don't contend on higher level sup to find worker sup
rework a place I missed
monitor session dials
more hax
add sidejob
rebar stuff
stop inbound workers anyway
shutdown rather than normal
stop worker properly
stop worker out of process, catch closes
inbound workers stop on failure
don't copy state around
remove some debug loggin
fix compilation
Send all new peers to any seed nodes we're connected to
handle sidejob overload
Cherry pick async identify
Fix starting inbound worker
rework peerbook repair and clearing
Clear the peerbook if it's not corrupted
Avoid p2p address conversions in gossip server
Stungun improvements:
Check for 1:1 port mappings with stungun
Often there's a port mapping manually configured such that the external
port is mapped through to the same internal port. When we detect
symmetric NAT, additionally try to use stungun on these fixed ports to
try to discover if this kind of mapping exists.
Improve stungun retries
Stungun will now try to continue resolving the NAT state until it gets
an answer (nodes will no longer get stuck in 'unknown').
Improved validation
Stungun now checks the session validating the external IP/Port comes
from an inbound session and ignores it otherwise.
Don't use match_delete on remove_pid, soft delete and use a GC
Do some more work upfront to speed up gc
We take a hit on potentially trying to delete some keys that don't
exist. Deleting against explicit pids means we won't delete anything
not yet marked. :fingers_crossed:
Re-work ETS GC to match all the keys it needs to and be single-pass
Handle overload when starting outbound gossip workers
Yolo spawn
Raise listen backlog
Move the gossip handling out into the worker
Cache gossip handler lookups for 1 minute, infinity timeout
1 minute cache is bad when block times are also 1 minute
make the workers accounting use maps
raise accept timeout
don't contend on higher level sup to find worker sup
rework a place I missed
monitor session dials
more hax
add sidejob
rebar stuff
stop inbound workers anyway
shutdown rather than normal
stop worker properly
stop worker out of process, catch closes
inbound workers stop on failure
don't copy state around
remove some debug loggin
fix compilation
Send all new peers to any seed nodes we're connected to
handle sidejob overload
Cherry pick async identify
Fix starting inbound worker
rework peerbook repair and clearing
Clear the peerbook if it's not corrupted
Avoid p2p address conversions in gossip server
Stungun improvements:
Check for 1:1 port mappings with stungun
Often there's a port mapping manually configured such that the external port is mapped through to the same internal port. When we detect symmetric NAT, additionally try to use stungun on these fixed ports to try to discover if this kind of mapping exists.
Improve stungun retries
Stungun will now try to continue resolving the NAT state until it gets an answer (nodes will no longer get stuck in 'unknown').
Improved validation
Stungun now checks the session validating the external IP/Port comes from an inbound session and ignores it otherwise.
Don't use match_delete on remove_pid, soft delete and use a GC
Do some more work upfront to speed up gc
We take a hit on potentially trying to delete some keys that don't exist. Deleting against explicit pids means we won't delete anything not yet marked. :fingers_crossed:
Re-work ETS GC to match all the keys it needs to and be single-pass
Handle overload when starting outbound gossip workers
Yolo spawn
Raise listen backlog
Move the gossip handling out into the worker
Cache gossip handler lookups for 1 minute, infinity timeout
1 minute cache is bad when block times are also 1 minute
Bump splicer to latest release
Fix gossip tuple stuff