* docs: update systemd example for the serve CLI with required node auth The unit and readme still described the old HTTP-API miner (MINER_PORT, MINER_ENGINE, no subcommand). Update ExecStart to 'serve' and document the now-required MINER_AUTH_TOKEN_FILE / MINER_TLS_CERT_SHA256_FILE, including how to copy the node's miner-auth-token and miner-tls-cert-sha256 files past ProtectHome=true, and refresh the env var reference to the current CLI. * docs: fix env-file example and CPU-worker claims per review - Move inline comments off the env assignments: systemd EnvironmentFile= keeps inline '# ...' text as part of the value, so the copied example fed Clap unparseable values and the unit crash-looped under Restart=always. - Describe the real worker behavior: CPUs are counted via the process affinity mask (num_cpus); unset MINER_CPU_WORKERS auto-detects ~50%, and an explicit value is used as-is — there is no clamp or warning. - Drop the 'logs the detected cpuset mask' claim from the unit and override comments; no such logging exists in the miner. * docs: metrics exporter is always on; purge-chain does not rotate credentials - The Prometheus exporter starts unconditionally and binds plaintext HTTP on 0.0.0.0:9900 by default; MINER_METRICS_PORT only changes the port. Say so everywhere instead of 'when metrics are enabled', and add a hardening note to firewall the port. - Rephrase the credential-rotation pitfall: purge-chain removes only the database, so the node reloads the same miner-auth-token and TLS cert; rotation only happens when the base path changes or the files themselves are deleted.
53 lines
2.0 KiB
Plaintext
53 lines
2.0 KiB
Plaintext
# Drop-in override: conservative CPU affinity for shared hardware (node + miner)
|
||
# Place this file at:
|
||
# /etc/systemd/system/quantus-miner.service.d/10-shared-hardware.conf
|
||
#
|
||
# Then reload and restart:
|
||
# sudo systemctl daemon-reload
|
||
# sudo systemctl restart quantus-miner
|
||
#
|
||
# Rationale:
|
||
# - On shared hosts where the node and miner co-exist, bind the miner to a subset
|
||
# of logical CPUs, leaving the remainder for the node (and the OS scheduler).
|
||
# - Use a slightly lower CPU priority and moderate CPU weight so the node remains
|
||
# responsive under load.
|
||
#
|
||
# Notes:
|
||
# - CPUAffinity must be a subset of the service’s allowed cpuset (e.g., cgroups).
|
||
# - The miner counts CPUs via the process affinity mask and exports a
|
||
# miner_effective_cpus gauge on the always-on metrics endpoint.
|
||
# - Adjust the CPU list and worker count to match your hardware and cpuset.
|
||
#
|
||
# Example below assumes an 8‑way logical CPU system (IDs 0‑7).
|
||
# Reserve CPUs 0‑3 for the node and bind the miner to CPUs 4‑7.
|
||
|
||
[Service]
|
||
# Bind miner to specific CPUs (edit to match your topology).
|
||
# Check `lscpu` and any cpuset constraints (cgroup v2: /sys/fs/cgroup/cpuset.cpus.effective).
|
||
CPUAffinity=4 5 6 7
|
||
|
||
# Keep the node responsive; slightly lower miner priority (higher niceness value).
|
||
Nice=5
|
||
|
||
# If using cgroup v2, weight the miner below default (1–10000, default ~100).
|
||
# This helps the node compete for CPU on shared hardware.
|
||
CPUWeight=70
|
||
|
||
# I/O scheduling: best-effort with moderate priority.
|
||
IOSchedulingClass=best-effort
|
||
IOSchedulingPriority=4
|
||
|
||
# Default time-sharing policy is fine for CPU-bound workloads when pinned.
|
||
CPUSchedulingPolicy=other
|
||
|
||
# Optional: pin worker count and change the metrics port (uncomment and adjust as needed).
|
||
# Environment=MINER_CPU_WORKERS=4
|
||
# Environment=MINER_METRICS_PORT=9900
|
||
# Environment=RUST_LOG=info,miner=debug
|
||
|
||
# Optional: tighten memory/task limits if desired (ensure they fit your environment).
|
||
# MemoryAccounting=true
|
||
# TasksAccounting=true
|
||
# MemoryMax=infinity
|
||
# TasksMax=infinity
|