The revalidateMax value is lowered to have a quicker ramp up of
the radiusCache + to keep it healthier.
The defaultMaxGossipNodes value is lowered because with the
current value a Nodes lookup is triggered almost always.
This value is dependant on the content replication value. This
is dependant on the network (and subnetwork) because of the amount
of nodes and their radius/storage capacity.
This bug would have as effect that our radius cache would not get
filled by any outgoing pings, causing:
- Node lookups to always be occurring on NH gossip
- POKEs to much more rarely
Also add metrics for the amount of offers done via POKE mechanism.
* Make stop functions wait for completion before return.
* Implement graceful shutdown.
* Shutdown rpc and metric servers if enabled.
* Move metrics and rpc servers out of PortalNode.
This avoid restarting the node always with a full radius, which
causes the node the be bombarded with offers which it later has
to delete anyhow.
In order to implement this functionality, several changes were
made as the radius needed to move from the Portal wire protocol
current location to the contentDB and beaconDB, which is
conceptually more correct anyhow.
So radius is now part of the database objects and a handler is
used in the portal wire protocol to access its value.
`initTable` is obsolete since nim 0.19 and can introduce significant
memory overhead while providing no benefit (since the table will be
grown to the default initial size on first use anyway).
In particular, aristo layers will not necessarily use all tables they
initialize, for exampe when many empty accounts are being created.
* Add nph check to fluffy CI lint
* Add a section on nph usage in the fluffy.guide
* Update copyright years for altered files
* Avoid chained methods formatting style in db code
* Update nph in CI to v0.5
* Remove leftover commented import
* Move comment to avoid nph turning complex list into simple list (nph bug)
* Update nph in CI to v0.5.1
* Formatting fluffy with nph v0.5.1
- The current setting of 0.25 was very big. Set to 0.5, which is
potentially still large. This change did expose some issues with
the current implementation and especially testing.
- General clean-up, renaming for consistency, and
re-ordering/restructuring/deletion of some code.
- Fixed several typos
- ...
This would lead to sending an AcceptMessage with a zeroed out
connection id but still potentially bits set in the bitlist.
Requesting nodes could try to open up an uTP connection that
would never be accepted on the responder side.
- Link network gossip validation to LC processor validation
- QuickFix put/get for optimistic and finality updates
- Minor fixes and clean-up
- Improve bridge gossip for new LC Updates
- Adjust local testnet script to be able to locally test this
- Adjust test and skip broken test
* Add discv5 provided ENR directly to Portal protocol routing table
Previous version of getting the ENR from the discv5 routing table
would not work due to the order of first calling the talk protocol
handler and only after that the addEnr to the disc5 routing table.
Instead of changing this order, pass along the ENR directly to
avoid this additional getNode call.
* Still request ENR from discv5 if it wasn't passed via handshake
* Reduce Nim 1.6 compiler warnings/hints for Fluffy and Nimbus proxy
Mostly raises Defect removals, TaintedString removal and some
unnecessary imports.
Also updating the copyright years alongside.
* Further reduce Nim 1.6 compiler warnings/hints for Nimbus
The portal_*Offer call was changed in the specs to actually do
gossip. Make it no longer possible to test purely an offer with
one node. Add OfferReal call for now until spec potentially gets
adjusted.
Also Add some Node information logging for FindContent and Offer
to be able to better debug failures and interoperability.
Kim De Mey
web3-developer
Jacek Sieka
BiKangNing
Daniel Sobol
andri lim