This fixes a regression introduced in #32518. In that PR, we removed the
slowdown logic that would throttle lookups when the table runs empty.
Said logic was originally added in #20389.
Usually it's fine, but there exist pathological cases, such as hive
tests, where the node can only discover one other node, so it can only
ever query that node and won't get any results. In cases like these, we
need to throttle the creation of lookups to avoid crazy CPU usage.
The lookup would add self into the replyBuffer if returned by another node.
Avoid doing that by marking self as seen.
With the changed initialization behavior of lookup, the lookupIterator needs to yield the
buffer right after creation. This fixes the smallNetConvergence test, where all results
are straight out of the local table.
Here we clean up internal uses of type discover.node, converting most code to use
enode.Node instead. The discover.node type used to be the canonical representation of
network hosts before ENR was introduced. Most code worked with *node to avoid conversions
when interacting with Table methods. Since *node also contains internal state of Table and
is a mutable type, using *node outside of Table code is prone to data races. It's also
cleaner not having to wrap/unwrap *enode.Node all the time.
discover.node has been renamed to tableNode to clarify its purpose.
While here, we also change most uses of net.UDPAddr into netip.AddrPort. While this is
technically a separate refactoring from the *node -> *enode.Node change, it is more
convenient because *enode.Node handles IP addresses as netip.Addr. The switch to package
netip in discovery would've happened very soon anyway.
The change to netip.AddrPort stops at certain interface points. For example, since package
p2p/netutil has not been converted to use netip.Addr yet, we still have to convert to
net.IP/net.UDPAddr in a few places.
Node discovery periodically revalidates the nodes in its table by sending PING, checking
if they are still alive. I recently noticed some issues with the implementation of this
process, which can cause strange results such as nodes dropping unexpectedly, certain
nodes not getting revalidated often enough, and bad results being returned to incoming
FINDNODE queries.
In this change, the revalidation process is improved with the following logic:
- We maintain two 'revalidation lists' containing the table nodes, named 'fast' and 'slow'.
- The process chooses random nodes from each list on a randomized interval, the interval being
faster for the 'fast' list, and performs revalidation for the chosen node.
- Whenever a node is newly inserted into the table, it goes into the 'fast' list.
Once validation passes, it transfers to the 'slow' list. If a request fails, or the
node changes endpoint, it transfers back into 'fast'.
- livenessChecks is incremented by one for successful checks. Unlike the old implementation,
we will not drop the node on the first failing check. We instead quickly decay the
livenessChecks give it another chance.
- Order of nodes in bucket doesn't matter anymore.
I am also adding a debug API endpoint to dump the node table content.
Co-authored-by: Martin HS <martin@swende.se>
This change improves discovery behavior in small networks. Very small
networks would often fail to bootstrap because all member nodes were
dropping table content due to findnode failure. The check is now changed
to avoid dropping nodes on findnode failure when their bucket is almost
empty. It also relaxes the liveness check requirement for FINDNODE/v4
response nodes, returning unverified nodes as results when there aren't
any verified nodes yet.
The "findnode failed" log now reports whether the node was dropped
instead of the number of results. The value of the "results" was
always zero by definition.
Co-authored-by: Felix Lange <fjl@twurst.com>
This adds an implementation of the current discovery v5 spec.
There is full integration with cmd/devp2p and enode.Iterator in this
version. In theory we could enable the new protocol as a replacement of
discovery v4 at any time. In practice, there will likely be a few more
changes to the spec and implementation before this can happen.
