01Three pain signals budget QA teams misread as routing bugs
Stable medians hide memory cliffs, port collisions, and disk churn that look like packet loss.
- City RTT blind spots: Probing only Los Angeles misses San Jose wins toward GitHub and Stripe-class APIs.
- OPENCLAW_HOME overlap: Two gateways on one tree poison caches before CPU charts move.
- Port 18789 debt: Finance funds one host while engineering binds duplicate listeners overnight.
02Los Angeles versus San Jose decision matrix
Match staff geography first, then revalidate with the DNS and VPN SKU your API clients use in production.
| Signal | Lean Los Angeles | Lean San Jose |
|---|---|---|
| Operator geography | Southern California staff and PST-heavy on-call | Bay Area vendors plus Pacific QA benches |
| Upstream API edges | Often closer to LA CDN footprints you already monitor | Often closer to GitHub and SaaS peering you measure nightly |
| Parallel OpenClaw slots | Prefer when invoices must stay isolated per gateway | Prefer when one procurement line covers shared CI bursts |
03Los Angeles and San Jose to GitHub and common SaaS RTT gates
Run identical curl or k6 probes from each neokvm city toward github.com and your production API host. Fail three consecutive runs before rejecting a pair.
| Target | Pass p95 from either city | Fail or escalate |
|---|---|---|
| github.com TLS handshake | At or below forty five milliseconds p95 | Above sixty five milliseconds sustained |
| Your SaaS API edge | At or below eighty milliseconds p95 | Above one hundred ten milliseconds with flat CPU |
| OAuth refresh jitter | Under twelve milliseconds during a five minute soak | Spikes above thirty milliseconds with flat load |
04Mac mini M4 sixteen gigabyte OPENCLAW_HOME session ceiling
Keep session counts identical when comparing cities so RAM noise does not skew RTT. Export OPENCLAW_HOME before install, onboard, or doctor runs.
- Safe trio: Two interactive gateways plus one headless API poller on sixteen gigabytes.
- Home paths: Use
~/openclaw-alpha,~/openclaw-beta, and~/openclaw-ci—never symlink two homes into one folder. - Parallel trigger: Add a slot when a third gateway stays warm twenty four seven or swap stays nonzero five minutes at Pacific peak.
05Parallel slots and port 18789 conflict avoidance
Start with listeners, not opinions. Encode offsets in LaunchDaemon plist environment blocks before debating city choice again.
| Instance | Port | OPENCLAW_HOME |
|---|---|---|
| Alpha gateway | 18789 | ~/openclaw-alpha |
| Beta canary | 18809 | ~/openclaw-beta |
| CI burst | 18829 | ~/openclaw-ci |
Verify listeners
lsof -nP -iTCP:18789 -sTCP:LISTEN
lsof -nP -iTCP:18809 -sTCP:LISTENIf finance needs isolated blast radius, rent a second parallel slot instead of stacking four gateways on one sixteen gigabyte host.
06One terabyte versus two terabyte disk watermarks and rent cost thresholds
Disk pressure masquerades as TLS timeouts. Align tier bumps on pricing before sprint promises harden.
| Condition | Stay on one terabyte | Move to two terabytes |
|---|---|---|
| Free space floor | One hundred twenty gigabytes free before dual gateways | Two hundred gigabytes free with three log streams |
| Log retention | Daily compress plus weekly purge | Thirty day traces for audit-heavy API monitors |
| Rent term trigger | Monthly list when monitors run always on | Daily only for incidents under seventy two hours |
07Daily weekly monthly rent posture for US West OpenClaw benches
Relative economics for one Mac mini M4 class. Replace placeholders with live rows on pricing before finance sign-off.
| Term | Best fit | Cost posture versus monthly list |
|---|---|---|
| Daily | Incident bridges under seventy two hours | Highest effective hourly burn, lowest commitment |
| Weekly | Mid sprint OpenClaw upgrades without quarter contracts | Mid effective hourly burn with moderate flexibility |
| Monthly | Always on API monitors plus stable QA benches | Lowest effective hourly burn on published monthly rows |
08Install, onboard, and openclaw doctor troubleshooting FAQ
Install or onboard fails? Retry Homebrew after xcode-select --install, export OPENCLAW_HOME, confirm plist port rows, bounce one LaunchAgent at a time. See gateway troubleshooting.
Doctor merge noise? Run doctor per home and merge keys only into that home json before restart.
09Six step runbook before you rent Los Angeles or San Jose metal
- Probe RTT: Record LA and SJ p95 toward github.com and your API edge with one DNS profile.
- Apply gates: Mark pass fail with the RTT table before city debate.
- Smoke homes: Two gateways plus one poller for twenty minutes at Pacific peak.
- Encode ports: Document 18789, 18809, and 18829 with lsof attachments.
- Disk tier: Compare one terabyte versus two terabytes using the watermark table.
- Checkout: US West purchase then console for SSH keys.
10Citable facts for architecture notes
11FAQ for leads who want numbers not vibes
Does San Jose always beat Los Angeles? No—measure both with one client, DNS, and VPN SKU.
Need APAC metal instead? Use the APAC versus US West matrix when operators sit in East Asia.
Quota limits? Confirm listener and egress caps in help before external promises.
12Paste ready go live checklist
- RTT log: LA and SJ p95 toward GitHub and SaaS with VPN noted.
- Ports: lsof sheet for 18789, 18809, and 18829.
- Checkout: Matrix sentence plus US West purchase.
Deploy OpenClaw on the US West node that clears your RTT gates
Read gateway ops and parallel FAQ. Reserve on US West purchase or purchase. Plans: pricing.