test gossiped peer reportedly seen long ago
Create a `MetaAddr` representing a gossiped peer that was reported to
last have been seen a long time ago. Check that the peer is not
considered gossipable.
* Test if just responded peer is gossipable
Create a `MetaAddr` representing a peer that has just responded and
check that it is considered gossipable.
* Test if recently responded peer is gossipable
Create a `MetaAddr` representing a peer that last responded within the
duration a peer is considered reachable. Verify that the peer is
considered gossipable.
* Test peer that responded long ago isn't gossipable
Create a `MetaAddr` representing a peer that last responded outside the
duration a peer is considered reachable. Verify that the peer is not
considered gossipable.
* Disable IPv6 tests when $ZEBRA_SKIP_IPV6_TESTS is set
This allows users to disable IPv6 tests in environments where IPv6 is not
configured.
* Add network test env var constants
* Replace env strings with constants
fastmod '"ZEBRA_SKIP_NETWORK_TESTS"' zebra_test::net::ZEBRA_SKIP_NETWORK_TESTS
fastmod '"ZEBRA_SKIP_IPV6_TESTS"' zebra_test::net::ZEBRA_SKIP_IPV6_TESTS
* Add functions to skip network tests
* Replace test network env var checks with test function
fastmod --fixed-strings 'env::var_os(zebra_test::net::ZEBRA_SKIP_NETWORK_TESTS).is_some()' 'zebra_test::net::zebra_skip_network_tests()'
fastmod --fixed-strings 'env::var_os(zebra_test::net::ZEBRA_SKIP_IPV6_TESTS).is_some()' 'zebra_test::net::zebra_skip_ipv6_tests()'
* Remove redundant logging and use statements
* Gossip dynamically allocated listener ports to peers
Previously, Zebra would either gossip port `0`, which is invalid, or skip
gossiping its own dynamically allocated listener port.
* Improve "no configured peers" warning
And downgrade from error to warning, because inbound-only nodes are a
valid use case.
* Move random_known_port to zebra-test
* Add tests for dynamic local listener ports and the AddressBook
Co-authored-by: Janito Vaqueiro Ferreira Filho <janito.vff@gmail.com>
* Always send our local listener with the latest time
Previously, whenever there was an inbound request for peers, we would
clone the address book and update it with the local listener.
This had two impacts:
- the listener could conflict with an existing entry,
rather than unconditionally replacing it, and
- the listener was briefly included in the address book metrics.
As a side-effect, this change also makes sanitization slightly faster,
because it avoids some useless peer filtering and sorting.
* Skip listeners that are not valid for outbound connections
* Filter sanitized addresses Zebra based on address state
This fix correctly prevents Zebra gossiping client addresses to peers,
but still keeps the client in the address book to avoid reconnections.
* Add a full set of DateTime32 and Duration32 calculation methods
* Refactor sanitize to use the new DateTime32/Duration32 methods
* Security: Use canonical SocketAddrs to avoid duplicate connections
If we allow multiple variants for each peer address, we can make multiple
connections to that peer.
Also make sure sanitized MetaAddrs are valid for outbound connections.
* Test that address books contain the local listener address
Co-authored-by: Janito Vaqueiro Ferreira Filho <janito.vff@gmail.com>
* Security: Limit reconnection rate to individual peers
Reconnection Rate
Limit the reconnection rate to each individual peer by applying the
liveness cutoff to the attempt, responded, and failure time fields.
If any field is recent, the peer is skipped.
The new liveness cutoff skips any peers that have recently been attempted
or failed. (Previously, the liveness check was only applied if the peer
was in the `Responded` state, which could lead to repeated retries of
`Failed` peers, particularly in small address books.)
Reconnection Order
Zebra prefers more useful peer states, then the earliest attempted,
failed, and responded times, then the most recent gossiped last seen
times.
Before this change, Zebra took the most recent time in all the peer time
fields, and used that time for liveness and ordering. This led to
confusion between trusted and untrusted data, and success and failure
times.
Unlike the previous order, the new order:
- tries all peers in each state, before re-trying any peer in that state,
and
- only checks the the gossiped untrusted last seen time
if all other times are equal.
