HL Node

Sync & blocks

Last applied height—

EVM block—

Block time (chain, UTC)—

Lag behind wall clock—

Blocks applied, last 60 s / 5 min—

Apply duration, 5-min mean—

Hardfork version—

visor_abci_state height—

visor consensus time—

visor wall-clock time—

visor state file written—

Initial height (this sync)—

Scheduled freeze height—

HL API clock skew (exchangeStatus − local)—

Block lag, last 24 h (s, log scale)

Blocks applied per minute, last 24 h

Process & resources

hl-visor.service—

visor up since / unit restarts—

hl-node child—

node CPU / RSS—

node threads / fds—

hl-node binary—

hl-visor binary—

signature / quorum files—

Busy % of the cores in the visor's CPUAffinity mask

/data volume

HL nodeother (hl-mm, PG…)usableroot-reserved

Growth /data used · 1 h / 24 h / 7 d—

Growth HL node dir · 1 h / 24 h / 7 d—

Usable space runs out in—

ENOSPC in visor journal, 24 h—

inodes—

hl-data-prune timer—

What	Size	% of /data

/data used, GB (storage sampler history + this service)

Prune rules (hl-data-prune, every minute)

Path	Keep

Disk guard & retention

Guard verdict—

/data used (df %) · thresholds—

Last check · action—

Checkpoint cap (hl-ckpt-prune)—

Checkpoints on disk—

Generations · size · span—

Last cap run—

Growth 24 h · node / PG / SQLite—

Growth 7 d · node / PG / SQLite—

Node tree	Size	Retention

Every tree under the node's data dir and the EVM state. Trees without a rule are flagged; the live EVM state is the one tree that must never be pruned.

Output streams (/data/hl/data) newest file age

Stream	Age	State	Newest

Peers & gossip

Peer	Dir	Sent	Recv	RTT	Flags

Consumers & SHM freshness WTI cockpit /hlmm/ →

Unit	Reads node	State	Up for	Restarts

SHM segment	Write age	Content lag	Verdict

Limits mirror /usr/local/bin/hl-bbo-freshness-check.py (write 30 s; content 60 s node / 120 s mempool).

Housekeeping	Last run	Result	Next

Network & validators

Validators active / total · jailed—

Gossip root IPs · API / our config—

Egress now / 24 h mean—

Projected egress—

Top validators by stake	Stake %	State

Box NIC tx, last 24 h (MB/s) — the node dominates egress while syncing

Visor journal, last 10 min

Cost & runbook

Data volume (gp3, 500 GB, 6000 IOPS)—

Node share of the volume—

Egress at current 24 h mean—

Node-attributable total—

hl-visor.service runs hl-visor run-non-validator as ubuntu; the visor spawns and auto-upgrades /data/hl/hl-node.
State: /data/hl/hyperliquid_data (EVM DB, abci_state.rmp). Streams: /data/hl/data/*, trimmed every minute by hl-data-prune.timer.
Peers: /data/hl/override_gossip_config.json, appended from the API every 4 h by hl-refresh-root-peers.timer; empirical list in known_good_peers.md.
Watchdog: hl-bbo-freshness-check.timer logs stale WTI segments every minute (log-only, no restarts).
Rule from /data/hl/CLAUDE.md: never pause the EC2 instance to save cost — hours of resync and a gap in replica_cmds. Pause trading by stopping MM units instead.
Logs: journalctl -u hl-visor.service (bounded with --since), child stderr under data/visor_child_stderr/.

Prices: egress $0.114/GB (ap-northeast-1, first 10 TB); gp3 $0.096/GB-mo + $0.006 per IOPS above 3000. Instance cost is shared with the MMs and not attributed here.

HL Node Hyperliquid non-validator · — · — · —