reymon.xyz · raspberry pi 5

darknode // recorded

What running a DarkFi node actually costs, measured from inside it. The block explorer sees what the network agreed on; this is the view only the operator has.

digest stale07-18 19:50Z → 08-18 05:26Z

recorded

30.4d

40,897 snapshots

coverage

93.4%

4 gaps >150s

zero-peer

episodes — network lost

wall hits

memory-ceiling episodes

restarts

memguard, all graceful

stalls

height frozen, peers up

01 // memory budget over time

working set (anon)+ reclaimable cacheMemoryHigh — softMemoryMax — hard wallmemguard restart

The sawtooth is the story. The heap climbs for days, squeezes the page cache toward zero, and the node stalls before it ever reaches the wall — memguard then restarts it gracefully rather than letting the cgroup OOM-kill sled mid-write. Every drop to the floor is one of those restarts.

02 // episodes

when	kind	lasted	detail
07-20 11:37Z	restart	—	before the cgroup wall
07-25 11:28Z	memory ceiling	4m	peak 4.07 G ·
07-25 13:54Z	stall	48m	stuck at 31,545 · height frozen with peers connected
07-26 04:27Z	stall	37m	stuck at 31,941 · height frozen with peers connected
07-26 04:57Z	memory ceiling	7m	peak 4.13 G ·
07-26 05:05Z	restart	—	recovered 84 blocks · before the cgroup wall
07-30 03:10Z	memory ceiling	7m	peak 4.10 G ·
07-30 03:18Z	restart	—	recovered 15 blocks · before the cgroup wall
08-02 06:39Z	memory ceiling	6h 29m	peak 4.19 G ·
08-02 10:12Z	stall	1h 0m	stuck at 37,080 · height frozen with peers connected
08-02 12:26Z	stall	42m	stuck at 37,107 · height frozen with peers connected
08-02 13:09Z	restart	—	recovered 188 blocks · before the cgroup wall
08-04 15:04Z	memory ceiling	1m	peak 3.94 G ·
08-04 15:32Z	memory ceiling	1m	peak 3.96 G ·
08-04 22:33Z	stall	31m	stuck at 39,020 · height frozen with peers connected
08-04 23:04Z	stall	55m	stuck at 39,021 · height frozen with peers connected
08-05 00:05Z	memory ceiling	8m	peak 4.19 G ·
08-05 00:14Z	restart	—	recovered 74 blocks · before the cgroup wall
08-07 14:48Z	stall	4h 19m	stuck at 40,965 · height frozen with peers connected
08-07 19:07Z	stall	17h 50m	stuck at 40,960 · height frozen with peers connected
08-08 12:56Z	restart	—	before the cgroup wall
08-09 08:14Z	restart	—	recovered 7 blocks · before the cgroup wall
08-10 12:11Z	restart	—	recovered 4 blocks · before the cgroup wall
08-13 06:47Z	restart	—	recovered 11 blocks · before the cgroup wall
08-14 05:29Z	restart	—	before the cgroup wall
08-14 06:08Z	stall	32m	stuck at 45,703 · height frozen with peers connected
08-14 06:40Z	stall	41m	stuck at 45,704 · height frozen with peers connected
08-14 07:27Z	stall	48m	stuck at 45,706 · height frozen with peers connected
08-14 08:55Z	stall	52m	stuck at 45,710 · height frozen with peers connected
08-14 09:47Z	stall	39m	stuck at 45,711 · height frozen with peers connected
08-14 11:21Z	stall	1h 4m	stuck at 45,716 · height frozen with peers connected
08-14 12:26Z	stall	40m	stuck at 45,717 · height frozen with peers connected
08-14 13:16Z	stall	55m	stuck at 45,719 · height frozen with peers connected
08-14 15:08Z	stall	1h 6m	stuck at 45,727 · height frozen with peers connected
08-14 16:25Z	stall	40m	stuck at 45,730 · height frozen with peers connected
08-14 17:04Z	stall	38m	stuck at 45,731 · height frozen with peers connected
08-14 17:42Z	stall	48m	stuck at 45,732 · height frozen with peers connected
08-14 18:31Z	stall	37m	stuck at 45,733 · height frozen with peers connected
08-16 03:45Z	restart	—	recovered 11 blocks · before the cgroup wall
08-17 10:21Z	memory ceiling	12m	peak 3.95 G ·
08-17 10:34Z	restart	—	recovered 11 blocks · before the cgroup wall

03 // peer count across the record

4 blackouts0–7 peers

0 peers0.0%

1 peer1.4%

2 peers17.3%

3 peers37.8%

4 peers37.1%

5 peers5.9%

6 peers0.5%

7 peers0.0%

Counts only, never addresses. A zero was unrepresentable here until 2026-08-08 — the exporter blanked it into a missing field — so the blackouts above are only those recorded since that fix.

the overlay, from inside the node

Everything above is what running the node costs. This is what the node can see that an explorer cannot: who it talks to, how long those connections survive, and who is talking to it without ever being asked.

endpoints

seen in the window

sessions

4,605

slot-managed only

median lifetime

33s

longest 20.3 h

rtt median

35ms

223,933 pings

04 // who talks to whom

outbound slotno slot lifecycleweight = share of traffic

05 // what the node says

getaddr28.5%

addr25.1%

ping15.7%

pong14.9%

version7.1%

verack6.9%

proposal1.8%

Share of 3,116,086 wire messages. A node whose job is syncing a chain spends most of its words asking who else is out there.

06 // endpoints

#	tracking	share	sessions	longest	rtt
1	slot	20.0%	1973	14.4 h	58.5 ms
2	slot	21.0%	1920	3 m	34.1 ms
3	slot	21.5%	712	20.3 h	34.1 ms
4	slot	1.0%	—	—	47.5 ms
5	no slot	18.2%	—	—	34.3 ms
6	no slot	18.2%	—	—	34.4 ms

Numbered by first appearance, never by a hash of the address — reversible against a candidate list on a network this small. darkfid instruments outbound slots only, so the 36% of traffic on the 2 untracked endpoints appears in no session or churn figure above.