* Preserve the later time if changes arrive out of order
* Update CandidateSet::next documentation
* Update CandidateSet state diagram
* Fix variant names in comments
* Explain why timestamps can be left out of MetaAddrChanges
* Add a simple test for the individual peer retry limit
* Only generate valid Arbitrary PeerServices values
* Add an individual peer retry limit AddressBook and CandidateSet test
* Stop deleting recently live addresses from the address book
If we delete recently live addresses from the address book, we can get a
new entry for them, and reconnect too rapidly.
* Rename functions to match similar tokio API
* Fix docs for service sorting
* Clarify a comment
* Cleanup a variable and comments
* Remove blank lines in the CandidateSet state diagram
* Add a multi-peer proptest that checks outbound attempt fairness
* Fix a comment typo
Co-authored-by: Janito Vaqueiro Ferreira Filho <janito.vff@gmail.com>
* Simplify time maths in MetaAddr
* Create a Duration32 type to simplify calculations and comparisons
* Rename variables for clarity
* Split a string constant into multiple lines
* Make constants match rustdoc order
Co-authored-by: Janito Vaqueiro Ferreira Filho <janito.vff@gmail.com>
* Security: stop gossiping failure and attempt times as last_seen times
Previously, Zebra had a single time field for peer addresses, which was
updated every time a peer was attempted, sent a message, or failed.
This is a security issue, because the `last_seen` time should be
"the last time [a peer] connected to that node", so that
"nodes can use the time field to avoid relaying old 'addr' messages".
So Zebra was sending incorrect peer information to other nodes.
As part of this change, we split the `last_seen` time into the
following fields:
- untrusted_last_seen: gossiped from other peers
- last_response: time we got a response from a directly connected peer
- last_attempt: time we attempted to connect to a peer
- last_failure: time a connection with a peer failed
* Implement Arbitrary and strategies for MetaAddrChange
Also replace the MetaAddr Arbitrary impl with a derive.
* Write proptests for MetaAddr and MetaAddrChange
MetaAddr:
- the only times that get included in serialized MetaAddrs are
the untrusted last seen and responded times
MetaAddrChange:
- the untrusted last seen time is never updated
- the services are only updated if there has been a handshake
* Only advance the outbound connection timer when it returns an address
Previously, we were advancing the timer even when we returned `None`.
This created large wait times when there were no eligible peers.
* Refactor to avoid overlapping sleep timers
* Add a maximum next peer delay test
Also refactor peer numbers into constants.
* Make the number of proptests overridable by the standard env var
Also cleanup the test constants.
* Test that skipping peer connections also skips their rate limits
* Allow an extra second after each sleep on loaded machines
macOS VMs seem to need this extra time to pass their tests.
* Restart test time bounds from the current time
This change avoids test failures due to cumulative errors.
Also use a single call to `Instant::now` for each test round.
And print the times when the tests fail.
* Stop generating invalid outbound peers in proptests
The candidate set proptests will fail if enough generated peers are
invalid for outbound connections.
Rust atomics have an API that's very easy to use incorrectly, leading to
hard to find bugs. For that reason, it's best to avoid it unless there's
a good reason not to.
* Rename field to `wait_next_handshake`
Make the name a bit more clear regarding to the field's purpose.
* Move `MIN_PEER_CONNECTION_INTERVAL` to `constants`
Move it to the `constants` module so that it is placed closer to other
constants for consistency and to make it easier to see any relationships
when changing them.
* Rate limit calls to `CandidateSet::update()`
This effectively rate limits requests asking for more peer addresses
sent to the same peer. A new `min_next_crawl` field was added to
`CandidateSet`, and `update` only sends requests for more peer addresses
if the call happens after the instant specified by that field. After
sending the requests, the field value is updated so that there is a
`MIN_PEER_GET_ADDR_INTERVAL` wait time until the next `update` call
sends requests again.
* Include `update_initial` in rate limiting
Move the rate limiting code from `update` to `update_timeout`, so that
both `update` and `update_initial` get rate limited.
* Test `CandidateSet::update` rate limiting
Create a `CandidateSet` that uses a mocked `PeerService`. The mocked
service always returns an empty list of peers, but it also checks that
the requests only happen after expected instants, determined by the
fanout amount and the rate limiting interval.
