Opened 2 years ago

Closed 2 years ago

#2435 closed defect (duplicate)

NTCP V2 inbound conns starved

Reported by: jogger Owned by: zzz
Priority: minor Milestone: undecided
Component: router/transport Version: 0.9.38
Keywords: Cc:
Parent Tickets: #2346 Sensitive: no

Description (last modified by jogger)

Seeing some inbound NTCP V2 conns starved after receiving a single sent with outgoing packets piling up in the queue. Those stay indefinitely on the /peers page with queued packets going up over time.

Like this:

RU 0hMe Inbound 2 36 min / 36 min 0,00 / 0,00 36 min 1000 ms 1 0 1


Change History (6)

comment:1 Changed 2 years ago by jogger

Description: modified (diff)

comment:2 Changed 2 years ago by jogger

Description: modified (diff)

comment:3 Changed 2 years ago by jogger

Similar effect for outbound, but those go into backlogged and finally disappear:

WARN [ GW pumper 2] .transport.ntcp.NTCPConnection: Too backlogged: size is 512, wantsWrite? true, currentOut set? true, id: 51, writeBufs: 1 on NTCP2 conn 13139 to port 29151 kI833f created 5m ago, last send 5m ago, last recv 5m ago, sent 49, rcvd 1

comment:4 Changed 2 years ago by jogger

Now have this seen also for V1 inbound. Most extreme case, marked as backlogged for hours on /peers:

?? 0hMe Inbound 2 6 hours / 6 hours 0,00 / 0,00 6 hours 1000 ms 1 0 340 ✔

for this conn I found this log 6 hours old:

WARN [P reader ½] .transport.ntcp.NTCPConnection: Old connection closed: NTCP2 conn 920 to port 21608 0hMeZN created 1289ms ago, last send 40ms ago, last recv 1289ms ago, sent 3, rcvd 1 replaced by NTCP2 conn 924 from / port 61816 0hMeZN created 175ms ago, last send 175ms ago, last recv 175ms ago, sent 1, rcvd 0

Last edited 2 years ago by jogger (previous) (diff)

comment:5 Changed 2 years ago by zzz

Component: unspecifiedrouter/transport
Owner: set to zzz

comment:6 Changed 2 years ago by zzz

Parent Tickets: 2346
Resolution: duplicate
Status: newclosed
Note: See TracTickets for help on using tickets.