* Refactor to create a `mock_peer_service` helper
Move the code from the test to a utility function so that another test
will be able to use it as well.
* Check number of times service was called
Use an `AtomicUsize` shared between the service and the test body that
the service increments on every call. The test can then verify if the
service was called the number of times it expected.
* Test calling `update` after `update_initial`
The call to `update` should be skipped because the call to
`update_initial` should also be considered in the rate limiting.
* Mention that call to `update` may be skipped
Make it clearer that in this case the rate limiting causes calls to be
skipped, and not that there's an internal sleep that happens.
Also remove "to the same peers", because it's more general than that.
Co-authored-by: teor <teor@riseup.net>
* Rate-limit new outbound peer connections
Set the rate-limiting sleep timer to use a delay added to the maximum
between the next peer connection instant and now. This ensures that the
timer always sleeps at least the time used for the delay.
This change fixes rate-limiting new outbound peer connections, since
before there could be a burst of attempts until the deadline progressed
to the current instant.
Fixes#2216
* Create `MetaAddr::alternate_node_strategy` helper
Creates arbitrary `MetaAddr`s as if they were network nodes that sent
their listening address.
* Test outbound peer connection rate limiting
Tests if connections are rate limited to 10 per second, and also tests
that sleeping before continuing with the attempts still respets the rate
limit and does not result in a burst of reconnection attempts.
Given a generated list of gossiped peers, ensure that after running the
`validate_addrs` function none of the resulting peers have a `last_seen`
time that's after the specified limit.
If the calculation to apply the compensation offset overflows or
underflows, the reported times are too distant apart, and could be sent
on purpose by a malicious peer, so all addresses from that peer should
be rejected.
Use some mock gossiped peers where some have `last_seen` times in the
past and some have times in the future. Check that all the peers have
an offset applied to them by the `validate_addrs` function.
This tests if the offset is applied to all peers that a malicious peer
gossiped to us.
Use some mock gossiped peers that all have `last_seen` times in the
past and check that they don't have any changes to the `last_seen` times
applied by the `validate_addrs` function.
If any of the times gossiped by a peer are in the future, apply the
necessary offset to all the times gossiped by that peer. This ensures
that all gossiped peers from a malicious peer are moved further back in
the queue.
Co-authored-by: teor <teor@riseup.net>
- Make the security impact clearer and in a separate section.
- Instead of listing an assumption as almost a side-note, describe it
clearly inside a `Panics` section.
Co-authored-by: teor <teor@riseup.net>
Due to clock skew, the peers could end up at the front of the
reconnection queue or far at the back. The solution to this is to offset
the reported times by the difference between the most recent reported
sight (in the remote clock) and the current time (in the local clock).
Returning `impl IntoIterator` means that the caller will always be
forced to call `.into_iter()`, and returning `impl Iterator` still
allows them to call `.into_iter()` because it becomes the identity
function.
* Refactor: Split CandidateSet::update into separate functions
* Security: Apply a timeout to the entire CandidateSet::update
* Security: Stop using very large fanout limits during initialization
Previously, Zebra used the number of resolved peer addresses.
So it was possible for all peers to fail, and for Zebra to hang on the
first update.
And Zebra could send a fanout for each initial peer, regardless
of whether their connection was successful.
Also:
- wait for at least one successful peer before trying an update
- warn if there are no successful initial peers
When peers ask for peer addresses, add our local listener address to the
set of addresses, sanitize, then truncate. Sanitize shuffles addresses,
so if there are lots of addresses in the address book, our address will
only be sent to some peers.
If there is a small number of initial peers, and they are slow, the
initial candidate set update can appear to hang. To avoid this issue,
limit the initial candidate set fanout to the number of initial peers.
Once the initial peers have sent us more peer addresses, there is no need
to limit the fanouts for future updates.
Reported by Niklas Long of Equilibrium.
* Instrument the crawl task
When we created the crawl task, we forgot to instrument it with the
global span. This fix makes sure that the git and network span appears on
crawl logs.
* Instrument the connector
* Improve handshake instrumentation
Make some spans debug, so there are not too many spans.
* Add the address to initial peer connection errors
- stop putting inbound addresses in the address book
- drop address book entries that can't be used for outbound connections
- distinguish between temporary inbound and permanent outbound peer
addresses
- also create variants to handle proxy connections
(but don't use them yet)
- avoid tracking connection state for isolated connections
- document security constraints for the address book and peer set
* Allow use listen address in config without port
* update comments
* remove not used alias
* use Network::default_port
* Move tests and use toml instead json
* change error message
* Make match more readable
Co-authored-by: teor <teor@riseup.net>
* Stop ignoring inbound message errors and handshake timeouts
To avoid hangs, Zebra needs to maintain the following invariants in the
handshake and heartbeat code:
- each handshake should run in a separate spawned task
(not yet implemented)
- every message, error, timeout, and shutdown must update the peer address state
- every await that depends on the network must have a timeout
Once the Connection is created, it should handle timeouts.
But we need to handle timeouts during handshake setup.
* Avoid hangs by adding a timeout to the candidate set update
Also increase the fanout from 1 to 2, to increase address diversity.
But only return permanent errors from `CandidateSet::update`, because
the crawler task exits if `update` returns an error.
Also log Peers response errors in the CandidateSet.
* Use the select macro in the crawler to reduce hangs
The `select` function is biased towards its first argument, risking
starvation.
As a side-benefit, this change also makes the code a lot easier to read
and maintain.
* Split CrawlerAction::Demand into separate actions
This refactor makes the code a bit easier to read, at the cost of
sometimes blocking the crawler on `candidates.next()`.
That's ok, because `next` only has a short (< 100 ms) delay. And we're
just about to spawn a separate task for each handshake.
* Spawn a separate task for each handshake
This change avoids deadlocks by letting each handshake make progress
independently.
* Move the dial task into a separate function
This refactor improves readability.
* Fix buggy future::select function usage
And document the correctness of the new code.
Log a "Trying..." message before each listener opens, to see if the
delay is inside Zebra, or in the test harness or OS.
Also report the configured and actual ports where possible, for better
diagnostics.
Design:
- Add a `PeerAddrState` to each `MetaAddr`
- Use a single peer set for all peers, regardless of state
- Implement time-based liveness as an `AddressBook` method, rather than
a `PeerAddrState` variant
- Delete `AddressBook.by_state`
Implementation:
- Simplify `AddressBook` changes using `update` and `take` modifier
methods
- Simplify the `AddressBook` iterator implementation, replacing it with
methods that are more obviously correct
- Consistently collect peer set metrics
Documentation:
- Expand and update the peer set documentation
We can optimise later, but for now we want simple code that is more
obviously correct.
* replace to_socket_addrs
* refactor `resolve()` into `resolve_host()`
* use `resolve_host()` to resolve config peers
* add DNS_LOOKUP_TIMEOUT constant
* don't block the main thread in initialize
* add hint for port error
* add issue filter for port panic
* add lock file hint
* add metrics endpoint port conflict hint
* add hint for tracing endpoint port conflict
* add acceptance test for resource conflics
* Split out common conflict test code into a function
* Add state, metrics, and tracing conflict tests
* Add a full set of stderr acceptance test functions
This change makes the stdout and stderr acceptance test interfaces
identical.
* move Zcash listener opening
* add todo about hint for disk full
* add constant for lock file
* match path in state cache
* don't match windows cache path
* Use Display for state path logs
Avoids weird escaping on Windows when using Debug
* Add Windows conflict error messages
* Turn PORT_IN_USE_ERROR into a regex
And add another alternative Windows-specific port error
Co-authored-by: teor <teor@riseup.net>
Co-authored-by: Jane Lusby <jane@zfnd.org>
## Motivation
This PR is motivated by the regression identified in https://github.com/ZcashFoundation/zebra/issues/1349. That PR notes that the metrics stopped working for most of the crates other than `zebrad`.
## Solution
This PR resolves the regression by deduplicating the `metrics` crate dependency. During a recent change we upgraded the metrics version in `zebrad` and a couple other of our crates, but we never updated the dependencies in `zebra-state`, `zebra-consensus`, or `zebra-network`. This caused the metrics macros to attempt to retrieve the current metrics exporter through the wrong function. We would install the metrics exporter in `0.13`, but then attempt to look it up through the `0.12` crate, which contains a different instance of the metrics exporter static variable which is unset. Doing this causes the metrics macros to return `None` for the current exporter after which they just silently give up.
## Related Issues
closes https://github.com/ZcashFoundation/zebra/issues/1349
## Follow Up Work
I noticed we have quite a few duplicate dependencies in our tree. We might be able to save some compilation time by auditing those and deduplicating them as much as possible.
- https://github.com/ZcashFoundation/zebra/issues/1582
Co-authored-by: teor <teor@riseup.net>
* warn: if there are no peers at all
* info: if there are no ready peers
* trace: the number of ready and unready peers for every request
Log at most one warn or info log per minute, to avoid flooding the
terminal with log lines. Suppress warn and info logs for the first
minute, while the peer set is starting up.
Closes#1183.
The peer set maintains a preselected ready service that it can use to
perform power-of-two-choices (p2c) routing of requests. Ready services
are stored by key (socket address) in an `IndexMap`, and the preselected
service is represented by an `Option<usize>` indexing that map. This
means that whenever the set of ready services changes (e.g., a service
is removed from the peer set, or a service is taken to be used to
process a request), the preselected index is invalidated. The original
P2C-only implementation maintained this invariant but did not document
it.
The change to inventory-based routing introduced a bug by failing to
maintain this invariant and appropriately invalidate the preselected
index. However, this was only noticeable approximately 1/N of the time
on the next request after an inventory-directed request, so the bug
occurred infrequently. Luckily, the use of `.expect` caused the bug to
be an immediate panic, making it possible to identify by inspecting all
uses of the ready service map.
This change is mostly mechanical, with the exception of the changes to the
`tower-batch` middleware. This middleware was adapted from `tower::buffer`,
and the `tower::buffer` code was changed to implement its own bounded queue,
because Tokio 0.3 removed the `mpsc::Sender::poll_send` method. See
ddc64e8d4d
for more context on the Tower changes. To match Tower as closely as possible
in order to be able to upstream `tower-batch`, those changes are copied from
`tower::Buffer` to `tower-batch`.
This addresses at least three pain points:
- we were affected by bugs that were already fixed in git, but not in
the released crate;
- we can use service combinators to transform requests and responses;
- we can use the hedge middleware.
The version in git is still marked as 0.3.1 but these changes will be
part of tower 0.4: https://github.com/tower-rs/tower/issues/431
The GetPeers requests sent while crawling the network are randomly
load-balanced over available peers. But at the very beginning, they may
be both routed to the same peer, causing network initialization to be
delayed while the second one times out (since zcashd only ever responds
to the first addr message).
Only sending one GetPeers request per candidate set update means we
crawl the network a little more slowly, but avoids hanging on start.
This commit makes several related changes to the network code:
- adds a `TransactionsByHash(HashSet<transaction::Hash>)` request and
`Transactions(Vec<Arc<Transaction>>)` response pair that allows
fetching transactions from a remote peer;
- adds a `PushTransaction(Arc<Transaction>)` request that pushes an
unsolicited transaction to a remote peer;
- adds an `AdvertiseTransactions(HashSet<transaction::Hash>)` request
that advertises transactions by hash to a remote peer;
- adds an `AdvertiseBlock(block::Hash)` request that advertises a block
by hash to a remote peer;
Then, it modifies the connection state machine so that outbound
requests to remote peers are handled properly:
- `TransactionsByHash` generates a `getdata` message and collects the
results, like the existing `BlocksByHash` request.
- `PushTransaction` generates a `tx` message, and returns `Nil` immediately.
- `AdvertiseTransactions` and `AdvertiseBlock` generate an `inv`
message, and return `Nil` immediately.
Next, it modifies the connection state machine so that messages
from remote peers generate requests to the inbound service:
- `getdata` messages generate `BlocksByHash` or `TransactionsByHash`
requests, depending on the content of the message;
- `tx` messages generate `PushTransaction` requests;
- `inv` messages generate `AdvertiseBlock` or `AdvertiseTransactions`
requests.
Finally, it refactors the request routing logic for the peer set to
handle advertisement messages, providing three routing methods:
- `route_p2c`, which uses p2c as normal (default);
- `route_inv`, which uses the inventory registry and falls back to p2c
(used for `BlocksByHash` or `TransactionsByHash`);
- `route_all`, which broadcasts a request to all ready peers (used for
`AdvertiseBlock` and `AdvertiseTransactions`).
* network: fix bug in inventory advertisement handling
The RFC https://zebra.zfnd.org/dev/rfcs/0003-inventory-tracking.html described
the use of a `broadcast` channel in place of an `mpsc` channel to get
ring-buffer behavior, keeping a bound on the size of the channel but dropping
old entries when the channel is full.
However, it didn't explicitly describe how this works (the `broadcast` channel
returns a `RecvError::Lagged(u64)` to inform receivers that they lost
messages), so the lag-handling wasn't implemented and I didn't notice in
review.
Instead, the ? operator bubbled the lag error all the way up from
`InventoryRegistry::poll_inventory` through `<PeerSet as Service>::poll_ready`
through various Tower wrappers to users of the peer set. The error propagation
is bad enough, because it caused client errors that shouldn't have happened,
but there's a worse interaction.
The `Service` contract distinguishes between request errors (from
`Service::call`, scoped to the request) and service errors (from
`Service::poll_ready`, scoped to the service). The `Service` contract
specifies that once a service returns an error from `poll_ready`, the service
can be assumed to be failed permanently.
I believe (but haven't tested or carefully worked through the details) that
this caused various tower middleware to report the entire peer set service as
permanently failed due to a transient inventory "error" (more of an indicator),
and I suspect that this is the cause of #1003, where all of the sync
component's requests end up failing because the peer set reported that it
failed permanently. I am able to reproduce #1003 locally before this change
and unable to reproduce it locally after this change, though I have not tested
exhaustively.
* network: add metric for dropped inventory advertisements
Co-authored-by: teor <teor@riseup.net>
Co-authored-by: teor <teor@riseup.net>
The relay flag in the version message is used in conjunction with BIP37 to
receive bloom-filtered transactions. When it is set to false, transactions are
not relayed until a bloom filter is set. Since we don't implement BIP37 (it's
not useful for shielded transactions), this means we'll never receive
transactions.
* network: move gossiped peer selection logic into address book.
* network: return BoxService from init.
* zebrad: add note on why we truncate thegossiped peer list
Co-authored-by: Jane Lusby <jlusby42@gmail.com>
* Remove unused .rustfmt.toml
Many of these options are never actually loaded by our CI because of a channel
mismatch, where they're not applied on stable but only on nightly (see the logs
from a rustfmt job). This means that we can get different settings when
running `cargo fmt` on the nightly and stable channels, which was causing a CI
failure on this PR. Reverting back to the default rustfmt settings avoids this
problem and keeps us in line with upstream rustfmt. There's no loss to us
since we were using the defaults anyways.
Co-authored-by: Jane Lusby <jlusby42@gmail.com>
Closes#536.
This removes:
- the user-agent (we can add a mechanism to specify extra BIP14 components later, if any users ask us for that feature);
- the EWMA parameters (these were put in the config just to avoid making a choice);
- the peer connection timeout (we can change the default value if anyone ever has a problem with it);
- the peer set request buffer size (setting this too low can make the application deadlock);
The new peer interval is left in.
Prior to this change, we required that services that are canceled do not
have a cancel handle in the `cancel_handles` list, based on the
assumption that the handle must have been removed in the process of
canceling this service.
This doesn't holding up though, because it is currently possible for us
to have the same peer connect to us multiple times, the second connect
removes the cancel handle of the original connect and inserts it's own
cancel handle in its place. In this scenario, when the first service is
polled for readiness it will see that it has been canceled and go to
clean itself up, but when it asserts that it doesn't have a cancel
handle it will see the cancel handle of the second connect event, which
uses the same key as the first connect, and fail its debug assertion.
This change removes that debug assert on the assumption that it is okay
for a peer to connect multiple times consecutively, and that the correct
behavior in that case is to just cancel the first connection and
continue as normal.
teor
Alfredo Garcia
Janito Vaqueiro Ferreira Filho
Kirill Fomichev
Jack Grigg
Jane Lusby
Henry de Valence
Deirdre Connolly
Jane Lusby
Dimitris Apostolou